"""
from wellcomeml.utils import throw_extra_import_message
from wellcomeml.ml.bert_vectorizer import BertVectorizer
required_modules = 'sklearn'
required_extras = 'core'
try:
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics import f1_score
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
except ImportError as e:
throw_extra_import_message(error=e,
required_modules=required_modules,
required_extras=required_extras)
class SimilarityEntityLinker:
def __init__(self, stopwords, embedding="tf-idf"):
"""
Input:
stopwords - list of stopwords
embedding - How to embed the documents
in order to find which document in the corpus
is most similar to the sentence.
embedding='tf-idf': Use a TFIDF vectoriser
embedding='bert': Use a BERT vectoriser
"""
from wellcomeml.utils import throw_extra_import_message
try:
import tensorflow as tf
except ImportError as e:
throw_extra_import_message(error=e,
required_modules='tensorflow',
extras='tensorflow')
class SelfAttention(tf.keras.layers.Layer):
"""https://papers.nips.cc/paper/7181-attention-is-all-you-need.pdf"""
def __init__(self, attention_dim=20):
super(SelfAttention, self).__init__()
self.attention_dim = attention_dim
def build(self, input_shape):
self.WQ = self.add_weight(
shape=(input_shape[-1], self.attention_dim),
trainable=True,
initializer="uniform",
)
self.WK = self.add_weight(
shape=(input_shape[-1], self.attention_dim),
trainable=True,
initializer="uniform",
)
self.WV = self.add_weight(
shape=(input_shape[-1], input_shape[-1]),
trainable=True,
initializer="uniform",
from wellcomeml.utils import throw_extra_import_message
required_module = 'nervaluate'
required_extras = 'core'
try:
from nervaluate import Evaluator
except ImportError as e:
throw_extra_import_message(e, required_module, required_extras)
def ner_classification_report(y_true, y_pred, groups, tags):
"""
Evaluate the model's performance for each grouping of data
for the NER labels given in 'tags'
Input:
y_pred: a list of predicted entities
y_true: a list of gold entities
groups: (str) the group each of the pred or gold entities belong to
Output:
report: evaluation metrics for each group
in a nice format for printing
"""
unique_groups = sorted(set(groups))
outputs = []
for group in unique_groups:
pred_doc_entities = [y_pred[i] for i, g in enumerate(groups) if g == group]
from wellcomeml.ml.keras_utils import CategoricalMetrics # , MetricMiniBatchHistory
from wellcomeml.logger import LOGGING_LEVEL, build_logger
try:
import tensorflow as tf
from transformers import BertConfig, BertTokenizer, \
TFBertForSequenceClassification
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.model_selection import train_test_split
from sklearn.utils.validation import check_is_fitted
from sklearn.exceptions import NotFittedError
except ImportError as e:
throw_extra_import_message(error=e,
required_modules='tensorflow,sklearn',
extras='core,tensorflow,transformers')
class SemanticEquivalenceClassifier(BaseEstimator, TransformerMixin):
"""
Class to fine-tune BERT-type models for semantic equivalence, for example
paraphrase, textual similarity and other NLU tasks
"""
def __init__(
self,
pretrained="bert",
batch_size=32,
eval_batch_size=32 * 2,
learning_rate=3e-5,
test_size=0.2,
"""
import logging
import re
from wellcomeml.utils import throw_extra_import_message
# Heavy dependencies go here
required_modules = 'spacy,sklearn,scipy'
required_extras = 'spacy,core'
try:
import spacy
from scipy import sparse
from sklearn.feature_extraction.text import TfidfVectorizer
except ImportError as e:
throw_extra_import_message(error=e,
required_modules='spacy',
extras='spacy')
logger = logging.getLogger(__name__)
class WellcomeTfidf(TfidfVectorizer):
"""
Class to wrap some basic transformation and text
vectorisation/embedding
"""
def __init__(self, use_regex=True, use_spacy_lemmatizer=True, **kwargs):
"""
Args:
Any sklearn "tfidfvectorizer" arguments (min_df, etc.)
from datetime import datetime
import math
from wellcomeml.ml.attention import HierarchicalAttention
from wellcomeml.utils import throw_extra_import_message
try:
import tensorflow as tf
from sklearn.model_selection import train_test_split
from sklearn.base import BaseEstimator, ClassifierMixin
from sklearn.metrics import f1_score
from scipy.sparse import csr_matrix, vstack
import numpy as np
except ImportError as e:
throw_extra_import_message(error=e,
required_modules='tensorflow,scipy,numpy',
extras='tensorflow,core')
TENSORBOARD_LOG_DIR = "logs/scalars/" + datetime.now().strftime(
"%Y%m%d-%H%M%S")
CALLBACK_DICT = {
'tensorboard': tf.keras.callbacks.TensorBoard(log_dir=TENSORBOARD_LOG_DIR)
}
class BiLSTMClassifier(BaseEstimator, ClassifierMixin):
def __init__(self,
learning_rate=0.01,
learning_rate_decay=1,
batch_size=32,
nb_epochs=5,