def _base(self, strings_left, strings_right):
input_ids_left, input_masks_left, segment_ids_left, _ = xlnet_tokenization(
self._tokenizer, strings_left)
input_ids_right, input_masks_right, segment_ids_right, _ = xlnet_tokenization(
self._tokenizer, strings_left)
r = self._execute(
inputs=[
input_ids_left,
segment_ids_left,
input_masks_left,
input_ids_right,
input_masks_right,
segment_ids_right,
],
input_labels=[
'Placeholder',
'Placeholder_1',
'Placeholder_2',
'Placeholder_3',
'Placeholder_4',
'Placeholder_5',
],
output_labels=['logits'],
)
return softmax(r['logits'], axis=-1)
def _vectorize(self, strings, method='first'):
method = method.lower()
if method not in ['first', 'last', 'mean', 'word']:
raise ValueError(
"method not supported, only support 'first', 'last', 'mean' and 'word'"
)
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, strings)
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['vectorizer'],
)
v = r['vectorizer']
if method == 'first':
v = v[:, 0]
elif method == 'last':
v = v[:, -1]
elif method == 'mean':
v = np.mean(v, axis=1)
else:
v = [
merge_sentencepiece_tokens(
list(zip(s_tokens[i], v[i][:len(s_tokens[i])])),
weighted=False,
vectorize=True,
model='xlnet',
) for i in range(len(v))
]
return v
def predict(self, string: str):
"""
Tag a string.
Parameters
----------
string : str
Returns
-------
result : Tuple[str, str]
"""
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string])
s_tokens = s_tokens[0]
predicted = self._sess.run(
self._logits,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)[0]
t = [self._settings['idx2tag'][d] for d in predicted]
merged = merge_sentencepiece_tokens_tagging(s_tokens, t, model='xlnet')
return list(zip(*merged))
def vectorize(self, strings):
"""
Vectorize string inputs using bert attention.
Parameters
----------
strings : str / list of str
Returns
-------
array: vectorized strings
"""
if isinstance(strings, list):
if not isinstance(strings[0], str):
raise ValueError('input must be a list of strings or a string')
else:
if not isinstance(strings, str):
raise ValueError('input must be a list of strings or a string')
if isinstance(strings, str):
strings = [strings]
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
return self._sess.run(
self.logits,
feed_dict={
self.X: input_ids,
self.segment_ids: segment_ids,
self.input_masks: input_masks,
},
)
def _vectorize(self, strings, method='first'):
method = method.lower()
if method not in ['first', 'last', 'mean', 'word']:
raise ValueError(
"method not supported, only support 'first', 'last', 'mean' and 'word'"
)
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, strings)
v = self._sess.run(
self._vectorizer,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
if method == 'first':
v = v[:, 0]
elif method == 'last':
v = v[:, -1]
elif method == 'mean':
v = np.mean(v, axis=1)
else:
v = [
merge_sentencepiece_tokens(
list(zip(s_tokens[i], v[i][:len(s_tokens[i])])),
weighted=False,
vectorize=True,
model='xlnet',
) for i in range(len(v))
]
return v
def vectorize(self, string: str):
"""
vectorize a string.
Parameters
----------
string: List[str]
Returns
-------
result: np.array
"""
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string])
s_tokens = s_tokens[0]
v = self._sess.run(
self._vectorizer,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
v = v[0]
return merge_sentencepiece_tokens(
list(zip(s_tokens, v[:len(s_tokens)])),
weighted=False,
vectorize=True,
model='xlnet',
)
def vectorize(self, string: str):
"""
vectorize a string.
Parameters
----------
string: List[str]
Returns
-------
result: np.array
"""
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string], space_after_punct=True)
s_tokens = s_tokens[0]
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['vectorizer'],
)
v = r['vectorizer']
v = v[0]
return merge_sentencepiece_tokens(
list(zip(s_tokens, v[:len(s_tokens)])),
weighted=False,
vectorize=True,
model='xlnet',
)
def vectorize(self, strings: List[str]):
"""
Vectorize string inputs.
Parameters
----------
strings : List[str]
Returns
-------
result: np.array
"""
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings
)
return self._sess.run(
self.logits,
feed_dict = {
self.X: input_ids,
self.segment_ids: segment_ids,
self.input_masks: input_masks,
},
)
def _tokenize(self, string):
if self._tok:
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization_token(
self._tokenizer, self._tok, [string])
else:
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string], space_after_punct=True)
s_tokens = s_tokens[0]
return input_ids, input_masks, segment_ids, s_tokens
def _classify(self, strings):
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['logits'],
)
return softmax(r['logits'], axis=-1)
def predict(self, string: str):
"""
Tag a string.
Parameters
----------
string : str
Returns
-------
result : Tuple
"""
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string], space_after_punct=True)
s_tokens = s_tokens[0]
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['logits', 'heads_seq'],
)
tagging, depend = r['logits'], r['heads_seq']
tagging = [self._idx2tag[i] for i in tagging[0]]
depend = depend[0] - self._minus
for i in range(len(depend)):
if depend[i] == 0 and tagging[i] != 'root':
tagging[i] = 'root'
elif depend[i] != 0 and tagging[i] == 'root':
depend[i] = 0
tagging = merge_sentencepiece_tokens_tagging(s_tokens,
tagging,
model='xlnet')
tagging = list(zip(*tagging))
indexing = merge_sentencepiece_tokens_tagging(s_tokens,
depend,
model='xlnet')
indexing = list(zip(*indexing))
result, indexing_ = [], []
for i in range(len(tagging)):
index = int(indexing[i][1])
if index > len(tagging):
index = len(tagging)
elif (i + 1) == index:
index = index + 1
elif index == -1:
index = i
indexing_.append((indexing[i][0], index))
result.append('%d\t%s\t_\t_\t_\t_\t%d\t%s\t_\t_' %
(i + 1, tagging[i][0], index, tagging[i][1]))
d = DependencyGraph('\n'.join(result), top_relation_label='root')
return d, tagging, indexing_
def _classify(self, strings):
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
return self._sess.run(
self._softmax,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
def predict(self, string: str):
"""
Tag a string.
Parameters
----------
string : str
Returns
-------
result : Tuple
"""
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string])
s_tokens = s_tokens[0]
tagging, depend = self._sess.run(
[self._logits, self._heads_seq],
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
tagging = [self._idx2tag[i] for i in tagging[0]]
depend = depend[0] - 1
for i in range(len(depend)):
if depend[i] == 0 and tagging[i] != 'root':
tagging[i] = 'root'
elif depend[i] != 0 and tagging[i] == 'root':
depend[i] = 0
tagging = merge_sentencepiece_tokens_tagging(s_tokens,
tagging,
model='xlnet')
tagging = list(zip(*tagging))
indexing = merge_sentencepiece_tokens_tagging(s_tokens,
depend,
model='xlnet')
indexing = list(zip(*indexing))
result, indexing_ = [], []
for i in range(len(tagging)):
index = int(indexing[i][1])
if index > len(tagging):
index = len(tagging)
indexing_.append((indexing[i][0], index))
result.append('%d\t%s\t_\t_\t_\t_\t%d\t%s\t_\t_' %
(i + 1, tagging[i][0], index, tagging[i][1]))
d = DependencyGraph('\n'.join(result), top_relation_label='root')
return d, tagging, indexing_
def vectorize(self, strings: List[str], method: str = 'first'):
"""
vectorize list of strings.
Parameters
----------
strings: List[str]
method : str, optional (default='first')
Vectorization layer supported. Allowed values:
* ``'last'`` - vector from last sequence.
* ``'first'`` - vector from first sequence.
* ``'mean'`` - average vectors from all sequences.
* ``'word'`` - average vectors based on tokens.
Returns
-------
result: np.array
"""
method = method.lower()
if method not in ['first', 'last', 'mean', 'word']:
raise ValueError(
"method not supported, only support 'first', 'last', 'mean' and 'word'"
)
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, strings)
v = self._sess.run(
self._vectorizer,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
if method == 'first':
v = v[:, 0]
elif method == 'last':
v = v[:, -1]
elif method == 'mean':
v = np.mean(v, axis=1)
else:
v = [
merge_sentencepiece_tokens(
list(zip(s_tokens[i], v[i][:len(s_tokens[i])])),
weighted=False,
vectorize=True,
model='xlnet',
) for i in range(len(v))
]
return v
def _predict(self, strings):
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
result = self._sess.run(
self._sigmoid,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
return result
def _attention(self, strings):
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, strings)
maxlen = max([len(s) for s in s_tokens])
s_tokens = padding_sequence(s_tokens, maxlen, pad_int='<cls>')
attentions = self._sess.run(
self.attention_nodes,
feed_dict={
self.X: input_ids,
self.segment_ids: segment_ids,
self.input_masks: input_masks,
},
)
return attentions, s_tokens, input_masks
def _predict(self, strings, add_neutral):
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
result = self._sess.run(
self._softmax,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
if add_neutral:
result = neutral(result)
return result
def vectorize(self, strings: List[str]):
"""
Vectorize list of strings.
Parameters
----------
strings : List[str]
Returns
-------
result: np.array
"""
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['xlnet/summary'],
)
return r['xlnet/summary']
def vectorize(self, strings: List[str]):
"""
Vectorize list of strings.
Parameters
----------
strings : List[str]
Returns
-------
result: np.array
"""
input_ids, input_masks, segment_ids, _ = xlnet_tokenization(
self._tokenizer, strings)
segment_ids = np.array(segment_ids)
segment_ids[segment_ids == 0] = 1
return self._sess.run(
self._vectorizer,
feed_dict={
self._X: input_ids,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
def predict_words(self,
string: str,
method: str = 'last',
visualization: bool = True):
"""
classify words.
Parameters
----------
string : str
method : str, optional (default='last')
Attention layer supported. Allowed values:
* ``'last'`` - attention from last layer.
* ``'first'`` - attention from first layer.
* ``'mean'`` - average attentions from all layers.
visualization: bool, optional (default=True)
If True, it will open the visualization dashboard.
Returns
-------
dictionary: results
"""
method = method.lower()
if method not in ['last', 'first', 'mean']:
raise Exception(
"method not supported, only support 'last', 'first' and 'mean'"
)
batch_x, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string])
result, attentions, words = self._sess.run(
[self._softmax, self._attns, self._softmax_seq],
feed_dict={
self._X: batch_x,
self._segment_ids: segment_ids,
self._input_masks: input_masks,
},
)
if method == 'first':
cls_attn = attentions[0][:, :, 0, :]
if method == 'last':
cls_attn = attentions[-1][:, :, 0, :]
if method == 'mean':
cls_attn = np.mean(attentions, axis=0).mean(axis=2)
cls_attn = np.mean(cls_attn, axis=1)
total_weights = np.sum(cls_attn, axis=-1, keepdims=True)
attn = cls_attn / total_weights
result = result[0]
words = words[0]
weights = []
merged = merge_sentencepiece_tokens(list(zip(s_tokens[0], attn[0])))
for i in range(words.shape[1]):
m = merge_sentencepiece_tokens(list(zip(s_tokens[0], words[:, i])),
weighted=False)
_, weight = zip(*m)
weights.append(weight)
w, a = zip(*merged)
words = np.array(weights).T
distribution_words = words[:, np.argmax(words.sum(axis=0))]
y_histogram, x_histogram = np.histogram(distribution_words,
bins=np.arange(0, 1, 0.05))
y_histogram = y_histogram / y_histogram.sum()
x_attention = np.arange(len(w))
left, right = np.unique(np.argmax(words, axis=1), return_counts=True)
left = left.tolist()
y_barplot = []
for i in range(len(self._label)):
if i not in left:
y_barplot.append(i)
else:
y_barplot.append(right[left.index(i)])
dict_result = {self._label[i]: result[i] for i in range(len(result))}
dict_result['alphas'] = {w: a[no] for no, w in enumerate(w)}
dict_result['word'] = {w: words[no] for no, w in enumerate(w)}
dict_result['histogram'] = {'x': x_histogram, 'y': y_histogram}
dict_result['attention'] = {'x': x_attention, 'y': np.array(a)}
dict_result['barplot'] = {'x': self._label, 'y': y_barplot}
dict_result['class_name'] = self._class_name
if visualization:
if self._class_name == 'relevancy':
_render_relevancy(dict_result)
else:
_render_emotion(dict_result)
else:
return dict_result
def _predict_words(self, string, method, visualization, add_neutral=False):
method = method.lower()
if method not in ['last', 'first', 'mean']:
raise ValueError(
"method not supported, only support 'last', 'first' and 'mean'"
)
if add_neutral:
label = self._label + ['neutral']
else:
label = self._label
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string])
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['logits', 'attention', 'logits_seq'],
)
result = softmax(r['logits'], axis=-1)
words = softmax(r['logits_seq'], axis=-1)
attentions = r['attention']
if method == 'first':
cls_attn = attentions[0][:, :, 0, :]
if method == 'last':
cls_attn = attentions[-1][:, :, 0, :]
if method == 'mean':
cls_attn = np.mean(attentions, axis=0).mean(axis=2)
cls_attn = np.mean(cls_attn, axis=1)
total_weights = np.sum(cls_attn, axis=-1, keepdims=True)
attn = cls_attn / total_weights
words = words[0]
if add_neutral:
result = neutral(result)
words = neutral(words)
result = result[0]
weights = []
merged = merge_sentencepiece_tokens(list(zip(s_tokens[0], attn[0])),
model='xlnet')
for i in range(words.shape[1]):
m = merge_sentencepiece_tokens(
list(zip(s_tokens[0], words[:, i])),
weighted=False,
model='xlnet',
)
_, weight = zip(*m)
weights.append(weight)
w, a = zip(*merged)
words = np.array(weights).T
distribution_words = words[:, np.argmax(words.sum(axis=0))]
y_histogram, x_histogram = np.histogram(distribution_words,
bins=np.arange(0, 1, 0.05))
y_histogram = y_histogram / y_histogram.sum()
x_attention = np.arange(len(w))
left, right = np.unique(np.argmax(words, axis=1), return_counts=True)
left = left.tolist()
y_barplot = []
for i in range(len(label)):
if i not in left:
y_barplot.append(i)
else:
y_barplot.append(right[left.index(i)])
dict_result = {label[i]: result[i] for i in range(len(result))}
dict_result['alphas'] = {w: a[no] for no, w in enumerate(w)}
dict_result['word'] = {w: words[no] for no, w in enumerate(w)}
dict_result['histogram'] = {'x': x_histogram, 'y': y_histogram}
dict_result['attention'] = {'x': x_attention, 'y': np.array(a)}
dict_result['barplot'] = {'x': label, 'y': y_barplot}
dict_result['module'] = self._module
if visualization:
render_dict[self._module](dict_result)
else:
return dict_result
def predict_words(self,
string: str,
method: str = 'last',
visualization: bool = True):
"""
classify words.
Parameters
----------
string : str
method : str, optional (default='last')
Attention layer supported. Allowed values:
* ``'last'`` - attention from last layer.
* ``'first'`` - attention from first layer.
* ``'mean'`` - average attentions from all layers.
visualization: bool, optional (default=True)
If True, it will open the visualization dashboard.
Returns
-------
dictionary: results
"""
method = method.lower()
if method not in ['last', 'first', 'mean']:
raise ValueError(
"method not supported, only support 'last', 'first' and 'mean'"
)
input_ids, input_masks, segment_ids, s_tokens = xlnet_tokenization(
self._tokenizer, [string])
r = self._execute(
inputs=[input_ids, segment_ids, input_masks],
input_labels=['Placeholder', 'Placeholder_1', 'Placeholder_2'],
output_labels=['logits', 'attention', 'logits_seq'],
)
result = sigmoid(r['logits'])
words = sigmoid(r['logits_seq'])
attentions = r['attention']
if method == 'first':
cls_attn = attentions[0][:, :, 0, :]
if method == 'last':
cls_attn = attentions[-1][:, :, 0, :]
if method == 'mean':
cls_attn = np.mean(attentions, axis=0).mean(axis=2)
cls_attn = np.mean(cls_attn, axis=1)
total_weights = np.sum(cls_attn, axis=-1, keepdims=True)
attn = cls_attn / total_weights
result = result[0]
words = words[0]
weights = []
merged = merge_sentencepiece_tokens(list(zip(s_tokens[0], attn[0])),
model='xlnet')
for i in range(words.shape[1]):
m = merge_sentencepiece_tokens(
list(zip(s_tokens[0], words[:, i])),
weighted=False,
model='xlnet',
)
_, weight = zip(*m)
weights.append(weight)
w, a = zip(*merged)
words = np.array(weights).T
distribution_words = words[:, np.argmax(words.sum(axis=0))]
y_histogram, x_histogram = np.histogram(distribution_words,
bins=np.arange(0, 1, 0.05))
y_histogram = y_histogram / y_histogram.sum()
x_attention = np.arange(len(w))
left, right = np.unique(np.argmax(words, axis=1), return_counts=True)
left = left.tolist()
y_barplot = []
for i in range(len(self._label)):
if i not in left:
y_barplot.append(i)
else:
y_barplot.append(right[left.index(i)])
dict_result = {self._label[i]: result[i] for i in range(len(result))}
dict_result['alphas'] = {w: a[no] for no, w in enumerate(w)}
dict_result['word'] = {w: words[no] for no, w in enumerate(w)}
dict_result['histogram'] = {'x': x_histogram, 'y': y_histogram}
dict_result['attention'] = {'x': x_attention, 'y': np.array(a)}
dict_result['barplot'] = {'x': self._label, 'y': y_barplot}
dict_result['module'] = self._module
if visualization:
_render_toxic(dict_result)
else:
return dict_result
