text
stringlengths
81
112k
Residual feed-forward layer with normalization at start. def ResidualFeedForward(feature_depth, feedforward_depth, dropout, mode): """Residual feed-forward layer with normalization at start.""" return layers.Residual( layers.LayerNorm(),...
Transformer encoder layer. The input to the encoder is a pair (embedded source, mask) where the mask is created from the original source to prevent attending to the padding part of the input. Args: feature_depth: int: depth of embedding feedforward_depth: int: depth of feed-forward layer num_head...
Transformer encoder. Args: vocab_size: int: vocab size num_classes: how many classes on output feature_depth: int: depth of embedding feedforward_depth: int: depth of feed-forward layer num_layers: int: number of encoder/decoder layers num_heads: int: number of attention heads dropout: f...
Transformer decoder layer. Args: feature_depth: int: depth of embedding feedforward_depth: int: depth of feed-forward layer num_heads: int: number of attention heads dropout: float: dropout rate (how much to drop out) mode: str: 'train' or 'eval' Returns: the layer. def DecoderLayer(feat...
Transformer language model (only uses the decoder part of Transformer). Args: vocab_size: int: vocab size feature_depth: int: depth of embedding feedforward_depth: int: depth of feed-forward layer num_layers: int: number of encoder/decoder layers num_heads: int: number of attention heads dro...
Transformer decoder layer operating on chunks. Args: feature_depth: int: depth of embedding feedforward_depth: int: depth of feed-forward layer num_heads: int: number of attention heads dropout: float: dropout rate (how much to drop out) chunk_selector: a function from chunk number to list of ch...
Transformer language model operating on chunks. The input to this model is a sequence presented as a list or tuple of chunks: (chunk1, chunk2, chunks3, ..., chunkN). Each chunk should have the same shape (batch, chunk-length) and together they represent a long sequence that's a concatenation chunk1,chunk2,....
Transformer model. Args: source_vocab_size: int: source vocab size target_vocab_size: int: target vocab size mode: str: 'train' or 'eval' num_layers: int: number of encoder/decoder layers feature_depth: int: depth of embedding feedforward_depth: int: depth of feed-forward layer num_heads...
Set of hyperparameters. def mtf_transformer_base(): """Set of hyperparameters.""" hparams = common_hparams.basic_params1() hparams.no_data_parallelism = True hparams.use_fixed_batch_size = True hparams.add_hparam("mtf_mode", True) hparams.batch_size = 64 hparams.max_length = 256 hparams.add_hparam("d_m...
Catch bugs locally... def mtf_transformer_tiny(): """Catch bugs locally...""" hparams = mtf_transformer_base() hparams.d_model = 128 hparams.d_ff = 512 hparams.batch_size = 8 hparams.encoder_layers = ["att", "drd"] * 2 hparams.decoder_layers = ["att", "enc_att", "drd"] * 2 hparams.num_heads = 8 # dat...
Config for language-model experiments. Train these on languagemodel_lm1b32k_packed for 136000 steps (10 epochs) The size parameter is an integer that controls the number of heads and the size of the size of the feedforward hidden layers. Increasing size by 1 doubles each of these. Results: size params...
Config for translation experiments. Train these on translate_enfr_wmt32k_packed for 154000 steps (3 epochs) The size parameter is an integer that controls the number of heads and the size of the size of the feedforward hidden layers. Increasing size by 1 doubles each of these. Args: size: an integer ...
Small language model to run on 1 TPU. Run this on 2x2 on languagemodel_lm1b32k_packed for 272000 steps (10 epochs) Results: params/10^9 log-ppl(per-token) 0.14 3.202 Returns: a hparams def mtf_transformer_lm_baseline(): """Small language model to run on 1 TPU. Run this on 2x...
Multihead scaled-dot-product attention with input/output transformations. Args: query_antecedent: a Tensor with shape [batch, length_q, channels] memory_antecedent: a Tensor with shape [batch, length_m, channels] or None bias: bias Tensor (see attention_bias()) total_key_depth: an integer total_v...
graph attention. Args: q: a Tensor with shape [batch, heads, length_q, depth_k] k: a Tensor with shape [batch, heads, length_kv, depth_k] v: a Tensor with shape [batch, heads, length_kv, depth_v] bias: bias Tensor (see attention_bias()) dropout_rate: a floating point number image_shapes: opti...
Helper function that computes transformation for keys and values. Let B be the number of batches. Let N be the number of nodes in the graph. Let D be the size of the node hidden states. Let K be the size of the attention keys/queries (total_key_depth). Let V be the size of the attention values (total_value_d...
Computes query, key and value for edge matrices. Let B be the number of batches. Let N be the number of nodes in the graph. Let D be the size of the node hidden states. Let K be the size of the attention keys/queries (total_key_depth). Let V be the size of the attention values (total_value_depth). Let T be...
Identical to sparse_ggnn except that each input has a batch dimension. B = The batch size. N = The number of nodes in each batch. H = The size of the hidden states. T = The number of edge types. Args: node_states: Initial states of each node in the graph. Shape: [B, N, H] adjacency_matrices: Adjacen...
One message-passing step for a GNN with a sparse adjacency matrix. Implements equation 2 (the message passing step) in [Li et al. 2015](https://arxiv.org/abs/1511.05493). N = The number of nodes in each batch. H = The size of the hidden states. T = The number of edge types. Args: node_states: Initial...
Multihead scaled-dot-product attention with input/output transformations. Let B be the number of batches. Let N be the number of nodes in the graph. Let D be the size of the node hidden states. Let K be the size of the attention keys/queries (total_key_depth). Let V be the size of the attention values (total...
Dot product attention with edge vectors. Let B be the number of batches. Let N be the number of nodes in the graph. Let K be the size of the attention keys/queries. Let V be the size of the attention values. Let T be the total number of transforms (num_transforms). Args: q: The query Tensor of shape [...
ggnn version of the MPNN from Gilmer et al. Let B be the number of batches. Let D be the size of the node hidden states. Let K be the size of the attention keys/queries. Let V be the size of the output of the ggnn. Let T be the number of transforms / edge types. Args: node_states: The value Tensor of ...
Compute values. If edge compatibilities is just adjacency, we get ggnn. Args: edge_compatibility: A tensor of shape [batch, num_transforms, length, depth] v: A tensor of shape [batch, num_transforms, length, depth] Returns: output: A [batch, length, depth] tensor def compute_values(edge_compatibility...
Precompute the a_in and a_out tensors. (we don't want to add to the graph everytime _fprop is called) Args: adjacency: placeholder of real valued vectors of shape [B, L, L, E] hparams: HParams object Returns: edge_matrices: [batch, L * D, L * D] the dense matrix for message passing viewed as a bl...
Computes a_t from h_{t-1}, see bottom of page 3 in the paper. Args: node_states: [B, L, D] tensor (h_{t-1}) edge_matrices (tf.float32): [B, L*D, L*D] Returns: messages (tf.float32): [B, L, D] For each pair of nodes in the graph a message is sent along both the incoming and outgoing edge. ...
Helper: build tf.Example from (string -> int/float/str list) dictionary. def to_example(dictionary): """Helper: build tf.Example from (string -> int/float/str list) dictionary.""" features = {} for (k, v) in six.iteritems(dictionary): if not v: raise ValueError("Empty generated field: %s" % str((k, v))...
generate_files but with a single writer writing to shard task_id. def generate_files_distributed(generator, output_name, output_dir, num_shards=1, max_cases=None, t...
Generate cases from a generator and save as TFRecord files. Generated cases are transformed to tf.Example protos and saved as TFRecords in sharded files named output_dir/output_name-00..N-of-00..M=num_shards. Args: generator: a generator yielding (string -> int/float/str list) dictionaries. output_filen...
Report hook for download progress. Args: count: current block number block_size: block size total_size: total size def download_report_hook(count, block_size, total_size): """Report hook for download progress. Args: count: current block number block_size: block size total_size: total si...
Download filename from uri unless it's already in directory. Copies a remote file to local if that local file does not already exist. If the local file pre-exists this function call, it does not check that the local file is a copy of the remote. Remote filenames can be filepaths, any URI readable by tensorfl...
Download filename from Google drive unless it's already in directory. Args: directory: path to the directory that will be used. filename: name of the file to download to (do nothing if it already exists). url: URL to download from. Returns: The path to the downloaded file. def maybe_download_from...
Unzips from gz_path into new_path. Args: gz_path: path to the zipped file. new_path: path to where the file will be unzipped. def gunzip_file(gz_path, new_path): """Unzips from gz_path into new_path. Args: gz_path: path to the zipped file. new_path: path to where the file will be unzipped. ""...
Inner implementation for vocab generators. Args: data_dir: The base directory where data and vocab files are stored. If None, then do not save the vocab even if it doesn't exist. vocab_filename: relative filename where vocab file is stored vocab_size: target size of the vocabulary constructed by Su...
Generate a vocabulary from the datasets in sources. def get_or_generate_vocab(data_dir, tmp_dir, vocab_filename, vocab_size, sources, file_byte_budget=1e6, max_subtoken_length=None): """Generate a vocabulary from the datasets in sources.""" vocab_generator = gen...
Generate lines for vocabulary generation. def generate_lines_for_vocab(tmp_dir, sources, file_byte_budget=1e6): """Generate lines for vocabulary generation.""" tf.logging.info("Generating vocab from: %s", str(sources)) for source in sources: url = source[0] filename = os.path.basename(url) compressed...
r"""Generate a vocabulary from a tabbed source file. The source is a file of source, target pairs, where each line contains a source string and a target string, separated by a tab ('\t') character. The index parameter specifies 0 for the source or 1 for the target. Args: data_dir: path to the data directo...
Generate a vocabulary from txt files with example-per-line. def get_or_generate_txt_vocab(data_dir, vocab_filename, vocab_size, filepatterns): """Generate a vocabulary from txt files with example-per-line.""" if isinstance(filepatterns, str): filepatterns = [filepatterns] def g...
Shuffle a single file of records. Args: fname: a string extra_fn: an optional function from list of TFRecords to list of TFRecords to be called after shuffling. def _shuffle_single(fname, extra_fn=None): """Shuffle a single file of records. Args: fname: a string extra_fn: an optional func...
Shuffles the dataset. Args: filenames: a list of strings extra_fn: an optional function from list of records to list of records to be called after shuffling a file. def shuffle_dataset(filenames, extra_fn=None): """Shuffles the dataset. Args: filenames: a list of strings extra_fn: an opti...
Pack examples into longer examples. If has_inputs=False, we are packing single-sequence examples with targets only and no inputs. In this case, we concatenate the targets from several examples to form each new example. We insert a number of zeros for spacing between the original sequences. This is to help...
Helper-function for packing a dataset which has already been batched. See pack_dataset() Relies on custom ops which require a custom compiled binary. Faster than _pack_with_tf_ops(), and denser packing. Args: dataset: a dataset containing padded batches of examples. keys: a list of strings (must have...
Make a temporary directory. def make_tmp_dir(suffix="", prefix="tmp", dir=None): # pylint: disable=redefined-builtin """Make a temporary directory.""" if dir is None: return tempfile.mkdtemp(suffix, prefix, dir) else: while True: rand_term = random.randint(1, 9999) tmp_dir = os.path.join(dir...
Iterate over the records on disk for the Problem. def tfrecord_iterator_for_problem(problem, data_dir, dataset_split=tf.estimator.ModeKeys.TRAIN): """Iterate over the records on disk for the Problem.""" filenames = tf.gfile.Glob(problem.filepattern(data_dir, mode=dataset_split)) ...
Yields records from TFRecord files. Args: filenames: list<str>, list of TFRecord filenames to read from. gzipped: bool, whether the TFRecord files are gzip-encoded. example_spec: dict<str feature name, tf.VarLenFeature/tf.FixedLenFeature>, if provided, will parse each record as a tensorflow.Example...
Create a fill-in-the-blanks training example from text. Split on spaces, then cut into segments at random points. Alternate segments are assigned to the two output strings. separator_symbol separates segments within each of the outputs. example: text="The quick brown fox jumps over the lazy dog." ret...
The core Neural GPU. def neural_gpu_body(inputs, hparams, name=None): """The core Neural GPU.""" with tf.variable_scope(name, "neural_gpu"): def step(state, inp): # pylint: disable=missing-docstring x = tf.nn.dropout(state, 1.0 - hparams.dropout) for layer in range(hparams.num_hidden_layers): ...
Improved Neural GPU as in https://arxiv.org/abs/1702.08727. def diagonal_neural_gpu(inputs, hparams, name=None): """Improved Neural GPU as in https://arxiv.org/abs/1702.08727.""" with tf.variable_scope(name, "diagonal_neural_gpu"): def step(state_tup, inp): """Single step of the improved Neural GPU.""" ...
Helper to determine the shape of reorder output. def _reorder_shape(input_shape, output=None): # pylint: disable=invalid-name """Helper to determine the shape of reorder output.""" if output is None: return input_shape return base.nested_map(output, lambda i: input_shape[i])
Reorder a tuple into another tuple. For example, we can re-order (x, y) into (y, x) or even (y, (x, y), y). The output argument specifies how to re-order, using integers that refer to indices in the input tuple. For example, if input = (x, y, z) then Reorder(input, output=(1, 0, 2)) = (y, x, z) ...
Helper: sum a list of arrays or nested arrays. def _nested_op(inputs, op): # pylint: disable=invalid-name """Helper: sum a list of arrays or nested arrays.""" # First the simple non-nested case. if not isinstance(inputs[0], (list, tuple)): return op(inputs) # In the nested case, sum on each axis separatel...
Implements a gating function on a (memory, gate, candidate) tuple. Final update is memory * gate + (1-gate) * candidate This gating equation may also be referred to as Highway Network. Highway Networks: https://arxiv.org/abs/1505.00387 Args: x: A tuple of (memory, gate, candidate) Returns: The res...
Helper to determine the shape of Concatenate output. def _concatenate_shape(input_shape, axis=-1): # pylint: disable=invalid-name """Helper to determine the shape of Concatenate output.""" ax = axis % len(input_shape[0]) concat_size = sum(shape[ax] for shape in input_shape) out_shape = input_shape[0][:ax] + (...
Constructs a residual version of layers, summing input to layers output. def Residual(*layers, **kwargs): """Constructs a residual version of layers, summing input to layers output.""" shortcut = kwargs.get('shortcut', Identity()) # pylint: disable=no-value-for-parameter if len(layers) > 1: return Serial( ...
Train. def train( self, env_fn, hparams, simulated, save_continuously, epoch, sampling_temp=1.0, num_env_steps=None, env_step_multiplier=1, eval_env_fn=None, report_fn=None ): """Train.""" raise NotImplementedError()
Adds default hparams for all of the variants of the Universal Transformer. Args: hparams: default hparams (usually one of the standard hparams from transformer model (like "transformer_base") Returns: hparams with default values for Universal Transformers hyper-parameters def update_hparams_for_uni...
Base parameters for Universal Transformer. def universal_transformer_base(): """Base parameters for Universal Transformer.""" hparams = transformer.transformer_base() # To have a similar capacity to the transformer_base with 6 layers, # we need to increase the size of the UT's layer # since, in fact, UT has ...
Multi-layer config for adaptive Transformer on TPU. def adaptive_universal_transformer_multilayer_tpu(): """Multi-layer config for adaptive Transformer on TPU.""" hparams = adaptive_universal_transformer_base_tpu() hparams.num_inrecurrence_layers = 2 hparams.mix_with_transformer = "before_ut,after_ut" hparam...
Multi-layer config for adaptive Transformer with hard attention. def adaptive_universal_transformer_multilayer_hard(): """Multi-layer config for adaptive Transformer with hard attention.""" hparams = adaptive_universal_transformer_multilayer_tpu() hparams.batch_size = 256 hparams.hard_attention_k = 8 hparams...
Range of hyperparameters. def universal_transformer_base_range(rhp): """Range of hyperparameters.""" # After starting from base, set intervals for some parameters. rhp.set_discrete("num_rec_steps", [6, 8, 10]) rhp.set_discrete("hidden_size", [1024, 2048, 4096]) rhp.set_discrete("filter_size", [2048, 4096, 81...
Range of hyperparameters. def adaptive_universal_transformer_base_range(rhp): """Range of hyperparameters.""" # After starting from base, set intervals for some parameters. rhp.set_discrete("act_max_steps", [8, 16, 32]) rhp.set_float("act_loss_weight", 0.0, 0.5) rhp.set_discrete("hidden_size", [1024, 2048, 4...
Split channels in 3 parts. Shifts 1st and 3rd sections to left/right. def DiagonalGate(x, params, **kwargs): """Split channels in 3 parts. Shifts 1st and 3rd sections to left/right.""" del params del kwargs # x : [batch, 1, length, depth] x = np.pad( x, [(0, 0), (0, 0), (1, 1), (0, 0)], mode='constant'...
Build convolutional GRU with diagonal gating as in ImprovedNGPU. def ConvDiagonalGRU(units, kernel_size=(3, 3)): """Build convolutional GRU with diagonal gating as in ImprovedNGPU.""" def BuildConv(): return layers.Conv(filters=units, kernel_size=kernel_size, padding='SAME') return layers.GeneralGRUCell( ...
Implementation of Neural GPU: https://arxiv.org/abs/1702.08727. Args: feature_depth: Number of memory channels steps: Number of times depthwise recurrence steps. vocab_size: Vocabulary size. Returns: A NeuralGPU Stax model. def NeuralGPU(feature_depth=96, steps=16, vocab_size=2): """Implementat...
Strip ids_to_strip from the end ids. def strip_ids(ids, ids_to_strip): """Strip ids_to_strip from the end ids.""" ids = list(ids) while ids and ids[-1] in ids_to_strip: ids.pop() return ids
Escape away underscores and OOV characters and append '_'. This allows the token to be expressed as the concatenation of a list of subtokens from the vocabulary. The underscore acts as a sentinel which allows us to invertibly concatenate multiple such lists. Args: token: A unicode string to be escaped. ...
Transform a human-readable string into a sequence of int ids. The ids should be in the range [num_reserved_ids, vocab_size). Ids [0, num_reserved_ids) are reserved. EOS is not appended. Args: s: human-readable string to be converted. Returns: ids: list of integers def encode(self, s...
Transform a sequence of int ids into a human-readable string. EOS is not expected in ids. Args: ids: list of integers to be converted. strip_extraneous: bool, whether to strip off extraneous tokens (EOS and PAD). Returns: s: human-readable string. def decode(self, ids, strip_ex...
Transform a sequence of int ids into a their string versions. This method supports transforming individual input/output ids to their string versions so that sequence to/from text conversions can be visualized in a human readable format. Args: ids: list of integers to be converted. Returns: ...
Converts a space-separated string of tokens to a list of ids. def encode(self, s): """Converts a space-separated string of tokens to a list of ids.""" sentence = s tokens = sentence.strip().split() if self._replace_oov is not None: tokens = [t if t in self._token_to_id else self._replace_oov ...
Load vocab from a file. Args: filename: The file to load vocabulary from. def _init_vocab_from_file(self, filename): """Load vocab from a file. Args: filename: The file to load vocabulary from. """ with tf.gfile.Open(filename) as f: tokens = [token.strip() for token in f.readlin...
Initialize tokens from a list of tokens. It is ok if reserved tokens appear in the vocab list. They will be removed. The set of tokens in vocab_list should be unique. Args: vocab_list: A list of tokens. def _init_vocab_from_list(self, vocab_list): """Initialize tokens from a list of tokens. ...
Initialize vocabulary with tokens from token_generator. def _init_vocab(self, token_generator, add_reserved_tokens=True): """Initialize vocabulary with tokens from token_generator.""" self._id_to_token = {} non_reserved_start_index = 0 if add_reserved_tokens: self._id_to_token.update(enumerate(...
Write vocab file to disk. Vocab files have one token per line. The file ends in a newline. Reserved tokens are written to the vocab file as well. Args: filename: Full path of the file to store the vocab to. def store_to_file(self, filename): """Write vocab file to disk. Vocab files have on...
Converts a sequence of subtoken ids to a native string. Args: ids: a list of integers in the range [0, vocab_size) strip_extraneous: bool, whether to strip off extraneous tokens (EOS and PAD). Returns: a native string def decode(self, ids, strip_extraneous=False): """Converts a ...
Converts a list of tokens to a list of subtoken ids. Args: tokens: a list of strings. Returns: a list of integers in the range [0, vocab_size) def _tokens_to_subtoken_ids(self, tokens): """Converts a list of tokens to a list of subtoken ids. Args: tokens: a list of strings. Retu...
Converts token to a list of subtoken ids. Args: token: a string. Returns: a list of integers in the range [0, vocab_size) def _token_to_subtoken_ids(self, token): """Converts token to a list of subtoken ids. Args: token: a string. Returns: a list of integers in the range [...
Converts a list of subtoken ids to a list of tokens. Args: subtokens: a list of integers in the range [0, vocab_size) Returns: a list of strings. def _subtoken_ids_to_tokens(self, subtokens): """Converts a list of subtoken ids to a list of tokens. Args: subtokens: a list of integers...
Converts a subtoken integer ID to a subtoken string. def _subtoken_id_to_subtoken_string(self, subtoken): """Converts a subtoken integer ID to a subtoken string.""" if 0 <= subtoken < self.vocab_size: return self._all_subtoken_strings[subtoken] return u""
Converts an escaped token string to a list of subtoken strings. Args: escaped_token: An escaped token as a unicode string. Returns: A list of subtokens as unicode strings. def _escaped_token_to_subtoken_strings(self, escaped_token): """Converts an escaped token string to a list of subtoken str...
Converts an escaped token string to a list of subtoken IDs. Args: escaped_token: An escaped token as a unicode string. Returns: A list of subtoken IDs as integers. def _escaped_token_to_subtoken_ids(self, escaped_token): """Converts an escaped token string to a list of subtoken IDs. Args:...
Builds a SubwordTextEncoder from the generated text. Args: generator: yields text. target_size: int, approximate vocabulary size to create. max_subtoken_length: Maximum length of a subtoken. If this is not set, then the runtime and memory use of creating the vocab is quadratic in ...
Builds a SubwordTextEncoder that has `vocab_size` near `target_size`. Uses simple recursive binary search to find a minimum token count that most closely matches the `target_size`. Args: target_size: Desired vocab_size to approximate. token_counts: A dictionary of token counts, mapping string ...
Train a SubwordTextEncoder based on a dictionary of word counts. Args: token_counts: a dictionary of Unicode strings to int. min_count: an integer - discard subtokens with lower counts. num_iterations: an integer. how many iterations of refinement. reserved_tokens: List of reserved tokens....
Debugging dump of the current subtoken vocabulary. def dump(self): """Debugging dump of the current subtoken vocabulary.""" subtoken_strings = [(i, s) for s, i in six.iteritems(self._subtoken_string_to_id)] print(u", ".join(u"{0} : '{1}'".format(i, s) for i, s i...
Initialize token information from a list of subtoken strings. Args: subtoken_strings: a list of subtokens reserved_tokens: List of reserved tokens. We must have `reserved_tokens` as None or the empty list, or else the global variable `RESERVED_TOKENS` must be a prefix of `reserved_token...
Load from a file object. Args: f: File object to load vocabulary from def _load_from_file_object(self, f): """Load from a file object. Args: f: File object to load vocabulary from """ subtoken_strings = [] for line in f: s = line.strip() # Some vocab files wrap words i...
Load from a vocab file. def _load_from_file(self, filename): """Load from a vocab file.""" if not tf.gfile.Exists(filename): raise ValueError("File %s not found" % filename) with tf.gfile.Open(filename) as f: self._load_from_file_object(f)
Transform a string with a filename into a list of RGB integers. Args: s: path to the file with an image. Returns: ids: list of integers def encode(self, s): """Transform a string with a filename into a list of RGB integers. Args: s: path to the file with an image. Returns: ...
Transform a sequence of int ids into an image file. Args: ids: list of integers to be converted. strip_extraneous: unused Returns: Path to the temporary file where the image was saved. Raises: ValueError: if the ids are not of the appropriate size. def decode(self, ids, strip_ext...
Transform sequence of float values into string (float values). Args: ids: array of floats to be converted. strip_extraneous: unused Returns: String having space separated float values. Raises: ValueError: if the ids are not of the appropriate size. def decode(self, ids, strip_ext...
Helper utility to make a tiled field of images from numpy arrays. Args: images: Image tensor in shape [N, W, H, C]. rows: Number of images per row in tiled image. cols: Number of images per column in tiled image. Returns: A tiled image of shape [W * rows, H * cols, C]. Truncates incomplete row...
Convert an operative config string to markdown format. def markdownify_operative_config_str(string): """Convert an operative config string to markdown format.""" # TODO(b/37527917): Total hack below. Implement more principled formatting. def process(line): """Convert a single line to markdown format.""" ...
Close SummaryWriter. Final! def close(self): """Close SummaryWriter. Final!""" if not self._closed: self._event_writer.close() self._closed = True del self._event_writer
Saves scalar value. Args: tag: str: label for this data value: int/float: number to log step: int: training step def scalar(self, tag, value, step=None): """Saves scalar value. Args: tag: str: label for this data value: int/float: number to log step: int: training step...
Saves RGB image summary from onp.ndarray [H,W], [H,W,1], or [H,W,3]. Args: tag: str: label for this data image: ndarray: [H,W], [H,W,1], [H,W,3] save image in greyscale or colors/ step: int: training step def image(self, tag, image, step=None): """Saves RGB image summary from onp.ndarray [H,...
Saves (rows, cols) tiled images from onp.ndarray. If either rows or cols aren't given, they are determined automatically from the size of the image batch, if neither are given a long column of images is produced. This truncates the image batch rather than padding if it doesn't fill the final row. ...
Saves matplotlib plot output to summary image. Args: tag: str: label for this data mpl_plt: matplotlib stateful pyplot object with prepared plotting state step: int: training step close_plot: bool: automatically closes plot def plot(self, tag, mpl_plt, step=None, close_plot=True): """S...
Saves audio. NB: single channel only right now. Args: tag: str: label for this data audiodata: ndarray [Nsamples,]: data between (-1.0,1.0) to save as wave step: int: training step sample_rate: sample rate of passed in audio buffer def audio(self, tag, audiodata, step=None, sample_rat...
Saves histogram of values. Args: tag: str: label for this data values: ndarray: will be flattened by this routine bins: number of bins in histogram, or array of bins for onp.histogram step: int: training step def histogram(self, tag, values, bins, step=None): """Saves histogram of valu...
Saves a text summary. Args: tag: str: label for this data textdata: string, or 1D/2D list/numpy array of strings step: int: training step Note: markdown formatting is rendered by tensorboard. def text(self, tag, textdata, step=None): """Saves a text summary. Args: tag: str: la...