def get_predictions(reader_outputs, params):
"""Get predictions."""
tokenization_info = bert_utils.get_tokenization_info(
params["reader_module_path"])
with tf.io.gfile.GFile(tokenization_info["vocab_file"]) as vocab_file:
vocab = tf.constant([l.strip() for l in vocab_file.readlines()])
# []
predicted_block_index = tf.argmax(tf.reduce_max(reader_outputs.logits, 1))
predicted_candidate = tf.argmax(tf.reduce_max(reader_outputs.logits, 0))
predicted_block = tf.gather(reader_outputs.blocks, predicted_block_index)
predicted_orig_block = tf.gather(reader_outputs.orig_blocks,
predicted_block_index)
predicted_orig_tokens = tf.gather(reader_outputs.orig_tokens,
predicted_block_index)
predicted_orig_start = tf.gather(
tf.gather(reader_outputs.candidate_orig_starts, predicted_block_index),
predicted_candidate)
predicted_orig_end = tf.gather(
tf.gather(reader_outputs.candidate_orig_ends, predicted_block_index),
predicted_candidate)
predicted_orig_answer = tf.reduce_join(
predicted_orig_tokens[predicted_orig_start:predicted_orig_end + 1],
separator=" ")
predicted_token_ids = tf.gather(reader_outputs.token_ids,
predicted_block_index)
predicted_tokens = tf.gather(vocab, predicted_token_ids)
predicted_start = tf.gather(reader_outputs.candidate_starts,
predicted_candidate)
predicted_end = tf.gather(reader_outputs.candidate_ends,
predicted_candidate)
predicted_normalized_answer = tf.reduce_join(
predicted_tokens[predicted_start:predicted_end + 1], separator=" ")
def _get_final_text(pred_text, orig_text):
pred_text = six.ensure_text(pred_text, errors="ignore")
orig_text = six.ensure_text(orig_text, errors="ignore")
return squad_lib.get_final_text(
pred_text=pred_text,
orig_text=orig_text,
do_lower_case=tokenization_info["do_lower_case"])
predicted_answer = tf.py_func(
func=_get_final_text,
inp=[predicted_normalized_answer, predicted_orig_answer],
Tout=tf.string)
return dict(block_index=predicted_block_index,
candidate=predicted_candidate,
block=predicted_block,
orig_block=predicted_orig_block,
orig_tokens=predicted_orig_tokens,
orig_start=predicted_orig_start,
orig_end=predicted_orig_end,
answer=predicted_answer)
def get_predictions(reader_outputs):
"""Get predictions."""
# []
predicted_block_index = tf.argmax(tf.reduce_max(reader_outputs.logits, 1))
predicted_candidate = tf.argmax(tf.reduce_max(reader_outputs.logits, 0))
predicted_block = tf.gather(reader_outputs.blocks, predicted_block_index)
predicted_orig_block = tf.gather(reader_outputs.orig_blocks,
predicted_block_index)
predicted_orig_tokens = tf.gather(reader_outputs.orig_tokens,
predicted_block_index)
predicted_orig_start = tf.gather(
tf.gather(reader_outputs.candidate_orig_start, predicted_block_index),
predicted_candidate)
predicted_orig_end = tf.gather(
tf.gather(reader_outputs.candidate_orig_end, predicted_block_index),
predicted_candidate)
predicted_answer = tf.reduce_join(
predicted_orig_tokens[predicted_orig_start:predicted_orig_end + 1],
separator=" ")
return dict(
block_index=predicted_block_index,
candidate=predicted_candidate,
block=predicted_block,
orig_block=predicted_orig_block,
orig_tokens=predicted_orig_tokens,
orig_start=predicted_orig_start,
orig_end=predicted_orig_end,
answer=predicted_answer)
def from_tokens(raw, lookup_):
gathered = tf.gather(lookup_, tf.cast(raw, tf.int32))
joined = tf.regex_replace(tf.reduce_join(gathered, axis=1), b"<EOS>.*",
b"")
cleaned = tf.regex_replace(joined, b"_", b" ")
tokens = tf.string_split(cleaned, " ")
return tokens
def from_characters(raw, lookup_):
"""Convert ascii+2 encoded codes to string-tokens."""
corrected = tf.bitcast(tf.clip_by_value(tf.subtract(raw, 2), 0, 255),
tf.uint8)
gathered = tf.gather(lookup_, tf.cast(corrected, tf.int32))[:, :, 0]
joined = tf.reduce_join(gathered, axis=1)
cleaned = tf.regex_replace(joined, b"\0", b"")
tokens = tf.string_split(cleaned, " ")
return tokens
def _get_prediction_text(args, window=5):
"""Get the prediction text for a single row in the batch."""
current_context, start, end = args
prediction_context_start = tf.maximum(start - window, 0)
prediction_context_end = tf.minimum(end + 1 + window,
tf.shape(current_context)[0])
before = current_context[prediction_context_start:start]
prediction = current_context[start:end + 1]
after = current_context[end + 1:prediction_context_end]
concat = tf.concat([before, ["**"], prediction, ["**"], after], 0)
return tf.reduce_join(concat, separator=" ")
def random_dna_sequence(template, shape=()):
"""Generate a random DNA sequence according to a template.
Args:
template: string of characters representing a template for making DNA
sequences. Each occurrence of an 'N' is replaced by a randomly chosen
base.
shape: optional tuple indicating the desired shape of the returned Tensor.
Returns:
tf.Tensor with dtype=strings.
"""
n_random = sum(base == 'N' for base in template)
all_bases = tf.constant([x.encode('utf8') for x in dna.DNA_BASES])
random_index = tf.random_uniform((n_random, ) + shape,
maxval=3,
dtype=tf.int32)
random_bases = tf.gather(all_bases, random_index)
iter_bases = iter(random_bases[i] for i in range(n_random))
bases = [
next(iter_bases) if base == 'N' else tf.fill(shape, base)
for base in template
]
return tf.reduce_join(bases, 0)
def get_text_summary(question, context, start_predictions, end_predictions):
"""Get a text summary of the question and the predicted answer."""
question_text = tf.reduce_join(question, -1, separator=" ")
def _get_prediction_text(args, window=5):
"""Get the prediction text for a single row in the batch."""
current_context, start, end = args
prediction_context_start = tf.maximum(start - window, 0)
prediction_context_end = tf.minimum(end + 1 + window,
tf.shape(current_context)[0])
before = current_context[prediction_context_start:start]
prediction = current_context[start:end + 1]
after = current_context[end + 1:prediction_context_end]
concat = tf.concat([before, ["**"], prediction, ["**"], after], 0)
return tf.reduce_join(concat, separator=" ")
prediction_text = tf.map_fn(
fn=_get_prediction_text,
elems=[context, start_predictions, end_predictions],
dtype=tf.string,
back_prop=False)
return tf.summary.text("predictions",
tf.stack([question_text, prediction_text], -1))