def AddMultiCurveSubplot(fig,
tensors,
paddings,
labels,
xlabels=None,
**kwargs):
"""Adds a multi curve subplot to Matplotlib figure.
Plots one line for each entry in tensors and assigns a plot label legend.
Args:
fig: The Matplotlib figure.
tensors: List of tensors of shape [batch, length]
paddings: Paddings for 'tensors' with shape [batch, length] with 0. in valid
positions and 1. in invalid.
labels: A list of tensor names (strings) of the same length as 'tensors'.
xlabels: A string tensor of shape [batch] with an xlabel per batch.
**kwargs: With optional, title, xlabel, ylabel, fontsize.
"""
data = []
row_labels = []
for t, l in zip(tensors, labels):
if t is not None:
data.append(py_utils.ApplyPadding(paddings, t))
row_labels.append(l)
shape = py_utils.GetShape(data[0], 2)
data = tf.reshape(tf.concat(data, -1), [shape[0], len(data), shape[1]])
args = [data, py_utils.LengthsFromPaddings(paddings)]
if xlabels is not None:
args.append(xlabels)
fig.AddSubplot(args,
plot_func=_AddMultiCurveRowPlots,
row_labels=row_labels,
**kwargs)
def tokenize_words(words_t):
padded_tokenized_t, _, paddings_t = str_to_vocab_tokens(
labels=words_t,
maxlen=longest_word_length,
append_eos=True,
pad_to_maxlen=True,
vocab_filepath=FLAGS.in_units_txt,
load_token_ids_from_vocab=False,
delimiter="",
)
# Either lengths or paddings are incorrect.
lengths_t = py_utils.LengthsFromPaddings(paddings_t)
ragged_tokenized_t = tf.RaggedTensor.from_tensor(padded_tokenized_t,
lengths=lengths_t)
# Drop start-of-sentence-token
ragged_tokenized_t = ragged_tokenized_t[:, 1:]
lengths_t -= 1
letters_t = vocab_id_to_token(
id=ragged_tokenized_t.flat_values,
vocab=vocab_tokens,
load_token_ids_from_vocab=False,
)
ragged_letters_t = tf.RaggedTensor.from_row_lengths(
letters_t, lengths_t)
# Is capatilizationt he problem?
return ragged_tokenized_t, ragged_letters_t