text
stringlengths
81
112k
Generator that yields series of consecutive audio frames comprising each utterence, separated by yielding a single None. Determines voice activity by ratio of frames in padding_ms. Uses a buffer to include padding_ms prior to being triggered. Example: (frame, ..., frame, None, frame, ..., frame, None, ...) |---utterence---| |---utterence---| def vad_collector(self, padding_ms=300, ratio=0.75, frames=None): """Generator that yields series of consecutive audio frames comprising each utterence, separated by yielding a single None. Determines voice activity by ratio of frames in padding_ms. Uses a buffer to include padding_ms prior to being triggered. Example: (frame, ..., frame, None, frame, ..., frame, None, ...) |---utterence---| |---utterence---| """ if frames is None: frames = self.frame_generator() num_padding_frames = padding_ms // self.frame_duration_ms ring_buffer = collections.deque(maxlen=num_padding_frames) triggered = False for frame in frames: is_speech = self.vad.is_speech(frame, self.sample_rate) if not triggered: ring_buffer.append((frame, is_speech)) num_voiced = len([f for f, speech in ring_buffer if speech]) if num_voiced > ratio * ring_buffer.maxlen: triggered = True for f, s in ring_buffer: yield f ring_buffer.clear() else: yield frame ring_buffer.append((frame, is_speech)) num_unvoiced = len([f for f, speech in ring_buffer if not speech]) if num_unvoiced > ratio * ring_buffer.maxlen: triggered = False yield None ring_buffer.clear()
Global `cut` function that supports parallel processing. Note that this only works using dt, custom POSTokenizer instances are not supported. def cut(sentence, HMM=True): """ Global `cut` function that supports parallel processing. Note that this only works using dt, custom POSTokenizer instances are not supported. """ global dt if jieba.pool is None: for w in dt.cut(sentence, HMM=HMM): yield w else: parts = strdecode(sentence).splitlines(True) if HMM: result = jieba.pool.map(_lcut_internal, parts) else: result = jieba.pool.map(_lcut_internal_no_hmm, parts) for r in result: for w in r: yield w
Change the module's `cut` and `cut_for_search` functions to the parallel version. Note that this only works using dt, custom Tokenizer instances are not supported. def enable_parallel(processnum=None): """ Change the module's `cut` and `cut_for_search` functions to the parallel version. Note that this only works using dt, custom Tokenizer instances are not supported. """ global pool, dt, cut, cut_for_search from multiprocessing import cpu_count if os.name == 'nt': raise NotImplementedError( "jieba: parallel mode only supports posix system") else: from multiprocessing import Pool dt.check_initialized() if processnum is None: processnum = cpu_count() pool = Pool(processnum) cut = _pcut cut_for_search = _pcut_for_search
The main function that segments an entire sentence that contains Chinese characters into separated words. Parameter: - sentence: The str(unicode) to be segmented. - cut_all: Model type. True for full pattern, False for accurate pattern. - HMM: Whether to use the Hidden Markov Model. def cut(self, sentence, cut_all=False, HMM=True): ''' The main function that segments an entire sentence that contains Chinese characters into separated words. Parameter: - sentence: The str(unicode) to be segmented. - cut_all: Model type. True for full pattern, False for accurate pattern. - HMM: Whether to use the Hidden Markov Model. ''' sentence = strdecode(sentence) if cut_all: re_han = re_han_cut_all re_skip = re_skip_cut_all else: re_han = re_han_default re_skip = re_skip_default if cut_all: cut_block = self.__cut_all elif HMM: cut_block = self.__cut_DAG else: cut_block = self.__cut_DAG_NO_HMM blocks = re_han.split(sentence) for blk in blocks: if not blk: continue if re_han.match(blk): for word in cut_block(blk): yield word else: tmp = re_skip.split(blk) for x in tmp: if re_skip.match(x): yield x elif not cut_all: for xx in x: yield xx else: yield x
Finer segmentation for search engines. def cut_for_search(self, sentence, HMM=True): """ Finer segmentation for search engines. """ words = self.cut(sentence, HMM=HMM) for w in words: if len(w) > 2: for i in xrange(len(w) - 1): gram2 = w[i:i + 2] if self.FREQ.get(gram2): yield gram2 if len(w) > 3: for i in xrange(len(w) - 2): gram3 = w[i:i + 3] if self.FREQ.get(gram3): yield gram3 yield w
Load personalized dict to improve detect rate. Parameter: - f : A plain text file contains words and their ocurrences. Can be a file-like object, or the path of the dictionary file, whose encoding must be utf-8. Structure of dict file: word1 freq1 word_type1 word2 freq2 word_type2 ... Word type may be ignored def load_userdict(self, f): ''' Load personalized dict to improve detect rate. Parameter: - f : A plain text file contains words and their ocurrences. Can be a file-like object, or the path of the dictionary file, whose encoding must be utf-8. Structure of dict file: word1 freq1 word_type1 word2 freq2 word_type2 ... Word type may be ignored ''' self.check_initialized() if isinstance(f, string_types): f_name = f f = open(f, 'rb') else: f_name = resolve_filename(f) for lineno, ln in enumerate(f, 1): line = ln.strip() if not isinstance(line, text_type): try: line = line.decode('utf-8').lstrip('\ufeff') except UnicodeDecodeError: raise ValueError('dictionary file %s must be utf-8' % f_name) if not line: continue # match won't be None because there's at least one character word, freq, tag = re_userdict.match(line).groups() if freq is not None: freq = freq.strip() if tag is not None: tag = tag.strip() self.add_word(word, freq, tag)
Add a word to dictionary. freq and tag can be omitted, freq defaults to be a calculated value that ensures the word can be cut out. def add_word(self, word, freq=None, tag=None): """ Add a word to dictionary. freq and tag can be omitted, freq defaults to be a calculated value that ensures the word can be cut out. """ self.check_initialized() word = strdecode(word) freq = int(freq) if freq is not None else self.suggest_freq(word, False) self.FREQ[word] = freq self.total += freq if tag: self.user_word_tag_tab[word] = tag for ch in xrange(len(word)): wfrag = word[:ch + 1] if wfrag not in self.FREQ: self.FREQ[wfrag] = 0 if freq == 0: finalseg.add_force_split(word)
Suggest word frequency to force the characters in a word to be joined or splitted. Parameter: - segment : The segments that the word is expected to be cut into, If the word should be treated as a whole, use a str. - tune : If True, tune the word frequency. Note that HMM may affect the final result. If the result doesn't change, set HMM=False. def suggest_freq(self, segment, tune=False): """ Suggest word frequency to force the characters in a word to be joined or splitted. Parameter: - segment : The segments that the word is expected to be cut into, If the word should be treated as a whole, use a str. - tune : If True, tune the word frequency. Note that HMM may affect the final result. If the result doesn't change, set HMM=False. """ self.check_initialized() ftotal = float(self.total) freq = 1 if isinstance(segment, string_types): word = segment for seg in self.cut(word, HMM=False): freq *= self.FREQ.get(seg, 1) / ftotal freq = max(int(freq * self.total) + 1, self.FREQ.get(word, 1)) else: segment = tuple(map(strdecode, segment)) word = ''.join(segment) for seg in segment: freq *= self.FREQ.get(seg, 1) / ftotal freq = min(int(freq * self.total), self.FREQ.get(word, 0)) if tune: add_word(word, freq) return freq
Tokenize a sentence and yields tuples of (word, start, end) Parameter: - sentence: the str(unicode) to be segmented. - mode: "default" or "search", "search" is for finer segmentation. - HMM: whether to use the Hidden Markov Model. def tokenize(self, unicode_sentence, mode="default", HMM=True): """ Tokenize a sentence and yields tuples of (word, start, end) Parameter: - sentence: the str(unicode) to be segmented. - mode: "default" or "search", "search" is for finer segmentation. - HMM: whether to use the Hidden Markov Model. """ if not isinstance(unicode_sentence, text_type): raise ValueError("jieba: the input parameter should be unicode.") start = 0 if mode == 'default': for w in self.cut(unicode_sentence, HMM=HMM): width = len(w) yield (w, start, start + width) start += width else: for w in self.cut(unicode_sentence, HMM=HMM): width = len(w) if len(w) > 2: for i in xrange(len(w) - 1): gram2 = w[i:i + 2] if self.FREQ.get(gram2): yield (gram2, start + i, start + i + 2) if len(w) > 3: for i in xrange(len(w) - 2): gram3 = w[i:i + 3] if self.FREQ.get(gram3): yield (gram3, start + i, start + i + 3) yield (w, start, start + width) start += width
Extract keywords from sentence using TextRank algorithm. Parameter: - topK: return how many top keywords. `None` for all possible words. - withWeight: if True, return a list of (word, weight); if False, return a list of words. - allowPOS: the allowed POS list eg. ['ns', 'n', 'vn', 'v']. if the POS of w is not in this list, it will be filtered. - withFlag: if True, return a list of pair(word, weight) like posseg.cut if False, return a list of words def textrank(self, sentence, topK=20, withWeight=False, allowPOS=('ns', 'n', 'vn', 'v'), withFlag=False): """ Extract keywords from sentence using TextRank algorithm. Parameter: - topK: return how many top keywords. `None` for all possible words. - withWeight: if True, return a list of (word, weight); if False, return a list of words. - allowPOS: the allowed POS list eg. ['ns', 'n', 'vn', 'v']. if the POS of w is not in this list, it will be filtered. - withFlag: if True, return a list of pair(word, weight) like posseg.cut if False, return a list of words """ self.pos_filt = frozenset(allowPOS) g = UndirectWeightedGraph() cm = defaultdict(int) words = tuple(self.tokenizer.cut(sentence)) for i, wp in enumerate(words): if self.pairfilter(wp): for j in xrange(i + 1, i + self.span): if j >= len(words): break if not self.pairfilter(words[j]): continue if allowPOS and withFlag: cm[(wp, words[j])] += 1 else: cm[(wp.word, words[j].word)] += 1 for terms, w in cm.items(): g.addEdge(terms[0], terms[1], w) nodes_rank = g.rank() if withWeight: tags = sorted(nodes_rank.items(), key=itemgetter(1), reverse=True) else: tags = sorted(nodes_rank, key=nodes_rank.__getitem__, reverse=True) if topK: return tags[:topK] else: return tags
Extract keywords from sentence using TF-IDF algorithm. Parameter: - topK: return how many top keywords. `None` for all possible words. - withWeight: if True, return a list of (word, weight); if False, return a list of words. - allowPOS: the allowed POS list eg. ['ns', 'n', 'vn', 'v','nr']. if the POS of w is not in this list,it will be filtered. - withFlag: only work with allowPOS is not empty. if True, return a list of pair(word, weight) like posseg.cut if False, return a list of words def extract_tags(self, sentence, topK=20, withWeight=False, allowPOS=(), withFlag=False): """ Extract keywords from sentence using TF-IDF algorithm. Parameter: - topK: return how many top keywords. `None` for all possible words. - withWeight: if True, return a list of (word, weight); if False, return a list of words. - allowPOS: the allowed POS list eg. ['ns', 'n', 'vn', 'v','nr']. if the POS of w is not in this list,it will be filtered. - withFlag: only work with allowPOS is not empty. if True, return a list of pair(word, weight) like posseg.cut if False, return a list of words """ if allowPOS: allowPOS = frozenset(allowPOS) words = self.postokenizer.cut(sentence) else: words = self.tokenizer.cut(sentence) freq = {} for w in words: if allowPOS: if w.flag not in allowPOS: continue elif not withFlag: w = w.word wc = w.word if allowPOS and withFlag else w if len(wc.strip()) < 2 or wc.lower() in self.stop_words: continue freq[w] = freq.get(w, 0.0) + 1.0 total = sum(freq.values()) for k in freq: kw = k.word if allowPOS and withFlag else k freq[k] *= self.idf_freq.get(kw, self.median_idf) / total if withWeight: tags = sorted(freq.items(), key=itemgetter(1), reverse=True) else: tags = sorted(freq, key=freq.__getitem__, reverse=True) if topK: return tags[:topK] else: return tags
Generates raw (English, other) pairs from a ParaCrawl V3.0 data file. Args: paracrawl_file: A ParaCrawl V3.0 en-.. data file. Yields: Pairs of (sentence_en, sentence_xx), as Unicode strings. Raises: StopIteration: If the file ends while this method is in the middle of creating a translation pair. def paracrawl_v3_pairs(paracrawl_file): """Generates raw (English, other) pairs from a ParaCrawl V3.0 data file. Args: paracrawl_file: A ParaCrawl V3.0 en-.. data file. Yields: Pairs of (sentence_en, sentence_xx), as Unicode strings. Raises: StopIteration: If the file ends while this method is in the middle of creating a translation pair. """ raw_sentences = _raw_sentences(paracrawl_file) for s_en in raw_sentences: try: s_xx = next(raw_sentences) if s_en and s_xx: # Prevent empty string examples. yield s_en, s_xx except StopIteration: tf.logging.error( 'Unmatched final sentence while reading in sentence pairs: [%s]', s_en)
Generates Unicode strings, one for each <seg> in a ParaCrawl data file. Also decodes some of the most common HTML entities found in ParaCrawl data. Args: paracrawl_file: A ParaCrawl V3.0 en-.. data file. Yields: One Unicode string for each <seg> element in the ParaCrawl data file. def _raw_sentences(paracrawl_file): """Generates Unicode strings, one for each <seg> in a ParaCrawl data file. Also decodes some of the most common HTML entities found in ParaCrawl data. Args: paracrawl_file: A ParaCrawl V3.0 en-.. data file. Yields: One Unicode string for each <seg> element in the ParaCrawl data file. """ for line_utf8 in paracrawl_file: line_uni = line_utf8.decode('UTF-8') text_match = re.match(r' +<seg>(.*)</seg>$', line_uni) if text_match: txt = text_match.group(1) txt = re.sub(r'&amp;', r'&', txt) txt = re.sub(r'& ?amp;', r'&', txt) txt = re.sub(r'& ?apos;', r"'", txt) txt = re.sub(r'& ?quot;', r'"', txt) txt = re.sub(r'& ?lt;', r'<', txt) txt = re.sub(r'& ?gt;', r'>', txt) yield txt
Generates a cleaned-up stream of (English, other) translation pairs. Cleaning includes both filtering and simplistic sentence splitting, with minimal assumptions on the non-English pair member: (1) All filtering is done based on the English member of the pair, and (2) sentence splitting assumes only that sentences can end with one of '.!?' and begin with an ASCII uppercase letter. Input pairs that would get split into different numbers of sentences (e.g., three English sentences vs. two German ones) are discarded. Args: en_xx_pairs: A stream (iterable) of Unicode string pairs. Each item in the stream should be a (sentence_en, sentence_xx) pair. Yields: Cleaned-up (sentence_en, sentence_xx) pairs. def clean_en_xx_pairs(en_xx_pairs): """Generates a cleaned-up stream of (English, other) translation pairs. Cleaning includes both filtering and simplistic sentence splitting, with minimal assumptions on the non-English pair member: (1) All filtering is done based on the English member of the pair, and (2) sentence splitting assumes only that sentences can end with one of '.!?' and begin with an ASCII uppercase letter. Input pairs that would get split into different numbers of sentences (e.g., three English sentences vs. two German ones) are discarded. Args: en_xx_pairs: A stream (iterable) of Unicode string pairs. Each item in the stream should be a (sentence_en, sentence_xx) pair. Yields: Cleaned-up (sentence_en, sentence_xx) pairs. """ for s1, s2 in en_xx_pairs: if _regex_filter(s1): continue s1_list, s2_list = _split_sentences(s1, s2) if len(s1_list) != len(s2_list): continue # discard this pair elif len(s1_list) == 1: yield s1, s2 else: for s1_subsentence, s2_subsentence in itertools.izip(s1_list, s2_list): if _regex_filter(s1_subsentence): continue yield s1_subsentence, s2_subsentence
Obtain a list of image paths corresponding to training or eval case. Args: tmp_dir: str, the root path to which raw images were written, at the top level having meta/ and raw/ subdirs. case: bool, whether obtaining file paths for training (true) or eval (false). training_fraction: float, the fraction of the sub-image path list to consider as the basis for training examples. Returns: list: A list of file paths. Raises: ValueError: if images not found in tmp_dir, or if training_fraction would leave no examples for eval. def _get_case_file_paths(tmp_dir, case, training_fraction=0.95): """Obtain a list of image paths corresponding to training or eval case. Args: tmp_dir: str, the root path to which raw images were written, at the top level having meta/ and raw/ subdirs. case: bool, whether obtaining file paths for training (true) or eval (false). training_fraction: float, the fraction of the sub-image path list to consider as the basis for training examples. Returns: list: A list of file paths. Raises: ValueError: if images not found in tmp_dir, or if training_fraction would leave no examples for eval. """ paths = tf.gfile.Glob("%s/*.jpg" % tmp_dir) if not paths: raise ValueError("Search of tmp_dir (%s) " % tmp_dir, "for subimage paths yielded an empty list, ", "can't proceed with returning training/eval split.") split_index = int(math.floor(len(paths)*training_fraction)) if split_index >= len(paths): raise ValueError("For a path list of size %s " "and a training_fraction of %s " "the resulting split_index of the paths list, " "%s, would leave no elements for the eval " "condition." % (len(paths), training_fraction, split_index)) if case: return paths[:split_index] else: return paths[split_index:]
Download a set of images from api.brain-map.org to `target_dir`. Args: image_ids: list, a list of image ids. target_dir: str, a directory to which to download the images. def maybe_download_image_dataset(image_ids, target_dir): """Download a set of images from api.brain-map.org to `target_dir`. Args: image_ids: list, a list of image ids. target_dir: str, a directory to which to download the images. """ tf.gfile.MakeDirs(target_dir) num_images = len(image_ids) for i, image_id in enumerate(image_ids): destination = os.path.join(target_dir, "%s.jpg" % i) tmp_destination = "%s.temp" % destination source_url = ("http://api.brain-map.org/api/v2/" "section_image_download/%s" % image_id) if tf.gfile.Exists(destination): tf.logging.info("Image with ID already present, " "skipping download (%s of %s)." % ( i+1, num_images )) continue tf.logging.info("Downloading image with id %s (%s of %s)" % ( image_id, i+1, num_images )) response = requests.get(source_url, stream=True) response.raise_for_status() with tf.gfile.Open(tmp_destination, "w") as f: for block in response.iter_content(1024): f.write(block) tf.gfile.Rename(tmp_destination, destination)
Create a numpy array with specified shape and masked fraction. Args: shape: tuple, shape of the mask to create. fraction: float, fraction of the mask area to populate with `mask_scalar`. Returns: numpy.array: A numpy array storing the mask. def random_square_mask(shape, fraction): """Create a numpy array with specified shape and masked fraction. Args: shape: tuple, shape of the mask to create. fraction: float, fraction of the mask area to populate with `mask_scalar`. Returns: numpy.array: A numpy array storing the mask. """ mask = np.ones(shape) patch_area = shape[0]*shape[1]*fraction patch_dim = np.int(math.floor(math.sqrt(patch_area))) if patch_area == 0 or patch_dim == 0: return mask x = np.random.randint(shape[0] - patch_dim) y = np.random.randint(shape[1] - patch_dim) mask[x:(x + patch_dim), y:(y + patch_dim), :] = 0 return mask
Base problem example generator for Allen Brain Atlas problems. Args: tmp_dir: str, a directory where raw example input data has been stored. training: bool, whether the mode of operation is training (or, alternatively, evaluation), determining whether examples in tmp_dir prefixed with train or dev will be used. size: int, the image size to add to the example annotation. training_fraction: float, the fraction of the sub-image path list to consider as the basis for training examples. Yields: A dictionary representing the images with the following fields: * image/encoded: The string encoding the image as JPEG. * image/format: The string "jpeg" indicating the image format. * image/height: The integer indicating the image height. * image/width: The integer indicating the image height. def _generator(tmp_dir, training, size=_BASE_EXAMPLE_IMAGE_SIZE, training_fraction=0.95): """Base problem example generator for Allen Brain Atlas problems. Args: tmp_dir: str, a directory where raw example input data has been stored. training: bool, whether the mode of operation is training (or, alternatively, evaluation), determining whether examples in tmp_dir prefixed with train or dev will be used. size: int, the image size to add to the example annotation. training_fraction: float, the fraction of the sub-image path list to consider as the basis for training examples. Yields: A dictionary representing the images with the following fields: * image/encoded: The string encoding the image as JPEG. * image/format: The string "jpeg" indicating the image format. * image/height: The integer indicating the image height. * image/width: The integer indicating the image height. """ maybe_download_image_dataset(_IMAGE_IDS, tmp_dir) image_files = _get_case_file_paths(tmp_dir=tmp_dir, case=training, training_fraction=training_fraction) image_obj = PIL_Image() tf.logging.info("Loaded case file paths (n=%s)" % len(image_files)) height = size width = size for input_path in image_files: img = image_obj.open(input_path) img = np.float32(img) shape = np.shape(img) for h_index in range(0, int(math.floor(shape[0]/size))): h_offset = h_index * size h_end = h_offset + size - 1 for v_index in range(0, int(math.floor(shape[1]/size))): v_offset = v_index * size v_end = v_offset + size - 1 # Extract a sub-image tile. subimage = np.uint8(img[h_offset:h_end, v_offset:v_end]) # pylint: disable=invalid-sequence-index # Filter images that are likely background (not tissue). if np.amax(subimage) < 230: continue subimage = image_obj.fromarray(subimage) buff = BytesIO() subimage.save(buff, format="JPEG") subimage_encoded = buff.getvalue() yield { "image/encoded": [subimage_encoded], "image/format": ["jpeg"], "image/height": [height], "image/width": [width] }
Set of hyperparameters. def transformer_moe_base(): """Set of hyperparameters.""" hparams = common_hparams.basic_params1() hparams.norm_type = "layer" hparams.hidden_size = 512 hparams.batch_size = 4096 hparams.max_length = 2001 hparams.max_input_seq_length = 2000 hparams.max_target_seq_length = 2000 hparams.dropout = 0.0 hparams.clip_grad_norm = 0. # i.e. no gradient clipping hparams.optimizer_adam_epsilon = 1e-9 hparams.learning_rate_decay_scheme = "noam" hparams.learning_rate = 0.1 hparams.learning_rate_warmup_steps = 2000 hparams.initializer_gain = 1.0 hparams.num_hidden_layers = 5 hparams.initializer = "uniform_unit_scaling" hparams.weight_decay = 0.0 hparams.optimizer_adam_beta1 = 0.9 hparams.optimizer_adam_beta2 = 0.98 hparams.num_sampled_classes = 0 hparams.label_smoothing = 0.0 hparams.shared_embedding_and_softmax_weights = True # According to noam, ("n", "da") seems better for harder-to-learn models hparams.layer_preprocess_sequence = "n" hparams.layer_postprocess_sequence = "da" # Hparams used by transformer_prepare_decoder() function hparams.add_hparam("pos", "timing") # timing, none hparams.add_hparam("proximity_bias", False) hparams.add_hparam("causal_decoder_self_attention", True) hparams = common_attention.add_standard_attention_hparams(hparams) # Decoder layers type. If set, num_decoder_layers parameter will be ignored # and the number of decoder layer will be deduced from the string # See top file comment for example of usage hparams.add_hparam("layer_types", "") # Default attention type (ex: a, loc, red,...) and feed-forward type (ex: fc, # sep, moe,...) hparams.add_hparam("default_att", "a") hparams.add_hparam("default_ff", "fc") return hparams
Hyper parameters specifics for long sequence generation. def transformer_moe_8k(): """Hyper parameters specifics for long sequence generation.""" hparams = transformer_moe_base() hparams.batch_size = 8192 hparams.max_length = 0 # max_length == batch_size hparams.eval_drop_long_sequences = True hparams.min_length_bucket = 256 # Avoid cyclic problems for big batches hparams.default_ff = "sep" hparams.hidden_size = 1024 return hparams
Base transformers model with moe. Will have the following architecture: * No encoder. * Layer 0: a - sep (self-attention - unmasked separable convolutions) * Layer 1: a - sep * Layer 2: a - sep * Layer 3: a - sep * Layer 4: a - sep * Decoder architecture: * Layer 0: a - a - sepm (self-attention - enco/deco-attention - masked sep) * Layer 1: a - a - sepm * Layer 2: a - a - moe (mixture of expert layers in the middle) * Layer 3: a - a - sepm * Layer 4: a - a - sepm Returns: hparams def transformer_moe_2k(): """Base transformers model with moe. Will have the following architecture: * No encoder. * Layer 0: a - sep (self-attention - unmasked separable convolutions) * Layer 1: a - sep * Layer 2: a - sep * Layer 3: a - sep * Layer 4: a - sep * Decoder architecture: * Layer 0: a - a - sepm (self-attention - enco/deco-attention - masked sep) * Layer 1: a - a - sepm * Layer 2: a - a - moe (mixture of expert layers in the middle) * Layer 3: a - a - sepm * Layer 4: a - a - sepm Returns: hparams """ hparams = transformer_moe_8k() hparams.batch_size = 2048 hparams.default_ff = "sep" # hparams.layer_types contains the network architecture: encoder_archi = "a/a/a/a/a" decoder_archi = "a-sepm/a-sepm/a-moe/a-sepm/a-sepm" hparams.layer_types = "{}#{}".format(encoder_archi, decoder_archi) return hparams
Model which formulate a seq2seq problem as language modeling. def transformer_moe_prepend_8k(): """Model which formulate a seq2seq problem as language modeling.""" hparams = transformer_moe_8k() hparams.prepend_mode = "prepend_inputs_masked_attention" hparams.eval_drop_long_sequences = False hparams.max_input_seq_length = 7500 hparams.default_ff = "sepm" hparams.layer_types = "locm/redm/locm-moe/redm/locm" hparams.moe_num_experts = 256 return hparams
Applies residual function for RevNet. Args: x: input tensor depth1: Number of output channels for the first and second conv layers. depth2: Number of output channels for the third conv layer. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. first_batch_norm: Whether to keep the first batch norm layer or not. Typically used in the first RevNet block. stride: Stride for the first conv filter. Note that this particular RevNet architecture only varies the stride for the first conv filter. The stride for the second conv filter is always set to 1. training: True for train phase, False for eval phase. bottleneck: If true, apply bottleneck 1x1 down/up sampling. padding: Padding for each conv layer. Returns: Output tensor after applying residual function for RevNet. def f(x, depth1, depth2, dim='2d', first_batch_norm=True, stride=1, training=True, bottleneck=True, padding='SAME'): """Applies residual function for RevNet. Args: x: input tensor depth1: Number of output channels for the first and second conv layers. depth2: Number of output channels for the third conv layer. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. first_batch_norm: Whether to keep the first batch norm layer or not. Typically used in the first RevNet block. stride: Stride for the first conv filter. Note that this particular RevNet architecture only varies the stride for the first conv filter. The stride for the second conv filter is always set to 1. training: True for train phase, False for eval phase. bottleneck: If true, apply bottleneck 1x1 down/up sampling. padding: Padding for each conv layer. Returns: Output tensor after applying residual function for RevNet. """ conv = CONFIG[dim]['conv'] with tf.variable_scope('f', reuse=tf.AUTO_REUSE): if first_batch_norm: net = tf.layers.batch_normalization(x, training=training) net = tf.nn.relu(net) else: net = x if bottleneck: net = conv(net, depth1, 1, strides=stride, padding=padding, activation=None) net = tf.layers.batch_normalization(net, training=training) net = tf.nn.relu(net) net = conv(net, depth1, 3, strides=1, padding=padding, activation=None) net = tf.layers.batch_normalization(net, training=training) net = tf.nn.relu(net) net = conv(net, depth2, 1, strides=1, padding=padding, activation=None) else: net = conv(net, depth2, 3, strides=stride, padding=padding, activation=None) net = tf.layers.batch_normalization(x, training=training) net = tf.nn.relu(net) net = conv(net, depth2, 3, strides=stride, padding=padding, activation=None) return net
Downsamples 'x' by `stride` using a 1x1 convolution filter. Args: x: input tensor of size [N, H, W, C] output_channels: Desired number of output channels. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. stride: What stride to use. Usually 1 or 2. scope: Optional variable scope. Returns: A downsampled tensor of size [N, H/2, W/2, output_channels] if stride is 2, else returns a tensor of size [N, H, W, output_channels] if stride is 1. def downsample_bottleneck(x, output_channels, dim='2d', stride=1, scope='h'): """Downsamples 'x' by `stride` using a 1x1 convolution filter. Args: x: input tensor of size [N, H, W, C] output_channels: Desired number of output channels. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. stride: What stride to use. Usually 1 or 2. scope: Optional variable scope. Returns: A downsampled tensor of size [N, H/2, W/2, output_channels] if stride is 2, else returns a tensor of size [N, H, W, output_channels] if stride is 1. """ conv = CONFIG[dim]['conv'] with tf.variable_scope(scope): x = conv(x, output_channels, 1, strides=stride, padding='SAME', activation=None) return x
Downsamples 'x' by `stride` using average pooling. Args: x: input tensor of size [N, H, W, C] output_channels: Desired number of output channels. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. stride: What stride to use. Usually 1 or 2. scope: Optional variable scope. Returns: A downsampled tensor of size [N, H/2, W/2, output_channels] if stride is 2, else returns a tensor of size [N, H, W, output_channels] if stride is 1. def downsample_residual(x, output_channels, dim='2d', stride=1, scope='h'): """Downsamples 'x' by `stride` using average pooling. Args: x: input tensor of size [N, H, W, C] output_channels: Desired number of output channels. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. stride: What stride to use. Usually 1 or 2. scope: Optional variable scope. Returns: A downsampled tensor of size [N, H/2, W/2, output_channels] if stride is 2, else returns a tensor of size [N, H, W, output_channels] if stride is 1. """ with tf.variable_scope(scope): if stride > 1: avg_pool = CONFIG[dim]['avg_pool'] x = avg_pool(x, pool_size=(stride, stride), strides=(stride, stride), padding='VALID') input_channels = tf.shape(x)[3] diff = output_channels - input_channels x = tf.pad( x, [[0, 0], [0, 0], [0, 0], [diff // 2, diff // 2]]) return x
Standard ResNet initial block used as first RevNet block. Args: images: [N, H, W, 3] tensor of input images to the model. num_channels: Output depth of convolutional layer in initial block. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. stride: stride for the convolution and pool layer. kernel_size: Size of the initial convolution filter maxpool: If true, apply a maxpool after the convolution training: True for train phase, False for eval phase. scope: Optional scope for the init block. Returns: Two [N, H, W, C] output activations from input images. def init(images, num_channels, dim='2d', stride=2, kernel_size=7, maxpool=True, training=True, scope='init'): """Standard ResNet initial block used as first RevNet block. Args: images: [N, H, W, 3] tensor of input images to the model. num_channels: Output depth of convolutional layer in initial block. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. stride: stride for the convolution and pool layer. kernel_size: Size of the initial convolution filter maxpool: If true, apply a maxpool after the convolution training: True for train phase, False for eval phase. scope: Optional scope for the init block. Returns: Two [N, H, W, C] output activations from input images. """ conv = CONFIG[dim]['conv'] pool = CONFIG[dim]['max_pool'] with tf.variable_scope(scope): net = conv(images, num_channels, kernel_size, strides=stride, padding='SAME', activation=None) net = tf.layers.batch_normalization(net, training=training) net = tf.nn.relu(net) if maxpool: net = pool(net, pool_size=3, strides=stride) x1, x2 = tf.split(net, 2, axis=CONFIG[dim]['split_axis']) return x1, x2
Implements bottleneck RevNet unit from authors' RevNet architecture. Args: x1: [N, H, W, C] tensor of network activations. x2: [N, H, W, C] tensor of network activations. block_num: integer ID of block depth: First depth in bottleneck residual unit. num_layers: Number of layers in the RevNet block. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. bottleneck: Should a bottleneck layer be used. first_batch_norm: Whether to keep the first batch norm layer or not. Typically used in the first RevNet block. stride: Stride for the residual function. training: True for train phase, False for eval phase. Returns: Two [N, H, W, C] output activation tensors. def unit(x1, x2, block_num, depth, num_layers, dim='2d', bottleneck=True, first_batch_norm=True, stride=1, training=True): """Implements bottleneck RevNet unit from authors' RevNet architecture. Args: x1: [N, H, W, C] tensor of network activations. x2: [N, H, W, C] tensor of network activations. block_num: integer ID of block depth: First depth in bottleneck residual unit. num_layers: Number of layers in the RevNet block. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. bottleneck: Should a bottleneck layer be used. first_batch_norm: Whether to keep the first batch norm layer or not. Typically used in the first RevNet block. stride: Stride for the residual function. training: True for train phase, False for eval phase. Returns: Two [N, H, W, C] output activation tensors. """ scope_name = 'unit_%d' % block_num if bottleneck: depth1 = depth depth2 = depth * 4 else: depth1 = depth2 = depth residual = wrapped_partial(f, depth1=depth1, depth2=depth2, dim=dim, training=training, bottleneck=bottleneck) with tf.variable_scope(scope_name): downsample = downsample_bottleneck if bottleneck else downsample_residual # Manual implementation of downsampling with tf.variable_scope('downsampling'): with tf.variable_scope('x1'): hx1 = downsample(x1, depth2, dim=dim, stride=stride) fx2 = residual(x2, stride=stride, first_batch_norm=first_batch_norm) x1 = hx1 + fx2 with tf.variable_scope('x2'): hx2 = downsample(x2, depth2, dim=dim, stride=stride) fx1 = residual(x1) x2 = hx2 + fx1 # Full block using memory-efficient rev_block implementation. with tf.variable_scope('full_block'): x1, x2 = tf.contrib.layers.rev_block(x1, x2, residual, residual, num_layers=num_layers) return x1, x2
Converts activations from last RevNet block to pre-logits. Args: x1: [NxHxWxC] tensor of network activations. x2: [NxHxWxC] tensor of network activations. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. training: True for train phase, False for eval phase. scope: Optional variable scope for the final block. Returns: [N, hidden_dim] pre-logits tensor from activations x1 and x2. def final_block(x1, x2, dim='2d', training=True, scope='final_block'): """Converts activations from last RevNet block to pre-logits. Args: x1: [NxHxWxC] tensor of network activations. x2: [NxHxWxC] tensor of network activations. dim: '2d' if 2-dimensional, '3d' if 3-dimensional. training: True for train phase, False for eval phase. scope: Optional variable scope for the final block. Returns: [N, hidden_dim] pre-logits tensor from activations x1 and x2. """ # Final batch norm and relu with tf.variable_scope(scope): y = tf.concat([x1, x2], axis=CONFIG[dim]['split_axis']) y = tf.layers.batch_normalization(y, training=training) y = tf.nn.relu(y) # Global average pooling net = tf.reduce_mean(y, CONFIG[dim]['reduction_dimensions'], name='final_pool', keep_dims=True) return net
Uses Tensor2Tensor memory optimized RevNet block to build a RevNet. Args: inputs: [NxHxWx3] tensor of input images to the model. hparams: HParams object that contains the following parameters, in addition to the parameters contained in the basic_params1() object in the common_hparams module: num_channels_first - A Python list where each element represents the depth of the first and third convolutional layers in the bottleneck residual unit for a given block. num_channels_second - A Python list where each element represents the depth of the second convolutional layer in the bottleneck residual unit for a given block. num_layers_per_block - A Python list containing the number of RevNet layers for each block. first_batch_norm - A Python list containing booleans representing the presence of a batch norm layer at the beginning of a given block. strides - A Python list containing integers representing the stride of the residual function for each block. num_channels_init_block - An integer representing the number of channels for the convolutional layer in the initial block. dimension - A string (either "2d" or "3d") that decides if the RevNet is 2-dimensional or 3-dimensional. reuse: Whether to reuse the default variable scope. Returns: [batch_size, hidden_dim] pre-logits tensor from the bottleneck RevNet. def revnet(inputs, hparams, reuse=None): """Uses Tensor2Tensor memory optimized RevNet block to build a RevNet. Args: inputs: [NxHxWx3] tensor of input images to the model. hparams: HParams object that contains the following parameters, in addition to the parameters contained in the basic_params1() object in the common_hparams module: num_channels_first - A Python list where each element represents the depth of the first and third convolutional layers in the bottleneck residual unit for a given block. num_channels_second - A Python list where each element represents the depth of the second convolutional layer in the bottleneck residual unit for a given block. num_layers_per_block - A Python list containing the number of RevNet layers for each block. first_batch_norm - A Python list containing booleans representing the presence of a batch norm layer at the beginning of a given block. strides - A Python list containing integers representing the stride of the residual function for each block. num_channels_init_block - An integer representing the number of channels for the convolutional layer in the initial block. dimension - A string (either "2d" or "3d") that decides if the RevNet is 2-dimensional or 3-dimensional. reuse: Whether to reuse the default variable scope. Returns: [batch_size, hidden_dim] pre-logits tensor from the bottleneck RevNet. """ training = hparams.mode == tf.estimator.ModeKeys.TRAIN with tf.variable_scope('RevNet', reuse=reuse): x1, x2 = init(inputs, num_channels=hparams.num_channels_init_block, dim=hparams.dim, kernel_size=hparams.init_kernel_size, maxpool=hparams.init_maxpool, stride=hparams.init_stride, training=training) for block_num in range(len(hparams.num_layers_per_block)): block = {'depth': hparams.num_channels[block_num], 'num_layers': hparams.num_layers_per_block[block_num], 'first_batch_norm': hparams.first_batch_norm[block_num], 'stride': hparams.strides[block_num], 'bottleneck': hparams.bottleneck} x1, x2 = unit(x1, x2, block_num, dim=hparams.dim, training=training, **block) pre_logits = final_block(x1, x2, dim=hparams.dim, training=training) return pre_logits
Default hparams for Revnet. def revnet_base(): """Default hparams for Revnet.""" hparams = common_hparams.basic_params1() hparams.add_hparam('num_channels', [64, 128, 256, 416]) hparams.add_hparam('num_layers_per_block', [1, 1, 10, 1]) hparams.add_hparam('bottleneck', True) hparams.add_hparam('first_batch_norm', [False, True, True, True]) hparams.add_hparam('init_stride', 2) hparams.add_hparam('init_kernel_size', 7) hparams.add_hparam('init_maxpool', True) hparams.add_hparam('strides', [1, 2, 2, 2]) hparams.add_hparam('num_channels_init_block', 64) hparams.add_hparam('dim', '2d') # Variable init hparams.initializer = 'normal_unit_scaling' hparams.initializer_gain = 2. # Optimization hparams.optimizer = 'Momentum' hparams.optimizer_momentum_momentum = 0.9 hparams.optimizer_momentum_nesterov = True hparams.weight_decay = 1e-4 hparams.clip_grad_norm = 0.0 # (base_lr=0.1) * (batch_size=128*8 (on TPU, or 8 GPUs)=1024) / (256.) hparams.learning_rate = 0.4 hparams.learning_rate_decay_scheme = 'cosine' # For image_imagenet224, 120k training steps, which effectively makes this a # cosine decay (i.e. no cycles). hparams.learning_rate_cosine_cycle_steps = 120000 # Can run with a batch size of 128 with Problem ImageImagenet224 hparams.batch_size = 128 return hparams
Tiny hparams suitable for CIFAR/etc. def revnet_cifar_base(): """Tiny hparams suitable for CIFAR/etc.""" hparams = revnet_base() hparams.num_channels_init_block = 32 hparams.first_batch_norm = [False, True, True] hparams.init_stride = 1 hparams.init_kernel_size = 3 hparams.init_maxpool = False hparams.strides = [1, 2, 2] hparams.batch_size = 128 hparams.weight_decay = 1e-4 hparams.learning_rate = 0.1 hparams.learning_rate_cosine_cycle_steps = 5000 return hparams
Tiny hparams suitable for CIFAR/etc. def revnet_110_cifar(): """Tiny hparams suitable for CIFAR/etc.""" hparams = revnet_cifar_base() hparams.bottleneck = False hparams.num_channels = [16, 32, 64] hparams.num_layers_per_block = [8, 8, 8] return hparams
Tiny hparams suitable for CIFAR/etc. def revnet_164_cifar(): """Tiny hparams suitable for CIFAR/etc.""" hparams = revnet_cifar_base() hparams.bottleneck = True hparams.num_channels = [16, 32, 64] hparams.num_layers_per_block = [8, 8, 8] return hparams
Hyperparameters for tuning revnet. def revnet_range(rhp): """Hyperparameters for tuning revnet.""" rhp.set_float('learning_rate', 0.05, 0.2, scale=rhp.LOG_SCALE) rhp.set_float('weight_decay', 1e-5, 1e-3, scale=rhp.LOG_SCALE) rhp.set_discrete('num_channels_init_block', [64, 128]) return rhp
Basic 2-frame conv model. def next_frame_basic_deterministic(): """Basic 2-frame conv model.""" hparams = base.next_frame_base() hparams.video_num_input_frames = 4 hparams.video_num_target_frames = 1 hparams.hidden_size = 64 hparams.batch_size = 4 hparams.num_hidden_layers = 2 hparams.optimizer = "Adafactor" hparams.learning_rate_constant = 1.5 hparams.learning_rate_warmup_steps = 8000 hparams.learning_rate_schedule = "linear_warmup * constant * rsqrt_decay" hparams.label_smoothing = 0.0 hparams.initializer = "uniform_unit_scaling" hparams.initializer_gain = 1.3 hparams.weight_decay = 0.0 hparams.clip_grad_norm = 1.0 hparams.dropout = 0.1 hparams.add_hparam("residual_dropout", 0.5) hparams.add_hparam("num_compress_steps", 6) hparams.add_hparam("filter_double_steps", 2) hparams.add_hparam("pixel_sampling_temperature", 0.0) hparams.add_hparam("concat_internal_states", False) hparams.add_hparam("do_autoregressive_rnn", False) hparams.add_hparam("autoregressive_rnn_lookback", 8) hparams.add_hparam("autoregressive_rnn_warmup_steps", 8000) hparams.add_hparam("activation_fn", "relu") hparams.bottom["inputs"] = modalities.video_identity_bottom hparams.bottom["targets"] = modalities.video_identity_bottom return hparams
Basic 2-frame conv model with pixel noise. def next_frame_pixel_noise(): """Basic 2-frame conv model with pixel noise.""" hparams = next_frame_basic_deterministic() hparams.add_hparam("video_modality_input_noise", 0.05) hparams.bottom["inputs"] = modalities.video_pixel_noise_bottom hparams.top["inputs"] = modalities.video_top return hparams
Basic conv model with scheduled sampling. def next_frame_sampling(): """Basic conv model with scheduled sampling.""" hparams = next_frame_basic_deterministic() hparams.scheduled_sampling_mode = "prob_inverse_exp" hparams.scheduled_sampling_max_prob = 1.0 hparams.scheduled_sampling_decay_steps = 10000 return hparams
Conv autoencoder. def next_frame_ae(): """Conv autoencoder.""" hparams = next_frame_basic_deterministic() hparams.bottom["inputs"] = modalities.video_bitwise_bottom hparams.top["inputs"] = modalities.video_top hparams.hidden_size = 256 hparams.batch_size = 8 hparams.num_hidden_layers = 4 hparams.num_compress_steps = 4 hparams.dropout = 0.4 return hparams
Conv autoencoder, tiny set for testing. def next_frame_ae_tiny(): """Conv autoencoder, tiny set for testing.""" hparams = next_frame_tiny() hparams.bottom["inputs"] = modalities.video_bitwise_bottom hparams.top["inputs"] = modalities.video_top hparams.batch_size = 8 hparams.dropout = 0.4 return hparams
Tiny for testing. def next_frame_tiny(): """Tiny for testing.""" hparams = next_frame_basic_deterministic() hparams.hidden_size = 32 hparams.num_hidden_layers = 1 hparams.num_compress_steps = 2 hparams.filter_double_steps = 1 return hparams
Basic conv model with L1 modality. def next_frame_l1(): """Basic conv model with L1 modality.""" hparams = next_frame_basic_deterministic() hparams.loss["targets"] = modalities.video_l1_loss hparams.top["targets"] = modalities.video_l1_top hparams.video_modality_loss_cutoff = 2.4 return hparams
Basic conv model with L2 modality. def next_frame_l2(): """Basic conv model with L2 modality.""" hparams = next_frame_basic_deterministic() hparams.loss["targets"] = modalities.video_l2_loss hparams.top["targets"] = modalities.video_l1_top hparams.video_modality_loss_cutoff = 2.4 return hparams
Basic tuning grid. def next_frame_base_range(rhp): """Basic tuning grid.""" rhp.set_float("dropout", 0.2, 0.6) rhp.set_discrete("hidden_size", [64, 128, 256]) rhp.set_int("num_compress_steps", 5, 8) rhp.set_discrete("batch_size", [4, 8, 16, 32]) rhp.set_int("num_hidden_layers", 1, 3) rhp.set_int("filter_double_steps", 1, 6) rhp.set_float("learning_rate_constant", 1., 4.) rhp.set_int("learning_rate_warmup_steps", 500, 3000) rhp.set_float("initializer_gain", 0.8, 1.8)
Autoencoder world model tuning grid. def next_frame_ae_range(rhp): """Autoencoder world model tuning grid.""" rhp.set_float("dropout", 0.3, 0.5) rhp.set_int("num_compress_steps", 1, 3) rhp.set_int("num_hidden_layers", 2, 6) rhp.set_float("learning_rate_constant", 1., 2.) rhp.set_float("initializer_gain", 0.8, 1.5) rhp.set_int("filter_double_steps", 2, 3)
Series of architectures for language modeling. def mqp_lm1b_base(): """Series of architectures for language modeling.""" hparams = mtf_transformer2.mtf_unitransformer_base() hparams.d_model = 1024 hparams.max_length = 256 hparams.batch_size = 256 # Parameters for my_layer_stack() hparams.num_hidden_layers = 6 hparams.d_ff = 8192 hparams.d_kv = 128 hparams.num_heads = 8 hparams.learning_rate_decay_steps = 13600 hparams.layout = "batch:batch;vocab:model;d_ff:model;heads:model" hparams.mesh_shape = "batch:32" return hparams
Initializes env_specs using the appropriate env. def initialize_env_specs(hparams, env_problem_name): """Initializes env_specs using the appropriate env.""" if env_problem_name: env = registry.env_problem(env_problem_name, batch_size=hparams.batch_size) else: env = rl_utils.setup_env(hparams, hparams.batch_size, hparams.eval_max_num_noops, hparams.rl_env_max_episode_steps, env_name=hparams.rl_env_name) env.start_new_epoch(0) return rl.make_real_env_fn(env)
Train. def train(hparams, output_dir, env_problem_name, report_fn=None): """Train.""" env_fn = initialize_env_specs(hparams, env_problem_name) tf.logging.vlog(1, "HParams in trainer_model_free.train : %s", misc_utils.pprint_hparams(hparams)) tf.logging.vlog(1, "Using hparams.base_algo: %s", hparams.base_algo) learner = rl_utils.LEARNERS[hparams.base_algo]( hparams.frame_stack_size, output_dir, output_dir, total_num_epochs=1 ) policy_hparams = trainer_lib.create_hparams(hparams.base_algo_params) rl_utils.update_hparams_from_hparams( policy_hparams, hparams, hparams.base_algo + "_" ) tf.logging.vlog(1, "Policy HParams : %s", misc_utils.pprint_hparams(policy_hparams)) # TODO(konradczechowski): remove base_algo dependance, when evaluation method # will be decided if hparams.base_algo == "ppo": total_steps = policy_hparams.epochs_num tf.logging.vlog(2, "total_steps: %d", total_steps) eval_every_epochs = policy_hparams.eval_every_epochs tf.logging.vlog(2, "eval_every_epochs: %d", eval_every_epochs) if eval_every_epochs == 0: eval_every_epochs = total_steps policy_hparams.eval_every_epochs = 0 metric_name = rl_utils.get_metric_name( sampling_temp=hparams.eval_sampling_temps[0], max_num_noops=hparams.eval_max_num_noops, clipped=False ) tf.logging.vlog(1, "metric_name: %s", metric_name) eval_metrics_dir = os.path.join(output_dir, "eval_metrics") eval_metrics_dir = os.path.expanduser(eval_metrics_dir) tf.gfile.MakeDirs(eval_metrics_dir) eval_metrics_writer = tf.summary.FileWriter(eval_metrics_dir) def evaluate_on_new_model(model_dir_path): global step eval_metrics = rl_utils.evaluate_all_configs(hparams, model_dir_path) tf.logging.info( "Agent eval metrics:\n{}".format(pprint.pformat(eval_metrics))) rl_utils.summarize_metrics(eval_metrics_writer, eval_metrics, step) if report_fn: report_fn(eval_metrics[metric_name], step) step += 1 policy_hparams.epochs_num = total_steps policy_hparams.save_models_every_epochs = eval_every_epochs else: def evaluate_on_new_model(model_dir_path): del model_dir_path raise NotImplementedError( "This function is currently implemented only for ppo") learner.train(env_fn, policy_hparams, simulated=False, save_continuously=True, epoch=0, model_save_fn=evaluate_on_new_model)
Compute the designated learning rate factor from hparams. def learning_rate_factor(name, step_num, hparams): """Compute the designated learning rate factor from hparams.""" if name == "constant": tf.logging.info("Base learning rate: %f", hparams.learning_rate_constant) return hparams.learning_rate_constant elif name == "linear_warmup": return tf.minimum(1.0, step_num / hparams.learning_rate_warmup_steps) elif name == "linear_decay": ret = (hparams.train_steps - step_num) / hparams.learning_rate_decay_steps return tf.minimum(1.0, tf.maximum(0.0, ret)) elif name == "cosdecay": # openai gpt in_warmup = tf.cast(step_num <= hparams.learning_rate_warmup_steps, dtype=tf.float32) ret = 0.5 * (1 + tf.cos( np.pi * step_num / hparams.learning_rate_decay_steps)) # if in warmup stage return 1 else return the decayed value return in_warmup * 1 + (1 - in_warmup) * ret elif name == "single_cycle_cos_decay": # Cosine decay to zero with a single cycle. This is different from # "cosdecay" because it starts at 1 when the warmup steps end. x = tf.maximum(step_num, hparams.learning_rate_warmup_steps) step = x - hparams.learning_rate_warmup_steps return tf.math.cos( step * np.pi / hparams.learning_rate_decay_steps) / 2.0 + 0.5 elif name == "rsqrt_decay": return tf.rsqrt(tf.maximum(step_num, hparams.learning_rate_warmup_steps)) elif name == "rsqrt_normalized_decay": scale = tf.sqrt(tf.to_float(hparams.learning_rate_warmup_steps)) return scale * tf.rsqrt(tf.maximum( step_num, hparams.learning_rate_warmup_steps)) elif name == "exp_decay": decay_steps = hparams.learning_rate_decay_steps warmup_steps = hparams.learning_rate_warmup_steps p = (step_num - warmup_steps) / decay_steps p = tf.maximum(p, 0.) if hparams.learning_rate_decay_staircase: p = tf.floor(p) return tf.pow(hparams.learning_rate_decay_rate, p) elif name == "rsqrt_hidden_size": return hparams.hidden_size ** -0.5 elif name == "legacy": return legacy_learning_rate_schedule(hparams) else: raise ValueError("unknown learning rate factor %s" % name)
Learning rate schedule based on hparams. def learning_rate_schedule(hparams): """Learning rate schedule based on hparams.""" mlperf_log.transformer_print(key=mlperf_log.OPT_LR, deferred=True) mlperf_log.transformer_print( key=mlperf_log.OPT_LR_WARMUP_STEPS, value=hparams.learning_rate_warmup_steps) step_num = _global_step(hparams) schedule_string = hparams.learning_rate_schedule names = schedule_string.split("*") names = [name.strip() for name in names if name.strip()] ret = tf.constant(1.0) for name in names: ret *= learning_rate_factor(name, step_num, hparams) return ret
Backwards-compatible learning-rate schedule. def legacy_learning_rate_schedule(hparams): """Backwards-compatible learning-rate schedule.""" step_num = _global_step(hparams) warmup_steps = tf.to_float(hparams.learning_rate_warmup_steps) if hparams.learning_rate_decay_scheme == "noam": ret = 5000.0 * hparams.hidden_size**-0.5 * tf.minimum( (step_num + 1) * warmup_steps**-1.5, (step_num + 1)**-0.5) else: warmup_steps = hparams.learning_rate_warmup_steps warmup = _learning_rate_warmup(warmup_steps, hparams=hparams) decay = _learning_rate_decay(hparams, warmup_steps) ret = tf.where(step_num < warmup_steps, warmup, decay) optimizer_correction = 0.002 if "adam" in hparams.optimizer else 1.0 tf.logging.info("Base learning rate: %f", hparams.learning_rate) return ret * optimizer_correction * hparams.learning_rate
Adjust global step if a multi-step optimizer is used. def _global_step(hparams): """Adjust global step if a multi-step optimizer is used.""" step = tf.to_float(tf.train.get_or_create_global_step()) multiplier = hparams.optimizer_multistep_accumulate_steps if not multiplier: return step tf.logging.info("Dividing global step by %d for multi-step optimizer." % multiplier) return step / tf.to_float(multiplier)
Scale learning rate according to the given schedule. Multipliers are not cumulative. Args: step: global step boundaries: List of steps to transition on. values: Multiplier to apply at each boundary transition. Returns: Scaled value for the learning rate. def _piecewise_learning_rate(step, boundaries, values): """Scale learning rate according to the given schedule. Multipliers are not cumulative. Args: step: global step boundaries: List of steps to transition on. values: Multiplier to apply at each boundary transition. Returns: Scaled value for the learning rate. """ values = [1.0] + values boundaries = [float(x) for x in boundaries] return tf.train.piecewise_constant( step, boundaries, values, name="piecewise_lr")
Learning rate decay multiplier. def _learning_rate_decay(hparams, warmup_steps=0): """Learning rate decay multiplier.""" scheme = hparams.learning_rate_decay_scheme warmup_steps = tf.to_float(warmup_steps) global_step = _global_step(hparams) if not scheme or scheme == "none": return tf.constant(1.) tf.logging.info("Applying learning rate decay: %s.", scheme) if scheme == "exp": decay_steps = hparams.learning_rate_decay_steps p = (global_step - warmup_steps) / decay_steps if hparams.learning_rate_decay_staircase: p = tf.floor(p) return tf.pow(hparams.learning_rate_decay_rate, p) if scheme == "piecewise": return _piecewise_learning_rate(global_step, hparams.learning_rate_boundaries, hparams.learning_rate_multiples) if scheme == "cosine": cycle_steps = hparams.learning_rate_cosine_cycle_steps cycle_position = global_step % (2 * cycle_steps) cycle_position = cycle_steps - tf.abs(cycle_steps - cycle_position) return 0.5 * (1 + tf.cos(np.pi * cycle_position / cycle_steps)) if scheme == "cyclelinear10x": # Cycle the rate linearly by 10x every warmup_steps, up and down. cycle_steps = warmup_steps cycle_position = global_step % (2 * cycle_steps) cycle_position = tf.to_float( # Normalize to the interval [-1, 1]. cycle_position - cycle_steps) / float(cycle_steps) cycle_position = 1.0 - tf.abs(cycle_position) # 0 to 1 and back to 0. return (cycle_position + 0.1) * 3.0 # 10x difference each cycle (0.3-3). if scheme == "sqrt": return _legacy_sqrt_decay(global_step - warmup_steps) raise ValueError("Unrecognized learning rate decay scheme: %s" % hparams.learning_rate_decay_scheme)
Learning rate warmup multiplier. def _learning_rate_warmup(warmup_steps, warmup_schedule="exp", hparams=None): """Learning rate warmup multiplier.""" if not warmup_steps: return tf.constant(1.) tf.logging.info("Applying %s learning rate warmup for %d steps", warmup_schedule, warmup_steps) warmup_steps = tf.to_float(warmup_steps) global_step = _global_step(hparams) if warmup_schedule == "exp": return tf.exp(tf.log(0.01) / warmup_steps)**(warmup_steps - global_step) else: assert warmup_schedule == "linear" start = tf.constant(0.35) return ((tf.constant(1.) - start) / warmup_steps) * global_step + start
Returns True if `find` is a subtree of `expr`. def is_in_expr(expr, find): """Returns True if `find` is a subtree of `expr`.""" return expr == find or (isinstance(expr, ExprNode) and expr.is_in(find))
Generate a random expression tree with a required variable. The required variable appears exactly once in the expression. Args: depth: At least one leaf will be this many levels down from the top. required_var: A char. This char is guaranteed to be placed exactly once at a leaf somewhere in the tree. This is the var to solve for. optional_list: A list of chars. These chars are randomly selected as leaf values. These are constant vars. ops: A list of ExprOp instances. Returns: An ExprNode instance which is the root of the generated expression tree. def random_expr_with_required_var(depth, required_var, optional_list, ops): """Generate a random expression tree with a required variable. The required variable appears exactly once in the expression. Args: depth: At least one leaf will be this many levels down from the top. required_var: A char. This char is guaranteed to be placed exactly once at a leaf somewhere in the tree. This is the var to solve for. optional_list: A list of chars. These chars are randomly selected as leaf values. These are constant vars. ops: A list of ExprOp instances. Returns: An ExprNode instance which is the root of the generated expression tree. """ if not depth: if required_var: return required_var return str(optional_list[random.randrange(len(optional_list))]) max_depth_side = random.randrange(2) other_side_depth = random.randrange(depth) required_var_side = random.randrange(2) left = random_expr_with_required_var( depth - 1 if max_depth_side else other_side_depth, required_var if required_var_side else None, optional_list, ops) right = random_expr_with_required_var( depth - 1 if not max_depth_side else other_side_depth, required_var if not required_var_side else None, optional_list, ops) op = ops[random.randrange(len(ops))] return ExprNode(left, right, op)
Generate a random expression tree. Args: depth: At least one leaf will be this many levels down from the top. vlist: A list of chars. These chars are randomly selected as leaf values. ops: A list of ExprOp instances. Returns: An ExprNode instance which is the root of the generated expression tree. def random_expr(depth, vlist, ops): """Generate a random expression tree. Args: depth: At least one leaf will be this many levels down from the top. vlist: A list of chars. These chars are randomly selected as leaf values. ops: A list of ExprOp instances. Returns: An ExprNode instance which is the root of the generated expression tree. """ if not depth: return str(vlist[random.randrange(len(vlist))]) max_depth_side = random.randrange(2) other_side_depth = random.randrange(depth) left = random_expr(depth - 1 if max_depth_side else other_side_depth, vlist, ops) right = random_expr(depth - 1 if not max_depth_side else other_side_depth, vlist, ops) op = ops[random.randrange(len(ops))] return ExprNode(left, right, op)
Solves for the value of the given var in an expression. Args: left: The root of the ExprNode tree on the left side of the equals sign. right: The root of the ExprNode tree on the right side of the equals sign. var: A char. The variable to solve for. solve_ops: A dictionary with the following properties. * For each operator in the expression, there is a rule that determines how to cancel out a value either to the left or the right of that operator. * For each rule, there is an entry in the dictionary. The key is two chars- the op char, and either 'l' or 'r' meaning rule for canceling out the left or right sides. For example, '+l', '+r', '-l', '-r'. * The value of each entry is a function with the following signature: (left, right, to_tree) -> (new_from_tree, new_to_tree) left- Expression on left side of the op. right- Expression on the right side of the op. to_tree- The tree on the other side of the equal sign. The canceled out expression will be moved here. new_from_tree- The resulting from_tree after the algebraic manipulation. new_to_tree- The resulting to_tree after the algebraic manipulation. Returns: The root of an ExprNode tree which holds the value of `var` after solving. Raises: ValueError: If `var` does not appear exactly once in the equation (which includes the left and right sides). def algebra_inverse_solve(left, right, var, solve_ops): """Solves for the value of the given var in an expression. Args: left: The root of the ExprNode tree on the left side of the equals sign. right: The root of the ExprNode tree on the right side of the equals sign. var: A char. The variable to solve for. solve_ops: A dictionary with the following properties. * For each operator in the expression, there is a rule that determines how to cancel out a value either to the left or the right of that operator. * For each rule, there is an entry in the dictionary. The key is two chars- the op char, and either 'l' or 'r' meaning rule for canceling out the left or right sides. For example, '+l', '+r', '-l', '-r'. * The value of each entry is a function with the following signature: (left, right, to_tree) -> (new_from_tree, new_to_tree) left- Expression on left side of the op. right- Expression on the right side of the op. to_tree- The tree on the other side of the equal sign. The canceled out expression will be moved here. new_from_tree- The resulting from_tree after the algebraic manipulation. new_to_tree- The resulting to_tree after the algebraic manipulation. Returns: The root of an ExprNode tree which holds the value of `var` after solving. Raises: ValueError: If `var` does not appear exactly once in the equation (which includes the left and right sides). """ is_in_left = is_in_expr(left, var) is_in_right = is_in_expr(right, var) if is_in_left == is_in_right: if is_in_left: raise ValueError("Solve-variable '%s' is on both sides of the equation. " "Only equations where the solve variable-appears once " "are supported by this solver. Left: '%s', right: '%s'" % (var, str(left), str(right))) else: raise ValueError("Solve-variable '%s' is not present in the equation. It " "must appear once. Left: '%s', right: '%s'" % (var, str(left), str(right))) from_tree = left if is_in_left else right to_tree = left if not is_in_left else right while from_tree != var: is_in_left = is_in_expr(from_tree.left, var) is_in_right = is_in_expr(from_tree.right, var) from_tree, to_tree = (solve_ops[str(from_tree.op) + ("l" if is_in_left else "r")]( from_tree.left, from_tree.right, to_tree)) return to_tree
Convert sympy expression into a string which can be encoded. Args: sympy_expr: Any sympy expression tree or string. functions: Defines special functions. A dict mapping human readable string names, like "log", "exp", "sin", "cos", etc., to single chars. Each function gets a unique token, like "L" for "log". Returns: A string representation of the expression suitable for encoding as a sequence input. def format_sympy_expr(sympy_expr, functions=None): """Convert sympy expression into a string which can be encoded. Args: sympy_expr: Any sympy expression tree or string. functions: Defines special functions. A dict mapping human readable string names, like "log", "exp", "sin", "cos", etc., to single chars. Each function gets a unique token, like "L" for "log". Returns: A string representation of the expression suitable for encoding as a sequence input. """ if functions is None: functions = {} str_expr = str(sympy_expr) result = str_expr.replace(" ", "") for fn_name, char in six.iteritems(functions): result = result.replace(fn_name, char) return result
Randomly generate an algebra inverse dataset sample. Given an input equation and variable, produce the expression equal to the variable. Args: vlist: Variable list. List of chars that can be used in the expression. ops: List of ExprOp instances. The allowed operators for the expression. solve_ops: See `solve_ops` documentation in `algebra_inverse_solve`. min_depth: Expression trees will not have a smaller depth than this. 0 means there is just a variable. 1 means there is one operation. max_depth: Expression trees will not have a larger depth than this. To make all trees have the same depth, set this equal to `min_depth`. Returns: sample: String representation of the input. Will be of the form 'solve_var:left_side=right_side'. target: String representation of the solution. def generate_algebra_inverse_sample(vlist, ops, solve_ops, min_depth, max_depth): """Randomly generate an algebra inverse dataset sample. Given an input equation and variable, produce the expression equal to the variable. Args: vlist: Variable list. List of chars that can be used in the expression. ops: List of ExprOp instances. The allowed operators for the expression. solve_ops: See `solve_ops` documentation in `algebra_inverse_solve`. min_depth: Expression trees will not have a smaller depth than this. 0 means there is just a variable. 1 means there is one operation. max_depth: Expression trees will not have a larger depth than this. To make all trees have the same depth, set this equal to `min_depth`. Returns: sample: String representation of the input. Will be of the form 'solve_var:left_side=right_side'. target: String representation of the solution. """ side = random.randrange(2) left_depth = random.randrange(min_depth if side else 0, max_depth + 1) right_depth = random.randrange(min_depth if not side else 0, max_depth + 1) var_index = random.randrange(len(vlist)) var = vlist[var_index] consts = vlist[:var_index] + vlist[var_index + 1:] left = random_expr_with_required_var(left_depth, var if side else None, consts, ops) right = random_expr_with_required_var(right_depth, var if not side else None, consts, ops) left_str = str(left) right_str = str(right) target = str(algebra_inverse_solve(left, right, var, solve_ops)) sample = "%s:%s=%s" % (var, left_str, right_str) return sample, target
Randomly generate an algebra simplify dataset sample. Given an input expression, produce the simplified expression. Args: vlist: Variable list. List of chars that can be used in the expression. ops: List of ExprOp instances. The allowed operators for the expression. min_depth: Expression trees will not have a smaller depth than this. 0 means there is just a variable. 1 means there is one operation. max_depth: Expression trees will not have a larger depth than this. To make all trees have the same depth, set this equal to `min_depth`. Returns: sample: String representation of the input. target: String representation of the solution. def generate_algebra_simplify_sample(vlist, ops, min_depth, max_depth): """Randomly generate an algebra simplify dataset sample. Given an input expression, produce the simplified expression. Args: vlist: Variable list. List of chars that can be used in the expression. ops: List of ExprOp instances. The allowed operators for the expression. min_depth: Expression trees will not have a smaller depth than this. 0 means there is just a variable. 1 means there is one operation. max_depth: Expression trees will not have a larger depth than this. To make all trees have the same depth, set this equal to `min_depth`. Returns: sample: String representation of the input. target: String representation of the solution. """ depth = random.randrange(min_depth, max_depth + 1) expr = random_expr(depth, vlist, ops) sample = str(expr) target = format_sympy_expr(sympy.simplify(sample)) return sample, target
Randomly generate a symbolic integral dataset sample. Given an input expression, produce the indefinite integral. Args: vlist: Variable list. List of chars that can be used in the expression. ops: List of ExprOp instances. The allowed operators for the expression. min_depth: Expression trees will not have a smaller depth than this. 0 means there is just a variable. 1 means there is one operation. max_depth: Expression trees will not have a larger depth than this. To make all trees have the same depth, set this equal to `min_depth`. functions: Defines special functions. A dict mapping human readable string names, like "log", "exp", "sin", "cos", etc., to single chars. Each function gets a unique token, like "L" for "log". Returns: sample: String representation of the input. Will be of the form 'var:expression'. target: String representation of the solution. def generate_calculus_integrate_sample(vlist, ops, min_depth, max_depth, functions): """Randomly generate a symbolic integral dataset sample. Given an input expression, produce the indefinite integral. Args: vlist: Variable list. List of chars that can be used in the expression. ops: List of ExprOp instances. The allowed operators for the expression. min_depth: Expression trees will not have a smaller depth than this. 0 means there is just a variable. 1 means there is one operation. max_depth: Expression trees will not have a larger depth than this. To make all trees have the same depth, set this equal to `min_depth`. functions: Defines special functions. A dict mapping human readable string names, like "log", "exp", "sin", "cos", etc., to single chars. Each function gets a unique token, like "L" for "log". Returns: sample: String representation of the input. Will be of the form 'var:expression'. target: String representation of the solution. """ var_index = random.randrange(len(vlist)) var = vlist[var_index] consts = vlist[:var_index] + vlist[var_index + 1:] depth = random.randrange(min_depth, max_depth + 1) expr = random_expr_with_required_var(depth, var, consts, ops) expr_str = str(expr) sample = var + ":" + expr_str target = format_sympy_expr( sympy.integrate(expr_str, sympy.Symbol(var)), functions=functions) return sample, target
Initializes required objects to generate symbolic math datasets. Produces token set, ExprOp instances, solve_op dictionary, encoders, and decoders needed to generate the algebra inverse dataset. Args: alphabet_size: How many possible variables there are. Max 52. digits: How many numerical digits to encode as tokens, "0" through str(digits-1), or None to encode no digits. functions: Defines special functions. A dict mapping human readable string names, like "log", "exp", "sin", "cos", etc., to single chars. Each function gets a unique token, like "L" for "log". WARNING, Make sure these tokens do not conflict with the list of possible variable names. Returns: AlgebraConfig instance holding all the objects listed above. Raises: ValueError: If `alphabet_size` is not in range [2, 52]. def math_dataset_init(alphabet_size=26, digits=None, functions=None): """Initializes required objects to generate symbolic math datasets. Produces token set, ExprOp instances, solve_op dictionary, encoders, and decoders needed to generate the algebra inverse dataset. Args: alphabet_size: How many possible variables there are. Max 52. digits: How many numerical digits to encode as tokens, "0" through str(digits-1), or None to encode no digits. functions: Defines special functions. A dict mapping human readable string names, like "log", "exp", "sin", "cos", etc., to single chars. Each function gets a unique token, like "L" for "log". WARNING, Make sure these tokens do not conflict with the list of possible variable names. Returns: AlgebraConfig instance holding all the objects listed above. Raises: ValueError: If `alphabet_size` is not in range [2, 52]. """ ops_list = ["+", "-", "*", "/"] ops = { "+": ExprOp("+", 0, True), "-": ExprOp("-", 0, False), "*": ExprOp("*", 1, True), "/": ExprOp("/", 1, False) } solve_ops = { "+l": lambda l, r, to: (l, ExprNode(to, r, ops["-"])), "+r": lambda l, r, to: (r, ExprNode(to, l, ops["-"])), "-l": lambda l, r, to: (l, ExprNode(to, r, ops["+"])), "-r": lambda l, r, to: (r, ExprNode(l, to, ops["-"])), "*l": lambda l, r, to: (l, ExprNode(to, r, ops["/"])), "*r": lambda l, r, to: (r, ExprNode(to, l, ops["/"])), "/l": lambda l, r, to: (l, ExprNode(to, r, ops["*"])), "/r": lambda l, r, to: (r, ExprNode(l, to, ops["/"])), } alphabet = ( [six.int2byte(ord("a") + c).decode("utf-8") for c in range(26)] + [six.int2byte(ord("A") + c).decode("utf-8") for c in range(26)]) if alphabet_size > 52: raise ValueError( "alphabet_size cannot be greater than 52. Got %s." % alphabet_size) if alphabet_size < 2: raise ValueError( "alphabet_size cannot be less than 2. Got %s." % alphabet_size) if digits is not None and not 1 <= digits <= 10: raise ValueError("digits cannot must be between 1 and 10. Got %s." % digits) vlist = alphabet[:alphabet_size] if digits is not None: dlist = [str(d) for d in range(digits)] else: dlist = [] if functions is None: functions = {} flist = sorted(functions.values()) pad = "_" tokens = [pad] + [":", "(", ")", "="] + ops_list + vlist + dlist + flist if len(tokens) != len(set(tokens)): raise ValueError("Duplicate token. Tokens: %s" % tokens) token_map = dict([(t, i) for i, t in enumerate(tokens)]) def int_encoder(sequence): return [token_map[s] for s in sequence] def int_decoder(tensor_1d): return "".join([tokens[i] for i in tensor_1d]) return AlgebraConfig( vlist=vlist, dlist=dlist, flist=flist, functions=functions, ops=ops, solve_ops=solve_ops, int_encoder=int_encoder, int_decoder=int_decoder)
Generate the algebra inverse dataset. Each sample is a symbolic math equation involving unknown variables. The task is to solve for the given variable. The target is the resulting expression. Args: alphabet_size: How many possible variables there are. Max 52. min_depth: Minimum depth of the expression trees on both sides of the equals sign in the equation. max_depth: Maximum depth of the expression trees on both sides of the equals sign in the equation. nbr_cases: The number of cases to generate. Yields: A dictionary {"inputs": input-list, "targets": target-list} where input-list are the tokens encoding the variable to solve for and the math equation, and target-list is a list of tokens encoding the resulting math expression after solving for the variable. Raises: ValueError: If `max_depth` < `min_depth`. def algebra_inverse(alphabet_size=26, min_depth=0, max_depth=2, nbr_cases=10000): """Generate the algebra inverse dataset. Each sample is a symbolic math equation involving unknown variables. The task is to solve for the given variable. The target is the resulting expression. Args: alphabet_size: How many possible variables there are. Max 52. min_depth: Minimum depth of the expression trees on both sides of the equals sign in the equation. max_depth: Maximum depth of the expression trees on both sides of the equals sign in the equation. nbr_cases: The number of cases to generate. Yields: A dictionary {"inputs": input-list, "targets": target-list} where input-list are the tokens encoding the variable to solve for and the math equation, and target-list is a list of tokens encoding the resulting math expression after solving for the variable. Raises: ValueError: If `max_depth` < `min_depth`. """ if max_depth < min_depth: raise ValueError("max_depth must be greater than or equal to min_depth. " "Got max_depth=%s, min_depth=%s" % (max_depth, min_depth)) alg_cfg = math_dataset_init(alphabet_size) for _ in range(nbr_cases): sample, target = generate_algebra_inverse_sample( alg_cfg.vlist, list(alg_cfg.ops.values()), alg_cfg.solve_ops, min_depth, max_depth) yield { "inputs": alg_cfg.int_encoder(sample), "targets": alg_cfg.int_encoder(target) }
Generate the algebra simplify dataset. Each sample is a symbolic math expression involving unknown variables. The task is to simplify the expression. The target is the resulting expression. Args: alphabet_size: How many possible variables there are. Max 52. min_depth: Minimum depth of the expression trees on both sides of the equals sign in the equation. max_depth: Maximum depth of the expression trees on both sides of the equals sign in the equation. nbr_cases: The number of cases to generate. Yields: A dictionary {"inputs": input-list, "targets": target-list} where input-list are the tokens encoding the expression to simplify, and target-list is a list of tokens encoding the resulting math expression after simplifying. Raises: ValueError: If `max_depth` < `min_depth`. def algebra_simplify(alphabet_size=26, min_depth=0, max_depth=2, nbr_cases=10000): """Generate the algebra simplify dataset. Each sample is a symbolic math expression involving unknown variables. The task is to simplify the expression. The target is the resulting expression. Args: alphabet_size: How many possible variables there are. Max 52. min_depth: Minimum depth of the expression trees on both sides of the equals sign in the equation. max_depth: Maximum depth of the expression trees on both sides of the equals sign in the equation. nbr_cases: The number of cases to generate. Yields: A dictionary {"inputs": input-list, "targets": target-list} where input-list are the tokens encoding the expression to simplify, and target-list is a list of tokens encoding the resulting math expression after simplifying. Raises: ValueError: If `max_depth` < `min_depth`. """ if max_depth < min_depth: raise ValueError("max_depth must be greater than or equal to min_depth. " "Got max_depth=%s, min_depth=%s" % (max_depth, min_depth)) alg_cfg = math_dataset_init(alphabet_size, digits=5) for _ in range(nbr_cases): sample, target = generate_algebra_simplify_sample( alg_cfg.vlist, list(alg_cfg.ops.values()), min_depth, max_depth) yield { "inputs": alg_cfg.int_encoder(sample), "targets": alg_cfg.int_encoder(target) }
Generate the calculus integrate dataset. Each sample is a symbolic math expression involving unknown variables. The task is to take the indefinite integral of the expression. The target is the resulting expression. Args: alphabet_size: How many possible variables there are. Max 26. min_depth: Minimum depth of the expression trees on both sides of the equals sign in the equation. max_depth: Maximum depth of the expression trees on both sides of the equals sign in the equation. nbr_cases: The number of cases to generate. Yields: A dictionary {"inputs": input-list, "targets": target-list} where input-list are the tokens encoding the variable to integrate with respect to and the expression to integrate, and target-list is a list of tokens encoding the resulting math expression after integrating. Raises: ValueError: If `max_depth` < `min_depth`, or if alphabet_size > 26. def calculus_integrate(alphabet_size=26, min_depth=0, max_depth=2, nbr_cases=10000): """Generate the calculus integrate dataset. Each sample is a symbolic math expression involving unknown variables. The task is to take the indefinite integral of the expression. The target is the resulting expression. Args: alphabet_size: How many possible variables there are. Max 26. min_depth: Minimum depth of the expression trees on both sides of the equals sign in the equation. max_depth: Maximum depth of the expression trees on both sides of the equals sign in the equation. nbr_cases: The number of cases to generate. Yields: A dictionary {"inputs": input-list, "targets": target-list} where input-list are the tokens encoding the variable to integrate with respect to and the expression to integrate, and target-list is a list of tokens encoding the resulting math expression after integrating. Raises: ValueError: If `max_depth` < `min_depth`, or if alphabet_size > 26. """ if max_depth < min_depth: raise ValueError("max_depth must be greater than or equal to min_depth. " "Got max_depth=%s, min_depth=%s" % (max_depth, min_depth)) # Don't allow alphabet to use capital letters. Those are reserved for function # names. if alphabet_size > 26: raise ValueError( "alphabet_size must not be greater than 26. Got %s." % alphabet_size) functions = {"log": "L"} alg_cfg = math_dataset_init(alphabet_size, digits=5, functions=functions) nbr_case = 0 while nbr_case < nbr_cases: try: sample, target = generate_calculus_integrate_sample( alg_cfg.vlist, list(alg_cfg.ops.values()), min_depth, max_depth, alg_cfg.functions) yield { "inputs": alg_cfg.int_encoder(sample), "targets": alg_cfg.int_encoder(target) } except: # pylint:disable=bare-except continue if nbr_case % 10000 == 0: print(" calculus_integrate: generating case %d." % nbr_case) nbr_case += 1
Returns True if `expr` is a subtree. def is_in(self, expr): """Returns True if `expr` is a subtree.""" if expr == self: return True is_in_left = is_in_expr(self.left, expr) is_in_right = is_in_expr(self.right, expr) return is_in_left or is_in_right
Preprocessing steps common to all models. def preprocess_example_common(example, mode, hparams): """Preprocessing steps common to all models.""" if "inputs" in example and hparams.max_input_seq_length > 0: example["inputs"] = example["inputs"][:hparams.max_input_seq_length] if hparams.prepend_mode != "none": if mode == tf.estimator.ModeKeys.PREDICT: example["partial_targets"] = tf.concat([example["inputs"], [0]], 0) else: example["targets"] = tf.concat( [example["inputs"], [0], example["targets"]], 0) if "targets" in example and hparams.max_target_seq_length > 0: example["targets"] = example["targets"][:hparams.max_target_seq_length] if hparams.split_to_length: new_example = {} for k, v in six.iteritems(example): if k == "targets" or k == "inputs": new_example[k] = tf.reshape(v, [-1, hparams.split_to_length, 1, 1]) else: tf.logging.warning("Dropping feature %s" % k) return tf.data.Dataset.from_tensor_slices(new_example) return example
Use input modality, vocab, and space id for target. def _copy_problem_hparams(p_hparams): """Use input modality, vocab, and space id for target.""" p = p_hparams # Duplicate input modality. p.modality["targets"] = p.modality["inputs"] # Duplicate input vocab size. p.vocab_size["targets"] = p.vocab_size["inputs"] # Duplicate input vocabulary. p.vocabulary["targets"] = p.vocabulary["inputs"] # Duplicate input space ids. p.target_space_id = p.input_space_id # Mark that p was reversed. p.was_copy = True
Swap input/output modalities, vocab, and space ids. def _reverse_problem_hparams(p_hparams): """Swap input/output modalities, vocab, and space ids.""" p = p_hparams # Swap modalities. # TODO(trandustin): Note this assumes target modalities have feature name # 'target', and each intended feature to swap has feature name 'input'. # In the future, remove need for this behavior. reversed_modality = {} for feature_name in p.modality: reversed_feature_name = feature_name.replace("target", "input") if "target" in feature_name and reversed_feature_name in p.modality: reversed_modality[feature_name] = p.modality[reversed_feature_name] reversed_modality[reversed_feature_name] = p.modality[feature_name] else: reversed_modality[feature_name] = p.modality[feature_name] p.modality = reversed_modality # Swap vocab sizes. reversed_vocab_size = {} for feature_name in p.vocab_size: reversed_feature_name = feature_name.replace("target", "input") if "target" in feature_name and reversed_feature_name in p.vocab_size: reversed_vocab_size[feature_name] = p.vocab_size[reversed_feature_name] reversed_vocab_size[reversed_feature_name] = p.vocab_size[feature_name] else: reversed_vocab_size[feature_name] = p.vocab_size[feature_name] p.vocab_size = reversed_vocab_size # Swap vocabularies. input_vocabulary = p.vocabulary.pop("inputs", None) target_vocabulary = p.vocabulary.pop("targets", None) if input_vocabulary is not None: p.vocabulary["targets"] = input_vocabulary if target_vocabulary is not None: p.vocabulary["inputs"] = target_vocabulary # Swap input/target space ids. input_space_id = p.input_space_id target_space_id = p.target_space_id if input_space_id is not None: p.target_space_id = input_space_id else: p.target_space_id = SpaceID.GENERIC if target_space_id is not None: p.input_space_id = target_space_id else: p.input_space_id = SpaceID.GENERIC # Mark that p was reversed. p.was_reversed = True
A set of basic model hyperparameters. def _default_hparams(): """A set of basic model hyperparameters.""" return hparam.HParams( # Use this parameter to get comparable perplexity numbers with different # tokenizations. This value should be set to the ratio of the number of # tokens in the test set according to the tokenization used to the number # of tokens in the test set in the "official" tokenization. For # example, if we are using a word-piece based model and we want to # compute per-word perplexity, then we set loss_multiplier to the number # of wordpieces per word in the test set. loss_multiplier=1.0, # Use this parameter to allow for larger sequences in the batch. Without # the use of this parameter, the size of the inner two dimensions will # be used to judge the sequence length. batch_size_multiplier=1, # During inference for autoregressive problems, if the batch_size is 1, # the inference will stop when the model predict a text_encoder.EOS_ID # token. stop_at_eos=False, # Modalities used to map from features to a space compatible with # chosen model architecture. It comprises key-value pairs of a feature # name (str) and its modality type. modality={}, vocab_size={}, # Identifiers used to tell the model which input/target space will be # expected. For example, it can tell that we expect French as characters # as output, or Spanish as sound. Spaces defined as constants in SpaceID # class. input_space_id=SpaceID.GENERIC, target_space_id=SpaceID.GENERIC)
Batch size in examples per TPU core. Args: model_hparams: model hyperparameters Returns: an integer def tpu_batch_size_per_shard(self, model_hparams): """Batch size in examples per TPU core. Args: model_hparams: model hyperparameters Returns: an integer """ if self.batch_size_means_tokens and not model_hparams.use_fixed_batch_size: return model_hparams.batch_size // self.max_length(model_hparams) else: return model_hparams.batch_size
Runtime preprocessing on the whole dataset. Return a tf.data.Datset -- the preprocessed version of the given one. By default this function calls preprocess_example. Args: dataset: the Dataset of already decoded but not yet preprocessed features. mode: tf.estimator.ModeKeys hparams: HParams, model hyperparameters interleave: bool, whether to use parallel_interleave, which is faster but will alter the order of samples non-deterministically, or flat_map, which is slower but will preserve the sample order. Returns: a Dataset def preprocess(self, dataset, mode, hparams, interleave=True): """Runtime preprocessing on the whole dataset. Return a tf.data.Datset -- the preprocessed version of the given one. By default this function calls preprocess_example. Args: dataset: the Dataset of already decoded but not yet preprocessed features. mode: tf.estimator.ModeKeys hparams: HParams, model hyperparameters interleave: bool, whether to use parallel_interleave, which is faster but will alter the order of samples non-deterministically, or flat_map, which is slower but will preserve the sample order. Returns: a Dataset """ def _preprocess(example): examples = self.preprocess_example(example, mode, hparams) if not isinstance(examples, tf.data.Dataset): examples = tf.data.Dataset.from_tensors(examples) return examples if interleave: dataset = dataset.apply( tf.data.experimental.parallel_interleave( _preprocess, sloppy=True, cycle_length=8)) else: dataset = dataset.flat_map(_preprocess) return dataset
Get filepattern for data files for mode. Matches mode to a suffix. * DatasetSplit.TRAIN: train * DatasetSplit.EVAL: dev * DatasetSplit.TEST: test * tf.estimator.ModeKeys.PREDICT: dev Args: data_dir: str, data directory. mode: DatasetSplit shard: int, if provided, will only read data from the specified shard. Returns: filepattern str def filepattern(self, data_dir, mode, shard=None): """Get filepattern for data files for mode. Matches mode to a suffix. * DatasetSplit.TRAIN: train * DatasetSplit.EVAL: dev * DatasetSplit.TEST: test * tf.estimator.ModeKeys.PREDICT: dev Args: data_dir: str, data directory. mode: DatasetSplit shard: int, if provided, will only read data from the specified shard. Returns: filepattern str """ path = os.path.join(data_dir, self.dataset_filename()) shard_str = "-%05d" % shard if shard is not None else "" if mode == DatasetSplit.TRAIN: suffix = "train" elif mode in [DatasetSplit.EVAL, tf.estimator.ModeKeys.PREDICT]: suffix = "dev" else: assert mode == DatasetSplit.TEST suffix = "test" return "%s-%s%s*" % (path, suffix, shard_str)
Returns problem_hparams. def get_hparams(self, model_hparams=None): """Returns problem_hparams.""" if self._hparams is not None: return self._hparams if model_hparams is None: model_hparams = default_model_hparams() if self._encoders is None: data_dir = (model_hparams and hasattr(model_hparams, "data_dir") and model_hparams.data_dir) or None self.get_feature_encoders(data_dir) hp = _default_hparams() ret = self.hparams(hp, model_hparams) if ret is not None: raise ValueError("The Problem subclass hparams function should mutate " "the defaults passed in and return None.") hp.add_hparam("vocabulary", self._encoders) hp.add_hparam("was_reversed", self._was_reversed) hp.add_hparam("was_copy", self._was_copy) if self._was_reversed: _reverse_problem_hparams(hp) if self._was_copy: _copy_problem_hparams(hp) self._hparams = hp return self._hparams
Reverse features between inputs and targets if the problem is '_rev'. def maybe_reverse_features(self, feature_map): """Reverse features between inputs and targets if the problem is '_rev'.""" if not self._was_reversed: return inputs = feature_map.pop("inputs", None) targets = feature_map.pop("targets", None) inputs_seg = feature_map.pop("inputs_segmentation", None) targets_seg = feature_map.pop("targets_segmentation", None) inputs_pos = feature_map.pop("inputs_position", None) targets_pos = feature_map.pop("targets_position", None) if inputs is not None: feature_map["targets"] = inputs if targets is not None: feature_map["inputs"] = targets if inputs_seg is not None: feature_map["targets_segmentation"] = inputs_seg if targets_seg is not None: feature_map["inputs_segmentation"] = targets_seg if inputs_pos is not None: feature_map["targets_position"] = inputs_pos if targets_pos is not None: feature_map["inputs_position"] = targets_pos
Build a Dataset for this problem. Args: mode: tf.estimator.ModeKeys; determines which files to read from. data_dir: directory that contains data files. num_threads: int, number of threads to use for decode and preprocess Dataset.map calls. output_buffer_size: int, how many elements to prefetch at end of pipeline. shuffle_files: whether to shuffle input files. Default behavior (i.e. when shuffle_files=None) is to shuffle if mode == TRAIN. hparams: HParams; hparams to be passed to Problem.preprocess_example and Problem.hparams. If None, will use a default set that is a no-op. preprocess: bool, whether to map the Dataset through Problem.preprocess_example. dataset_split: DatasetSplit, which split to read data from (TRAIN:"-train", EVAL:"-dev", "test":"-test"). Defaults to mode. shard: int, if provided, will only read data from the specified shard. partition_id: integer - which partition of the dataset to read from num_partitions: how many partitions in the dataset shuffle_buffer_size: if shuffle_files is True, this is the buffer size used to shuffle records. max_records: int, number of records to truncate to. Returns: Dataset containing dict<feature name, Tensor>. Raises: ValueError: if num_partitions is greater than the number of data files. def dataset(self, mode, data_dir=None, num_threads=None, output_buffer_size=None, shuffle_files=None, hparams=None, preprocess=True, dataset_split=None, shard=None, partition_id=0, num_partitions=1, shuffle_buffer_size=1024, max_records=-1): """Build a Dataset for this problem. Args: mode: tf.estimator.ModeKeys; determines which files to read from. data_dir: directory that contains data files. num_threads: int, number of threads to use for decode and preprocess Dataset.map calls. output_buffer_size: int, how many elements to prefetch at end of pipeline. shuffle_files: whether to shuffle input files. Default behavior (i.e. when shuffle_files=None) is to shuffle if mode == TRAIN. hparams: HParams; hparams to be passed to Problem.preprocess_example and Problem.hparams. If None, will use a default set that is a no-op. preprocess: bool, whether to map the Dataset through Problem.preprocess_example. dataset_split: DatasetSplit, which split to read data from (TRAIN:"-train", EVAL:"-dev", "test":"-test"). Defaults to mode. shard: int, if provided, will only read data from the specified shard. partition_id: integer - which partition of the dataset to read from num_partitions: how many partitions in the dataset shuffle_buffer_size: if shuffle_files is True, this is the buffer size used to shuffle records. max_records: int, number of records to truncate to. Returns: Dataset containing dict<feature name, Tensor>. Raises: ValueError: if num_partitions is greater than the number of data files. """ is_training = mode == tf.estimator.ModeKeys.TRAIN shuffle_files = shuffle_files or shuffle_files is None and is_training dataset_split = dataset_split or mode assert data_dir if hparams is None: hparams = default_model_hparams() if not hasattr(hparams, "data_dir"): hparams.add_hparam("data_dir", data_dir) if not hparams.data_dir: hparams.data_dir = data_dir # Construct the Problem's hparams so that items within it are accessible _ = self.get_hparams(hparams) data_filepattern = self.filepattern(data_dir, dataset_split, shard=shard) tf.logging.info("Reading data files from %s", data_filepattern) data_files = sorted(tf.contrib.slim.parallel_reader.get_data_files( data_filepattern)) # Functions used in dataset transforms below. `filenames` can be either a # `tf.string` tensor or `tf.data.Dataset` containing one or more filenames. def _load_records_and_preprocess(filenames): """Reads files from a string tensor or a dataset of filenames.""" # Load records from file(s) with an 8MiB read buffer. dataset = tf.data.TFRecordDataset(filenames, buffer_size=8 * 1024 * 1024) # Decode. dataset = dataset.map(self.decode_example, num_parallel_calls=num_threads) # Preprocess if requested. # Note that preprocessing should happen per-file as order may matter. if preprocess: dataset = self.preprocess(dataset, mode, hparams, interleave=shuffle_files) return dataset if len(data_files) < num_partitions: raise ValueError( "number of data files (%d) must be at least the number of hosts (%d)" % (len(data_files), num_partitions)) data_files = [f for (i, f) in enumerate(data_files) if i % num_partitions == partition_id] tf.logging.info( "partition: %d num_data_files: %d" % (partition_id, len(data_files))) if shuffle_files: mlperf_log.transformer_print(key=mlperf_log.INPUT_ORDER) random.shuffle(data_files) dataset = tf.data.Dataset.from_tensor_slices(tf.constant(data_files)) # Create data-set from files by parsing, pre-processing and interleaving. if shuffle_files: dataset = dataset.apply( tf.data.experimental.parallel_interleave( _load_records_and_preprocess, sloppy=True, cycle_length=8)) else: dataset = _load_records_and_preprocess(dataset) dataset = dataset.map( self.maybe_reverse_and_copy, num_parallel_calls=num_threads) dataset = dataset.take(max_records) ## Shuffle records only for training examples. if shuffle_files and is_training: dataset = dataset.shuffle(shuffle_buffer_size) if hparams.get("pack_dataset", False): dataset = generator_utils.pack_dataset( dataset, hparams.max_length, keys=["inputs", "targets"], use_custom_ops=hparams.get("use_custom_ops", False)) if output_buffer_size: dataset = dataset.prefetch(output_buffer_size) return dataset
Return a dict of Tensors from a serialized tensorflow.Example. def decode_example(self, serialized_example): """Return a dict of Tensors from a serialized tensorflow.Example.""" data_fields, data_items_to_decoders = self.example_reading_spec() # Necessary to rejoin examples in the correct order with the Cloud ML Engine # batch prediction API. data_fields["batch_prediction_key"] = tf.FixedLenFeature([1], tf.int64, 0) if data_items_to_decoders is None: data_items_to_decoders = { field: tf.contrib.slim.tfexample_decoder.Tensor(field) for field in data_fields } decoder = tf.contrib.slim.tfexample_decoder.TFExampleDecoder( data_fields, data_items_to_decoders) decode_items = list(sorted(data_items_to_decoders)) decoded = decoder.decode(serialized_example, items=decode_items) return dict(zip(decode_items, decoded))
Retrieve dict<feature name, FeatureInfo>. Must first call Problem.get_hparams or Problem.dataset to have the problem's internal hparams already constructed. Returns: dict<feature name, FeatureInfo> def feature_info(self): """Retrieve dict<feature name, FeatureInfo>. Must first call Problem.get_hparams or Problem.dataset to have the problem's internal hparams already constructed. Returns: dict<feature name, FeatureInfo> """ if self._feature_info is not None: return self._feature_info assert self._hparams is not None hp = self.get_hparams() if self.has_inputs: in_id = hp.input_space_id out_id = hp.target_space_id features = collections.defaultdict(FeatureInfo) for feature_name, modality_cls in six.iteritems(hp.modality): finfo = features[feature_name] finfo.modality = modality_cls finfo.vocab_size = hp.vocab_size[feature_name] vocabs = hp.vocabulary for name, encoder in six.iteritems(vocabs): features[name].encoder = encoder if self.has_inputs: features["inputs"].space_id = in_id features["targets"].space_id = out_id self._feature_info = features return features
Return input_fn wrapped for Estimator. def make_estimator_input_fn(self, mode, hparams, data_dir=None, force_repeat=False, prevent_repeat=False, dataset_kwargs=None): """Return input_fn wrapped for Estimator.""" def estimator_input_fn(params, config): return self.input_fn( mode, hparams, data_dir=data_dir, params=params, config=config, force_repeat=force_repeat, prevent_repeat=prevent_repeat, dataset_kwargs=dataset_kwargs) return estimator_input_fn
Which part of the training data to read. If there are multiple parallel calls to input_fn (multiple TPU hosts), then we want each one to read from a separate partition of the training data. Args: mode: tf.estimator.ModeKeys config: RunConfig params: A dict that contains parameters. Returns: partition_id: an integer num_partitions: an integer def _dataset_partition(self, mode, config, params): """Which part of the training data to read. If there are multiple parallel calls to input_fn (multiple TPU hosts), then we want each one to read from a separate partition of the training data. Args: mode: tf.estimator.ModeKeys config: RunConfig params: A dict that contains parameters. Returns: partition_id: an integer num_partitions: an integer """ if mode != tf.estimator.ModeKeys.TRAIN or not hasattr(config, "tpu_config"): # Reset in the case when using TPU but alternating TRAIN and EVAL. self._next_partition_id = 0 return 0, 1 phift = config.tpu_config.per_host_input_for_training # This is the mesh-tensorflow case. if (hasattr(tpu_config.InputPipelineConfig, "BROADCAST") and phift == tpu_config.InputPipelineConfig.BROADCAST): return 0, 1 if phift: num_hosts = (params["context"].num_hosts if "context" in params else config.tpu_config.num_shards // 8) num_partitions = max(num_hosts, 1) else: num_partitions = config.tpu_config.num_shards partition_id = getattr(self, "_next_partition_id", 0) self._next_partition_id = partition_id + 1 tf.logging.info("num_partitions = %d partition_id = %d" % (num_partitions, partition_id)) assert partition_id < num_partitions return partition_id, num_partitions
Builds input pipeline for problem. Args: mode: tf.estimator.ModeKeys hparams: HParams, model hparams data_dir: str, data directory; if None, will use hparams.data_dir params: dict, may include "batch_size" config: RunConfig; should have the data_parallelism attribute if not using TPU force_repeat: bool, whether to repeat the data even if not training prevent_repeat: bool, whether to not repeat when in training mode. Overrides force_repeat. dataset_kwargs: dict, if passed, will pass as kwargs to self.dataset method when called Returns: (features_dict<str name, Tensor feature>, Tensor targets) def input_fn(self, mode, hparams, data_dir=None, params=None, config=None, force_repeat=False, prevent_repeat=False, dataset_kwargs=None): """Builds input pipeline for problem. Args: mode: tf.estimator.ModeKeys hparams: HParams, model hparams data_dir: str, data directory; if None, will use hparams.data_dir params: dict, may include "batch_size" config: RunConfig; should have the data_parallelism attribute if not using TPU force_repeat: bool, whether to repeat the data even if not training prevent_repeat: bool, whether to not repeat when in training mode. Overrides force_repeat. dataset_kwargs: dict, if passed, will pass as kwargs to self.dataset method when called Returns: (features_dict<str name, Tensor feature>, Tensor targets) """ partition_id, num_partitions = self._dataset_partition(mode, config, params) is_training = mode == tf.estimator.ModeKeys.TRAIN if config and config.use_tpu: num_threads = 64 else: num_threads = data_reader.cpu_count() if is_training else 1 data_dir = data_dir or (hasattr(hparams, "data_dir") and hparams.data_dir) dataset_kwargs = dataset_kwargs or {} dataset_kwargs.update({ "mode": mode, "data_dir": data_dir, "num_threads": num_threads, "hparams": hparams, "partition_id": partition_id, "num_partitions": num_partitions, }) return data_reader.input_fn( self.dataset(**dataset_kwargs), self.filepattern(data_dir, mode), self.skip_random_fraction_when_training, self.batch_size_means_tokens, self.get_hparams().batch_size_multiplier, self.max_length(hparams), mode, hparams, data_dir=data_dir, params=params, config=config, force_repeat=force_repeat, prevent_repeat=prevent_repeat)
Input fn for serving export, starting from serialized example. def serving_input_fn(self, hparams, decode_hparams=None, use_tpu=False): """Input fn for serving export, starting from serialized example.""" mode = tf.estimator.ModeKeys.PREDICT serialized_example = tf.placeholder( dtype=tf.string, shape=[None], name="serialized_example") dataset = tf.data.Dataset.from_tensor_slices(serialized_example) dataset = dataset.map(self.decode_example) dataset = dataset.map(lambda ex: self.preprocess_example(ex, mode, hparams)) dataset = dataset.map(data_reader.cast_ints_to_int32) if use_tpu: padded_shapes = data_reader.pad_for_tpu(dataset.output_shapes, hparams, hparams.max_length) batch_size = 1 if not decode_hparams else getattr(decode_hparams, "batch_size", 1) dataset = dataset.padded_batch( batch_size, padded_shapes, drop_remainder=False) dataset = dataset.map( functools.partial(data_reader.pad_batch, batch_multiple=batch_size)) else: dataset = dataset.padded_batch( tf.shape(serialized_example, out_type=tf.int64)[0], dataset.output_shapes) dataset = dataset.map(data_reader.standardize_shapes) features = tf.data.experimental.get_single_element(dataset) if self.has_inputs: features.pop("targets", None) return tf.estimator.export.ServingInputReceiver( features=features, receiver_tensors=serialized_example)
Get hyper-parameters file path. def _get_hparams_path(): """Get hyper-parameters file path.""" hparams_path = None if FLAGS.output_dir: hparams_path = os.path.join(FLAGS.output_dir, "hparams.json") else: tf.logging.warning( "--output_dir not specified. Hyper-parameters will be infered from" "--hparams_set and --hparams only. These may not match training time" "hyper-parameters.") return hparams_path
Exports given checkpoint as tfhub module with given spec. def export_module_spec_with_checkpoint(module_spec, checkpoint_path, export_path, scope_prefix=""): """Exports given checkpoint as tfhub module with given spec.""" # The main requirement is that it is possible to know how to map from # module variable name to checkpoint variable name. # This is trivial if the original code used variable scopes, # but can be messy if the variables to export are interwined # with variables not export. with tf.Graph().as_default(): m = hub.Module(module_spec) assign_map = { scope_prefix + name: value for name, value in m.variable_map.items() } tf.train.init_from_checkpoint(checkpoint_path, assign_map) init_op = tf.initializers.global_variables() with tf.Session() as session: session.run(init_op) m.export(export_path, session)
Exports the last checkpoint from the directory as tfhub module. It creates the Module spec and signature (based on T2T problem information), which is later used to create and export the hub module. Module will be saved inside the ckpt_dir. Args: model_name: name of the model to be exported. hparams: T2T parameters, model graph will be based on them. decode_hparams: T2T parameters for decoding. problem: the name of the problem checkpoint_path: path to the checkpoint to be exported. export_dir: Directory to write the exported model to. def export_as_tfhub_module(model_name, hparams, decode_hparams, problem, checkpoint_path, export_dir): """Exports the last checkpoint from the directory as tfhub module. It creates the Module spec and signature (based on T2T problem information), which is later used to create and export the hub module. Module will be saved inside the ckpt_dir. Args: model_name: name of the model to be exported. hparams: T2T parameters, model graph will be based on them. decode_hparams: T2T parameters for decoding. problem: the name of the problem checkpoint_path: path to the checkpoint to be exported. export_dir: Directory to write the exported model to. """ def hub_module_fn(): """Creates the TF graph for the hub module.""" model_fn = t2t_model.T2TModel.make_estimator_model_fn( model_name, hparams, decode_hparams=decode_hparams, use_tpu=FLAGS.use_tpu) features = problem.serving_input_fn( hparams, decode_hparams, use_tpu=FLAGS.use_tpu).features # we must do a copy of the features, as the model_fn can add additional # entries there (like hyperparameter settings etc). original_features = features.copy() spec = model_fn(features, labels=None, mode=tf.estimator.ModeKeys.PREDICT) hub.add_signature( inputs=original_features, outputs=spec.export_outputs["serving_default"].outputs) # TFHub doesn't support the following collections. drop_collections = [tf.GraphKeys.LOSSES, tf.GraphKeys.SUMMARIES, tf.GraphKeys.LOCAL_VARIABLES] module_spec = hub.create_module_spec( hub_module_fn, drop_collections=drop_collections) # Loads the weights from the checkpoint using the model above # and saves it in the export_path. export_module_spec_with_checkpoint( module_spec, checkpoint_path=checkpoint_path, export_path=export_dir, scope_prefix="")
Build the graph required to fetch the attention weights. Args: hparams_set: HParams set to build the model with. model_name: Name of model. data_dir: Path to directory containing training data. problem_name: Name of problem. beam_size: (Optional) Number of beams to use when decoding a translation. If set to 1 (default) then greedy decoding is used. Returns: Tuple of ( inputs: Input placeholder to feed in ids to be translated. targets: Targets placeholder to feed to translation when fetching attention weights. samples: Tensor representing the ids of the translation. att_mats: Tensors representing the attention weights. ) def build_model(hparams_set, model_name, data_dir, problem_name, beam_size=1): """Build the graph required to fetch the attention weights. Args: hparams_set: HParams set to build the model with. model_name: Name of model. data_dir: Path to directory containing training data. problem_name: Name of problem. beam_size: (Optional) Number of beams to use when decoding a translation. If set to 1 (default) then greedy decoding is used. Returns: Tuple of ( inputs: Input placeholder to feed in ids to be translated. targets: Targets placeholder to feed to translation when fetching attention weights. samples: Tensor representing the ids of the translation. att_mats: Tensors representing the attention weights. ) """ hparams = trainer_lib.create_hparams( hparams_set, data_dir=data_dir, problem_name=problem_name) translate_model = registry.model(model_name)( hparams, tf.estimator.ModeKeys.EVAL) inputs = tf.placeholder(tf.int32, shape=(1, None, 1, 1), name="inputs") targets = tf.placeholder(tf.int32, shape=(1, None, 1, 1), name="targets") translate_model({ "inputs": inputs, "targets": targets, }) # Must be called after building the training graph, so that the dict will # have been filled with the attention tensors. BUT before creating the # inference graph otherwise the dict will be filled with tensors from # inside a tf.while_loop from decoding and are marked unfetchable. att_mats = get_att_mats(translate_model) with tf.variable_scope(tf.get_variable_scope(), reuse=True): samples = translate_model.infer({ "inputs": inputs, }, beam_size=beam_size)["outputs"] return inputs, targets, samples, att_mats
Get's the tensors representing the attentions from a build model. The attentions are stored in a dict on the Transformer object while building the graph. Args: translate_model: Transformer object to fetch the attention weights from. Returns: Tuple of attention matrices; ( enc_atts: Encoder self attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, inp_len, inp_len) dec_atts: Decoder self attetnion weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, out_len) encdec_atts: Encoder-Decoder attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, inp_len) ) def get_att_mats(translate_model): """Get's the tensors representing the attentions from a build model. The attentions are stored in a dict on the Transformer object while building the graph. Args: translate_model: Transformer object to fetch the attention weights from. Returns: Tuple of attention matrices; ( enc_atts: Encoder self attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, inp_len, inp_len) dec_atts: Decoder self attetnion weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, out_len) encdec_atts: Encoder-Decoder attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, inp_len) ) """ enc_atts = [] dec_atts = [] encdec_atts = [] prefix = "transformer/body/" postfix_self_attention = "/multihead_attention/dot_product_attention" if translate_model.hparams.self_attention_type == "dot_product_relative": postfix_self_attention = ("/multihead_attention/" "dot_product_attention_relative") postfix_encdec = "/multihead_attention/dot_product_attention" for i in range(translate_model.hparams.num_hidden_layers): enc_att = translate_model.attention_weights[ "%sencoder/layer_%i/self_attention%s" % (prefix, i, postfix_self_attention)] dec_att = translate_model.attention_weights[ "%sdecoder/layer_%i/self_attention%s" % (prefix, i, postfix_self_attention)] encdec_att = translate_model.attention_weights[ "%sdecoder/layer_%i/encdec_attention%s" % (prefix, i, postfix_encdec)] enc_atts.append(enc_att) dec_atts.append(dec_att) encdec_atts.append(encdec_att) return enc_atts, dec_atts, encdec_atts
Input str to features dict, ready for inference. def encode(self, input_str): """Input str to features dict, ready for inference.""" inputs = self.encoders["inputs"].encode(input_str) + [EOS_ID] batch_inputs = np.reshape(inputs, [1, -1, 1, 1]) # Make it 3D. return batch_inputs
List of ints to str. def decode(self, integers): """List of ints to str.""" integers = list(np.squeeze(integers)) return self.encoders["inputs"].decode(integers)
List of ints to list of str. def decode_list(self, integers): """List of ints to list of str.""" integers = list(np.squeeze(integers)) return self.encoders["inputs"].decode_list(integers)
Constructs the data needed for visualizing attentions. Args: sess: A tf.Session object. input_string: The input sentence to be translated and visualized. Returns: Tuple of ( output_string: The translated sentence. input_list: Tokenized input sentence. output_list: Tokenized translation. att_mats: Tuple of attention matrices; ( enc_atts: Encoder self attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, inp_len, inp_len) dec_atts: Decoder self attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, out_len) encdec_atts: Encoder-Decoder attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, inp_len) ) def get_vis_data_from_string(self, sess, input_string): """Constructs the data needed for visualizing attentions. Args: sess: A tf.Session object. input_string: The input sentence to be translated and visualized. Returns: Tuple of ( output_string: The translated sentence. input_list: Tokenized input sentence. output_list: Tokenized translation. att_mats: Tuple of attention matrices; ( enc_atts: Encoder self attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, inp_len, inp_len) dec_atts: Decoder self attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, out_len) encdec_atts: Encoder-Decoder attention weights. A list of `num_layers` numpy arrays of size (batch_size, num_heads, out_len, inp_len) ) """ encoded_inputs = self.encode(input_string) # Run inference graph to get the translation. out = sess.run(self.samples, { self.inputs: encoded_inputs, }) # Run the decoded translation through the training graph to get the # attention tensors. att_mats = sess.run(self.att_mats, { self.inputs: encoded_inputs, self.targets: np.reshape(out, [1, -1, 1, 1]), }) output_string = self.decode(out) input_list = self.decode_list(encoded_inputs) output_list = self.decode_list(out) return output_string, input_list, output_list, att_mats
Glow Hparams. def glow_hparams(): """Glow Hparams.""" hparams = common_hparams.basic_params1() hparams.clip_grad_norm = None hparams.weight_decay = 0.0 hparams.learning_rate_constant = 3e-4 hparams.batch_size = 32 # can be prev_level, prev_step or normal. # see: glow_ops.merge_level_and_latent_dist hparams.add_hparam("level_scale", "prev_level") hparams.add_hparam("n_levels", 3) hparams.add_hparam("n_bits_x", 8) hparams.add_hparam("depth", 32) # Activation - Relu or Gatu hparams.add_hparam("activation", "relu") # Coupling layer, additive or affine. hparams.add_hparam("coupling", "affine") hparams.add_hparam("coupling_width", 512) hparams.add_hparam("coupling_dropout", 0.0) hparams.add_hparam("top_prior", "single_conv") # init_batch_size denotes the number of examples used for data-dependent # initialization. A higher init_batch_size is required for training # stability especially when hparams.batch_size is low. hparams.add_hparam("init_batch_size", 256) hparams.add_hparam("temperature", 1.0) return hparams
Shifts and pads with zero along an axis. Example: shift_and_pad([1, 2, 3, 4], 2) --> [0, 0, 1, 2] shift_and_pad([1, 2, 3, 4], -2) --> [3, 4, 0, 0] Args: tensor: Tensor; to be shifted and padded. shift: int; number of positions to shift by. axis: int; along which axis to shift and pad. Returns: A Tensor with the same shape as the input tensor. def shift_and_pad(tensor, shift, axis=0): """Shifts and pads with zero along an axis. Example: shift_and_pad([1, 2, 3, 4], 2) --> [0, 0, 1, 2] shift_and_pad([1, 2, 3, 4], -2) --> [3, 4, 0, 0] Args: tensor: Tensor; to be shifted and padded. shift: int; number of positions to shift by. axis: int; along which axis to shift and pad. Returns: A Tensor with the same shape as the input tensor. """ shape = tensor.shape rank = len(shape) assert 0 <= abs(axis) < rank length = int(shape[axis]) assert 0 <= abs(shift) < length paddings = [(0, 0)] * rank begin = [0] * rank size = [-1] * rank if shift > 0: paddings[axis] = (shift, 0) size[axis] = length - shift elif shift < 0: paddings[axis] = (0, -shift) begin[axis] = -shift ret = tf.pad(tf.slice(tensor, begin, size), paddings) return ret
Set of hyperparameters. def transformer_aux_base(): """Set of hyperparameters.""" hparams = transformer.transformer_base() hparams.shared_embedding_and_softmax_weights = False hparams.add_hparam("shift_values", "1,2,3,4") return hparams
Set of hyperparameters. def transformer_aux_tiny(): """Set of hyperparameters.""" hparams = transformer.transformer_tiny() hparams.shared_embedding_and_softmax_weights = False hparams.add_hparam("shift_values", "1,2") return hparams
Given frame_logits from a per-pixel softmax, generate colors. def pixels_from_softmax(frame_logits, pure_sampling=False, temperature=1.0, gumbel_noise_factor=0.2): """Given frame_logits from a per-pixel softmax, generate colors.""" # If we're purely sampling, just sample each pixel. if pure_sampling or temperature == 0.0: return common_layers.sample_with_temperature(frame_logits, temperature) # Gumbel-sample from the pixel sofmax and average by pixel values. pixel_range = tf.to_float(tf.range(256)) for _ in range(len(frame_logits.get_shape().as_list()) - 1): pixel_range = tf.expand_dims(pixel_range, axis=0) frame_logits = tf.nn.log_softmax(frame_logits) gumbel_samples = discretization.gumbel_sample( common_layers.shape_list(frame_logits)) * gumbel_noise_factor frame = tf.nn.softmax((frame_logits + gumbel_samples) / temperature, axis=-1) result = tf.reduce_sum(frame * pixel_range, axis=-1) # Round on the forward pass, not on the backward one. return result + tf.stop_gradient(tf.round(result) - result)
Common HParams for next_frame models. def next_frame_base(): """Common HParams for next_frame models.""" hparams = common_hparams.basic_params1() # Loss cutoff. hparams.add_hparam("video_modality_loss_cutoff", 0.01) # Additional resizing the frames before feeding them to model. hparams.add_hparam("preprocess_resize_frames", None) # How many data points to suffle. Ideally should be part of problem not model! hparams.add_hparam("shuffle_buffer_size", 128) # Tiny mode. For faster tests. hparams.add_hparam("tiny_mode", False) # In case a model supports smaller/faster version. hparams.add_hparam("small_mode", False) # In case a model has stochastic version. hparams.add_hparam("stochastic_model", False) # Internal loss for recurrent models. hparams.add_hparam("internal_loss", True) # choose from: concat, multiplicative, multi_additive hparams.add_hparam("action_injection", "multi_additive") # Scheduled sampling method. Choose between # ground_truth_only, prediction_only, prob, count, prob_inverse_exp. hparams.add_hparam("scheduled_sampling_mode", "prediction_only") hparams.add_hparam("scheduled_sampling_decay_steps", 10000) hparams.add_hparam("scheduled_sampling_max_prob", 1.0) hparams.add_hparam("scheduled_sampling_k", 900.0) return hparams
Removes top level TimeLimit Wrapper. Removes TimeLimit Wrapper from top level if exists, throws error if any other TimeLimit Wrapper is present in stack. Args: env: environment Returns: the env with removed time limit wrapper. def remove_time_limit_wrapper(env): """Removes top level TimeLimit Wrapper. Removes TimeLimit Wrapper from top level if exists, throws error if any other TimeLimit Wrapper is present in stack. Args: env: environment Returns: the env with removed time limit wrapper. """ if isinstance(env, gym.wrappers.TimeLimit): env = env.env env_ = env while isinstance(env_, gym.Wrapper): if isinstance(env_, gym.wrappers.TimeLimit): raise ValueError("Can remove only top-level TimeLimit gym.Wrapper.") env_ = env_.env return env
Wraps a gym environment. see make_gym_env for details. def gym_env_wrapper(env, rl_env_max_episode_steps, maxskip_env, rendered_env, rendered_env_resize_to, sticky_actions): """Wraps a gym environment. see make_gym_env for details.""" # rl_env_max_episode_steps is None or int. assert ((not rl_env_max_episode_steps) or isinstance(rl_env_max_episode_steps, int)) wrap_with_time_limit = ((not rl_env_max_episode_steps) or rl_env_max_episode_steps >= 0) if wrap_with_time_limit: env = remove_time_limit_wrapper(env) if sticky_actions: env = StickyActionEnv(env) if maxskip_env: env = MaxAndSkipEnv(env) # pylint: disable=redefined-variable-type if rendered_env: env = RenderedEnv(env, resize_to=rendered_env_resize_to) if wrap_with_time_limit: env = gym.wrappers.TimeLimit( env, max_episode_steps=rl_env_max_episode_steps) return env