def predict(sentence):
if sentence:
model1 = NERModel('bert',
'NERMODEL1',
labels=[
"B-sector", "I-sector", "B-funda", "O",
"operator", "threshold", "Join", "B-attr",
"I-funda", "TPQty", "TPUnit", "Sortby", "B-eco",
"I-eco", "B-index", "Capitalization", "I-",
"funda", "B-security", 'I-security', 'Number',
'Sector', 'TPMonth', 'TPYr', 'TPRef'
],
args={
"save_eval_checkpoints": False,
"save_steps": -1,
"output_dir": "NERMODEL",
'overwrite_output_dir': True,
"save_model_every_epoch": False,
'reprocess_input_data': True,
"train_batch_size": 10,
'num_train_epochs': 15,
"max_seq_length": 64
},
use_cuda=False)
predictions, raw_outputs = model1.predict([sentence])
result = json.dumps(predictions[0])
return result
def test_named_entity_recognition():
# Creating train_df and eval_df for demonstration
train_data = [
[0, "Simple", "B-MISC"],
[0, "Transformers", "I-MISC"],
[0, "started", "O"],
[1, "with", "O"],
[0, "text", "O"],
[0, "classification", "B-MISC"],
[1, "Simple", "B-MISC"],
[1, "Transformers", "I-MISC"],
[1, "can", "O"],
[1, "now", "O"],
[1, "perform", "O"],
[1, "NER", "B-MISC"],
]
train_df = pd.DataFrame(train_data,
columns=["sentence_id", "words", "labels"])
eval_data = [
[0, "Simple", "B-MISC"],
[0, "Transformers", "I-MISC"],
[0, "was", "O"],
[1, "built", "O"],
[1, "for", "O"],
[0, "text", "O"],
[0, "classification", "B-MISC"],
[1, "Simple", "B-MISC"],
[1, "Transformers", "I-MISC"],
[1, "then", "O"],
[1, "expanded", "O"],
[1, "to", "O"],
[1, "perform", "O"],
[1, "NER", "B-MISC"],
]
eval_df = pd.DataFrame(eval_data,
columns=["sentence_id", "words", "labels"])
# Create a NERModel
model = NERModel(
"bert",
"bert-base-cased",
args={
"no_save": True,
"overwrite_output_dir": True,
"reprocess_input_data": False
},
use_cuda=False,
)
# Train the model
model.train_model(train_df)
# Evaluate the model
result, model_outputs, predictions = model.eval_model(eval_df)
# Predictions on arbitary text strings
predictions, raw_outputs = model.predict(["Some arbitary sentence"])
def predict():
if request.method == 'POST':
test_sents = pd.read_csv(request.files.get('file'))
aspect_model = NERModel("bert", "/Users/mutaz/Desktop/Mutaz Thesis bert/hotel_reviews", use_cuda=False,
labels=["B-A", "I-A", "O"],
args={"use_cuda": False,
"save_eval_checkpoints": False,
"save_steps": -1,
"output_dir": "MODEL",
'overwrite_output_dir': True,
"save_model_every_epoch": False,
})
predictions, raw_outputs = aspect_model.predict(test_sents["sentence"])
pred = [[tag__ for word_ in s for word__, tag__ in word_.items()] for s in [sentence_ for sentence_ in predictions]]
print(pred)
# connect(test_sents, aspects_prds)
return redirect('https://dub01.online.tableau.com/#/site/absa/views/absa_project/MAIN?:iid=20&:original_view=y')
args.learning_rate=1e-4
args.overwrite_output_dir=True
args.train_batch_size=32
args.eval_batch_size=32
model=NERModel('bert', 'bert-base-cased', labels=label, args=args)
model.train_model(train_data, eval_data=test_data, acc=accuracy_score)
result, model_outputs, preds_list=model.eval_model(test_data)
result
prediction, model_output=model.predict(["This is Nishi"])
prediction
!pip install bert-extractive-summarizer
!pip install wikipedia
import wikipedia
wiki = wikipedia.page('Amsterdam')
article=wiki.content
print(article)
from summarizer import Summarizer
model = Summarizer()
class NerModel:
def __init__(self, modelname="", dataset=None, use_saved_model=False):
self.dataset = dataset
#labels_list = ["O", "B-ACT", "I-ACT", "B-OBJ", "I-OBJ", "B-VAL", "I-VAL", "B-VAR", "I-VAR"]
#labels_list = dataset.get_labels_list()
labels_list = dataset['labels_list']
output_dir = "outputs_{}".format(modelname)
# Create a NERModel
model_args = {
'output_dir': output_dir,
'overwrite_output_dir': True,
'reprocess_input_data': True,
'save_eval_checkpoints': False,
'save_steps': -1,
'save_model_every_epoch': False,
'train_batch_size': 10, # 10
'num_train_epochs': 10, # 5
'max_seq_length': 256,
'gradient_accumulation_steps': 8,
'labels_list': labels_list
}
if use_saved_model:
self.model = NERModel("bert", output_dir, use_cuda=False, args=model_args)
else:
self.model = NERModel("bert", "bert-base-cased", use_cuda=False, args=model_args)
# args={"overwrite_output_dir": True, "reprocess_input_data": True}
def train(self):
# # Train the model
if self.dataset:
self.model.train_model(self.dataset['train'])
else:
raise Exception("dataset is None")
def eval(self):
# # Evaluate the model
if self.dataset:
result, model_outputs, predictions = self.model.eval_model(self.dataset['val'])
print("Evaluation result:", result)
else:
raise Exception("dataset is None")
def simple_test(self):
# Predictions on arbitary text strings
sentences = ["Some arbitary sentence", "Simple Transformers sentence"]
predictions, raw_outputs = self.model.predict(sentences)
print(predictions)
# More detailed preditctions
for n, (preds, outs) in enumerate(zip(predictions, raw_outputs)):
print("\n___________________________")
print("Sentence: ", sentences[n])
for pred, out in zip(preds, outs):
key = list(pred.keys())[0]
new_out = out[key]
preds = list(softmax(np.mean(new_out, axis=0)))
print(key, pred[key], preds[np.argmax(preds)], preds)
def predict(self, sentences):
predictions, raw_outputs = self.model.predict(sentences)
return predictions
class DependencyModel:
def __init__(self, ner_path: str, deps_path: str):
"""Create/Load a new DependencyModel
Args:
ner_path (str): directory of NER model. (if not exists, create a new model)
deps_path (str): directory of classification model.
"""
self.ner_path = ner_path
try:
self.ner_model = NERModel("distilbert", ner_path, use_cuda=False)
except:
self.ner_model = NERModel("distilbert", "distilbert-base-uncased",
labels=custom_labels, use_cuda=False)
self.deps_model = ClassifyModel(deps_path)
def train_ner(self, ner_data: Iterator[Tuple[str, str, str]] ):
"""Train the NER model
Args:
ner_data (Iterator[Tuple[str, str, str]]): iterator of tuple of (sentence, entity1, entity2).
"""
train_data = []
tokenizer = nltk.RegexpTokenizer(r"[A-Za-z']+")
for i, (sentence, e1, e2) in enumerate(ner_data):
for word in tokenizer.tokenize(sentence):
label = "O"
if word.lower() == e1.lower() or word.lower() == e2.lower():
label = "E"
train_data.append((i, word, label))
train_data = pd.DataFrame(train_data)
train_data.columns = ["sentence_id", "words", "labels"]
self.ner_model.train(train_data, output_dir=self.ner_path)
def train_deps(self, train_data: Iterator[Tuple[str, Tuple[str, str, str]]]):
"""Train the dependency tree path classification model.
Args:
train_data (Iterator[Tuple[str, Tuple[str, str, str]]]): iterator of (sentence, (relation, entity1, entity2))
"""
classify_data: List[str, str] = []
for sentence, (relation, e1, e2) in train_data:
tree = DepTree(sentence)
classify_data.append((" ".join(tree.shortest_path(e1,e2)), relation_list.index(relation)))
self.deps_model.train(train_data)
def train(self, train_data: Iterator[Tuple[str, Tuple[str, str, str]]]):
"""Train both the NER model AND the classification model.
Args:
train_data (Iterator[Tuple[str, Tuple[str, str, str]]]): iterator of (sentence, (relation, entity1, entity2))
"""
train_data = list(train_data)
self.train_ner([ (sent, e1, e2) for sent, (rela, e1, e2) in train_data ])
self.train_deps(train_data)
def predict(self, data: Iterable[str]) -> List[Tuple[str, Tuple[str, str]]]:
"""Predict the relation extracted from input sentences.
Args:
data (Iterable[str]): list of input sentences.
Returns:
List[Tuple[str, Tuple[str, str]]]: list of (relation, (entity1, entity2))
"""
data = list(data)
predictions, raws = self.ner_model.predict(data)
predict_data = []
entities = []
for raw, sentence in zip(raws, data):
[e1, e2] = get_entity(raw)
tree = DepTree(sentence)
entities.append((e1, e2))
predict_data.append(" ".join(tree.shortest_path(e1,e2)))
return list(zip(self.deps_model.predict(predict_data), entities))
if __name__ == '__main__':
train_df = txt_to_df("/Users/mutaz/Desktop/Mutaz Thesis bert/Data/raw/train.txt")
test_df = txt_to_df("/Users/mutaz/Desktop/Mutaz Thesis bert/Data/raw/test.txt")
test_df.reset_index(drop=True, inplace=True)
# test_df["Sentence #"] = test_df["Sentence #"].values
logging.basicConfig(level=logging.INFO)
transformers_logger = logging.getLogger("transformers")
transformers_logger.setLevel(logging.WARNING)
train_df = pd.DataFrame(train_df.values, columns=['sentence_id', 'words', 'labels'])
test_df = pd.DataFrame(test_df.values, columns=['sentence_id', 'words', 'labels'])
model_used = "/Users/mutaz/Desktop/Mutaz Thesis bert/hotel_reviews"
mm, epp, lrr, f11, results = [], [], [], [], {}
aspect_model = NERModel("bert", "{}".format(model_used),
labels=["B-A", "I-A", "O"],
args={"save_eval_checkpoints": False, "save_steps": -1, 'overwrite_output_dir': True,
"save_model_every_epoch": False,
'reprocess_input_data': True, "train_batch_size": 5, 'num_train_epochs': 2,
"gradient_accumulation_steps": 5,
"output_dir": "/Users/mutaz/Desktop/Mutaz Thesis bert"
, "learning_rate": 0.0001}, use_cuda=False)
# aspect_model.train_model(train_df)
test_df = test_df.iloc[:100]
a = test_df.groupby("sentence_id")
gps = [a.get_group(key) for key, item in a]
actual = [list(g.labels.values) for g in gps]
sentences = [" ".join(i) for i in [[w for w in g.words.values] for g in gps]]
predictions, raw_outputs = aspect_model.predict(sentences)
pred = [[tag__ for word_ in s for word__, tag__ in word_.items()] for s in [sentence_ for sentence_ in predictions]]
clear_output()
f1 = f1_score(actual, pred, mode='strict', scheme=IOB2)
print(f1)
line = line.split('\t')
if len(line) == 3:
eval_data.append(line)
eval_df = pd.DataFrame(eval_data,
columns=['sentence_id', 'words', 'labels'])
# test = eval_df.groupby('sentence_id').count()[['words']]
# print(test.query("words > 128"))
# Create a NERModel
model = NERModel('albert',
'albert-base-v2',
use_cuda=False,
args={
'overwrite_output_dir': True,
'reprocess_input_data': True
})
# # Train the model
model.train_model(train_df)
# Evaluate the model
result, model_outputs, predictions = model.eval_model(eval_df)
# Predictions on arbitary text strings
predictions, raw_outputs = model.predict(
["Peter er lige begyndt til badminton."])
print(predictions)
class CustomSlotAnalysis(BaseEstimator, TransformerMixin):
def __init__(self):
device = str(torch.device('cuda' if torch.cuda.is_available() else 'cpu'))
if device=='cpu':
use_cuda = False
else:
use_cuda = True
model_args = {"use_multiprocessing": False}
self.labels = ['B-aircraft_code',
'B-airline_code',
'B-airline_name',
'B-airport_code',
'B-airport_name',
'B-arrive_date.date_relative',
'B-arrive_date.day_name',
'B-arrive_date.day_number',
'B-arrive_date.month_name',
'B-arrive_date.today_relative',
'B-arrive_time.end_time',
'B-arrive_time.period_mod',
'B-arrive_time.period_of_day',
'B-arrive_time.start_time',
'B-arrive_time.time',
'B-arrive_time.time_relative',
'B-booking_class',
'B-city_name',
'B-class_type',
'B-compartment',
'B-connect',
'B-cost_relative',
'B-day_name',
'B-day_number',
'B-days_code',
'B-depart_date.date_relative',
'B-depart_date.day_name',
'B-depart_date.day_number',
'B-depart_date.month_name',
'B-depart_date.today_relative',
'B-depart_date.year',
'B-depart_time.end_time',
'B-depart_time.period_mod',
'B-depart_time.period_of_day',
'B-depart_time.start_time',
'B-depart_time.time',
'B-depart_time.time_relative',
'B-economy',
'B-fare_amount',
'B-fare_basis_code',
'B-flight',
'B-flight_days',
'B-flight_mod',
'B-flight_number',
'B-flight_stop',
'B-flight_time',
'B-fromloc.airport_code',
'B-fromloc.airport_name',
'B-fromloc.city_name',
'B-fromloc.state_code',
'B-fromloc.state_name',
'B-meal',
'B-meal_code',
'B-meal_description',
'B-mod',
'B-month_name',
'B-or',
'B-period_of_day',
'B-restriction_code',
'B-return_date.date_relative',
'B-return_date.day_name',
'B-return_date.day_number',
'B-return_date.month_name',
'B-return_date.today_relative',
'B-return_time.period_mod',
'B-return_time.period_of_day',
'B-round_trip',
'B-state_code',
'B-state_name',
'B-stoploc.airport_code',
'B-stoploc.airport_name',
'B-stoploc.city_name',
'B-stoploc.state_code',
'B-time',
'B-time_relative',
'B-today_relative',
'B-toloc.airport_code',
'B-toloc.airport_name',
'B-toloc.city_name',
'B-toloc.country_name',
'B-toloc.state_code',
'B-toloc.state_name',
'B-transport_type',
'I-airline_name',
'I-airport_name',
'I-arrive_date.date_relative',
'I-arrive_date.day_number',
'I-arrive_time.end_time',
'I-arrive_time.period_of_day',
'I-arrive_time.start_time',
'I-arrive_time.time',
'I-arrive_time.time_relative',
'I-city_name',
'I-class_type',
'I-cost_relative',
'I-depart_date.date_relative',
'I-depart_date.day_number',
'I-depart_date.today_relative',
'I-depart_time.end_time',
'I-depart_time.period_of_day',
'I-depart_time.start_time',
'I-depart_time.time',
'I-depart_time.time_relative',
'I-economy',
'I-fare_amount',
'I-fare_basis_code',
'I-flight_mod',
'I-flight_number',
'I-flight_stop',
'I-flight_time',
'I-fromloc.airport_name',
'I-fromloc.city_name',
'I-fromloc.state_name',
'I-meal_code',
'I-meal_description',
'I-period_of_day',
'I-restriction_code',
'I-return_date.date_relative',
'I-return_date.day_number',
'I-return_date.today_relative',
'I-round_trip',
'I-state_name',
'I-stoploc.city_name',
'I-time',
'I-today_relative',
'I-toloc.airport_name',
'I-toloc.city_name',
'I-toloc.state_name',
'I-transport_type',
'O']
self.model = NERModel('bert', '../models/checkpoint-1120-epoch-2',use_cuda=use_cuda,labels=self.labels,args=model_args)
def fit(self):
print('\n>>>>>>>fit() called. \n')
return self
def convert_to_dict(self,flag,ent,text):
l=[]
for i in range(len(flag)):
l.append({"slot":flag[i],
"value":ent[i],
"extractor": "slot_extractor"})
#x={"entity":l}
#print(l)
text['entities']=l
return text
def transform(self,text):
slot_res=text['correctness']['value']
#print(text['intent']['value'][0])
#print(slot_res)
prediction = self.model.predict([text['text']])
#print('15:slot_pred',prediction)
#label=message.get('correctness')[0].get('value')
prediction = prediction[0][0]
# print(prediction)
#text = ''
ent=[]
flag=[]
# ent=''
# flag=''
for i in prediction:
key = list(i.keys())[0]
if i[key]!='O':
flag.append(i[key])
ent.append(key.split('{')[0])
if slot_res=='correct':
entities =self.convert_to_dict(flag,ent,text)
return entities
else:
entities =self.convert_to_dict('-','-',text)
return entities
transformers_logger.setLevel(logging.WARNING)
# Creating train_df and eval_df for demonstration
train_data = [
[0, 'Simple', 'B-MISC'], [0, 'Transformers', 'I-MISC'], [0, 'started', 'O'], [1, 'with', 'O'], [0, 'text', 'O'], [0, 'classification', 'B-MISC'],
[1, 'Simple', 'B-MISC'], [1, 'Transformers', 'I-MISC'], [1, 'can', 'O'], [1, 'now', 'O'], [1, 'perform', 'O'], [1, 'NER', 'B-MISC']
]
train_df = pd.DataFrame(train_data, columns=['sentence_id', 'words', 'labels'])
eval_data = [
[0, 'Simple', 'B-MISC'], [0, 'Transformers', 'I-MISC'], [0, 'was', 'O'], [1, 'built', 'O'], [1, 'for', 'O'], [0, 'text', 'O'], [0, 'classification', 'B-MISC'],
[1, 'Simple', 'B-MISC'], [1, 'Transformers', 'I-MISC'], [1, 'then', 'O'], [1, 'expanded', 'O'], [1, 'to', 'O'], [1, 'perform', 'O'], [1, 'NER', 'B-MISC']
]
eval_df = pd.DataFrame(eval_data, columns=['sentence_id', 'words', 'labels'])
# print(train_df)
# print(eval_df)
# Create a NERModel
model = NERModel('bert', 'bert-base-cased', args={'overwrite_output_dir': True, 'reprocess_input_data': True})
# Train the model
model.train_model(train_df)
# # Evaluate the model
result, model_outputs, predictions = model.eval_model(eval_df)
print(result, predictions)
# # Predictions on arbitary text strings
predictions, raw_outputs = model.predict(["Simple Transformers started with text classification"])
print(predictions)
# print(raw_outputs)
class RestorePuncts:
def __init__(self, wrds_per_pred=250):
self.wrds_per_pred = wrds_per_pred
self.overlap_wrds = 30
self.valid_labels = [
'OU', 'OO', '.O', '!O', ',O', '.U', '!U', ',U', ':O', ';O', ':U',
"'O", '-O', '?O', '?U'
]
self.model = NERModel("bert",
"felflare/bert-restore-punctuation",
labels=self.valid_labels,
args={
"silent": True,
"max_seq_length": 512
})
# use_cuda isnt working and this hack seems to load the model correctly to the gpu
self.model.device = torch.device("cuda:1")
# dummy punctuate to load the model onto gpu
self.punctuate("hello how are you")
def punctuate(self, text: str, batch_size: int = 32, lang: str = ''):
"""
Performs punctuation restoration on arbitrarily large text.
Detects if input is not English, if non-English was detected terminates predictions.
Overrride by supplying `lang='en'`
Args:
- text (str): Text to punctuate, can be few words to as large as you want.
- lang (str): Explicit language of input text.
"""
#if not lang and len(text) > 10:
# lang = detect(text)
#if lang != 'en':
# raise Exception(F"""Non English text detected. Restore Punctuation works only for English.
# If you are certain the input is English, pass argument lang='en' to this function.
# Punctuate received: {text}""")
def chunks(L, n):
return [L[x:x + n] for x in range(0, len(L), n)]
# plit up large text into bert digestable chunks
splits = self.split_on_toks(text, self.wrds_per_pred,
self.overlap_wrds)
texts = [i["text"] for i in splits]
batches = chunks(texts, batch_size)
preds_lst = []
for batch in batches:
batch_preds, _ = self.model.predict(batch)
preds_lst.extend(batch_preds)
# predict slices
# full_preds_lst contains tuple of labels and logits
#full_preds_lst = [self.predict(i['text']) for i in splits]
# extract predictions, and discard logits
#preds_lst = [i[0][0] for i in full_preds_lst]
# join text slices
combined_preds = self.combine_results(text, preds_lst)
# create punctuated prediction
punct_text = self.punctuate_texts(combined_preds)
return punct_text
def predict(self, input_slice):
"""
Passes the unpunctuated text to the model for punctuation.
"""
predictions, raw_outputs = self.model.predict([input_slice])
return predictions, raw_outputs
@staticmethod
def split_on_toks(text, length, overlap):
"""
Splits text into predefined slices of overlapping text with indexes (offsets)
that tie-back to original text.
This is done to bypass 512 token limit on transformer models by sequentially
feeding chunks of < 512 toks.
Example output:
[{...}, {"text": "...", 'start_idx': 31354, 'end_idx': 32648}, {...}]
"""
wrds = text.replace('\n', ' ').split(" ")
resp = []
lst_chunk_idx = 0
i = 0
while True:
# words in the chunk and the overlapping portion
wrds_len = wrds[(length * i):(length * (i + 1))]
wrds_ovlp = wrds[(length * (i + 1)):((length * (i + 1)) + overlap)]
wrds_split = wrds_len + wrds_ovlp
# Break loop if no more words
if not wrds_split:
break
wrds_str = " ".join(wrds_split)
nxt_chunk_start_idx = len(" ".join(wrds_len))
lst_char_idx = len(" ".join(wrds_split))
resp_obj = {
"text": wrds_str,
"start_idx": lst_chunk_idx,
"end_idx": lst_char_idx + lst_chunk_idx,
}
resp.append(resp_obj)
lst_chunk_idx += nxt_chunk_start_idx + 1
i += 1
logging.info(f"Sliced transcript into {len(resp)} slices.")
return resp
@staticmethod
def combine_results(full_text: str, text_slices):
"""
Given a full text and predictions of each slice combines predictions into a single text again.
Performs validataion wether text was combined correctly
"""
split_full_text = full_text.replace('\n', ' ').split(" ")
split_full_text = [i for i in split_full_text if i]
split_full_text_len = len(split_full_text)
output_text = []
index = 0
if len(text_slices[-1]) <= 3 and len(text_slices) > 1:
text_slices = text_slices[:-1]
for _slice in text_slices:
slice_wrds = len(_slice)
for ix, wrd in enumerate(_slice):
# print(index, "|", str(list(wrd.keys())[0]), "|", split_full_text[index])
if index == split_full_text_len:
break
if split_full_text[index] == str(list(wrd.keys())[0]) and \
ix <= slice_wrds - 3 and text_slices[-1] != _slice:
index += 1
pred_item_tuple = list(wrd.items())[0]
output_text.append(pred_item_tuple)
elif split_full_text[index] == str(list(
wrd.keys())[0]) and text_slices[-1] == _slice:
index += 1
pred_item_tuple = list(wrd.items())[0]
output_text.append(pred_item_tuple)
assert [i[0] for i in output_text] == split_full_text
return output_text
@staticmethod
def punctuate_texts(full_pred: list):
"""
Given a list of Predictions from the model, applies the predictions to text,
thus punctuating it.
"""
punct_resp = ""
for i in full_pred:
word, label = i
if label[-1] == "U":
punct_wrd = word.capitalize()
else:
punct_wrd = word
if label[0] != "O":
punct_wrd += label[0]
punct_resp += punct_wrd + " "
punct_resp = punct_resp.strip()
# Append trailing period if doesnt exist.
if punct_resp[-1].isalnum():
punct_resp += "."
return punct_resp
labels = f.readlines()
labels = [i.strip() for i in labels]
train_args = {
"output_dir": "ner_output",
"overwrite_output_dir": True,
"use_multiprocessing": False,
"save_steps": 0,
"use_early_stopping": True,
"early_stopping_patience": 4,
"evaluate_during_training": True,
"reprocess_input_data": False,
"use_cached_eval_features": True,
"fp16": False,
"num_train_epochs": 10,
"evaluate_during_training_steps": 10000,
"train_batch_size": 32,
'cross_entropy_ignore_index': 0,
'classification_report': True
}
model = NERModel("electra",
"ner_output/checkpoint-150000",
args=train_args,
labels=labels,
use_cuda=True,
crf=True)
result = model.predict([list('发烧头痛3天'), list('见盲肠底，升结肠近肝曲')],
split_on_space=False)
print(result)
class NerModel:
def __init__(self,
modelname="",
dataset=None,
use_saved_model=False,
input_dir=None,
output_dir=None):
#pretrained_model_name = "lm_outputs_test/from_scratch/best_model"
self.tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
self.dataset = dataset
#labels_list = ["O", "B-ACT", "I-ACT", "B-OBJ", "I-OBJ", "B-VAL", "I-VAL", "B-VAR", "I-VAR"]
#labels_list = dataset.get_labels_list()
labels_list = dataset['labels_list']
#labels_list = ['O', 'B-ACT', 'I-ACT', 'B-OBJ', 'I-OBJ', 'B-CNT', 'I-CNT',
# 'B-OPE', 'I-OPE', 'B-ORD', 'B-PRE', 'I-PRE', 'B-TYP',
# 'B-VAL', 'I-VAL', 'B-ATT', 'I-ATT', 'B-VAR', 'I-VAR']
#output_dir = "outputs_{}".format(modelname)
os.system("{} -rf".format(output_dir))
use_cuda = torch.cuda.is_available()
# Create a NERModel
model_args = {
'labels_list': labels_list,
'output_dir': output_dir,
'overwrite_output_dir': True,
'reprocess_input_data': True,
'save_eval_checkpoints': False,
'save_steps': -1,
'save_model_every_epoch': False,
#'no_save' : True,
#'no_cache': True,
'evaluate_during_training': True,
'num_train_epochs': 15, # 5
'train_batch_size': 10, # 10 (<=10 for bert, <=5 for longformer)
'eval_batch_size': 10,
'max_seq_length': 128, # default 128
'gradient_accumulation_steps': 8,
'learning_rate':
0.0001, # default 4e-5; a good value is 0.0001 for albert
#'max_position_embeddings': 64,
}
#self.model = NERModel("bert", pretrained_model_name, use_cuda=False, args=model_args)
#self.model = NERModel("bert", "bert-base-uncased", use_cuda=False, args=model_args)
#self.model = NERModel("longformer", "allenai/longformer-base-4096", use_cuda=False, args=model_args)
#self.model = NERModel("longformer", pretrained_model_name, use_cuda=False, args=model_args)
#self.model = NERModel("xlmroberta", "xlm-roberta-base", use_cuda=False, args=model_args)
#self.model = NERModel("albert", "albert-base-v2", use_cuda=False, args=model_args)
#self.model = NERModel("electra", 'google/electra-small-generator', use_cuda=False, args=model_args)
#self.model = NERModel("layoutlm", 'microsoft/layoutlm-base-uncased', use_cuda=False, args=model_args)
#self.model = NERModel("distilbert", "distilbert-base-cased-distilled-squad", use_cuda=False, args=model_args)
#model_type, english_model_name = "longformer", "allenai/longformer-base-4096"
#model_type, english_model_name = "mpnet", "microsoft/mpnet-base"
#model_type, english_model_name = "electra", "google/electra-small-discriminator"
#model_type, english_model_name = "squeezebert", "squeezebert/squeezebert-uncased"
#model_type, english_model_name = "albert", "albert-base-v2"
#model_type, english_model_name = "xlmroberta", "xlm-roberta-base"
model_type, english_model_name = "roberta", "distilroberta-base"
#model_type, english_model_name = "bert", "bert-base-uncased"
#model_type, english_model_name = "distilbert", "distilbert-base-uncased"
if input_dir:
# Use a previously trained model (on NER or LM tasks)
self.model = NERModel(model_type,
input_dir,
use_cuda=use_cuda,
args=model_args)
else:
# Use a pre-trained (English) model
self.model = NERModel(model_type,
english_model_name,
use_cuda=use_cuda,
args=model_args) # force_download=True
"""
if use_saved_model:
if path:
# Use a model located in a given folder
self.model = NERModel("longformer", path, use_cuda=False, args=model_args)
else:
# Use a previously trained model (on NER or LM tasks)
self.model = NERModel("longformer", output_dir, use_cuda=False, args=model_args)
else:
# Use a pre-trained (English) model
self.model = NERModel("longformer", "allenai/longformer-base-4096", use_cuda=False, args=model_args)
"""
"""
if use_saved_model:
self.model = NERModel("bert", output_dir, use_cuda=False, args=model_args)
else:
self.model = NERModel("bert", pretrained_model_name, use_cuda=False, args=model_args)
# args={"overwrite_output_dir": True, "reprocess_input_data": True}
"""
self.model_info = {
'model_type': model_type,
'english_model_name': english_model_name,
'input_dir': input_dir
}
def train(self):
# # Train the model
if self.dataset:
global_step, training_details = self.model.train_model(
self.dataset['train'], eval_data=self.dataset['val'])
else:
raise Exception("dataset is None")
print("global_step:", global_step)
print("training_details:", training_details)
#training_details: {'global_step': [4], 'precision': [0.6987951807228916], 'recall': [0.402777777777777
#8], 'f1_score': [0.5110132158590308], 'train_loss': [0.41127926111221313], 'eval_loss': [0.63655577600
#00229]}
# it contains f1_score only for the validation dataset
return training_details
def eval(self):
# # Evaluate the model
if self.dataset:
res_train, model_outputs, predictions = self.model.eval_model(
self.dataset['train'])
res_val, model_outputs, predictions = self.model.eval_model(
self.dataset['val'])
print("Evaluation")
#print("On train data:", result)
#{'eval_loss': 0.8920, 'precision': 0.0833, 'recall': 0.027, 'f1_score': 0.0416}
print("train loss: {:.3f}; prec/recall/f1: {:.3f}/{:.3f}/{:.3f}".
format(res_train['eval_loss'], res_train['precision'],
res_train['recall'], res_train['f1_score']))
#print("On validation data:", result)
print("valid loss: {:.3f}; prec/recall/f1: {:.3f}/{:.3f}/{:.3f}".
format(res_val['eval_loss'], res_val['precision'],
res_val['recall'], res_val['f1_score']))
print(
"Summary. Loss (train/val): {:.3f}/{:.3f}, F1: {:.3f}/{:.3f}".
format(res_train['eval_loss'], res_val['eval_loss'],
res_train['f1_score'], res_val['f1_score']))
else:
raise Exception("dataset is None")
print("model_info:", self.model_info)
return res_val
def test(self):
sentence_id = self.dataset['test']['sentence_id']
words = self.dataset['test']['words']
labels = self.dataset['test']['labels']
prev_id = 0
s_words = []
s_labels = []
samples = []
for i in range(len(sentence_id)):
s_id = sentence_id[i]
word = words[i]
label = labels[i]
if s_id != prev_id:
sentence = " ".join(s_words)
#print("sentence id={}: {}".format(prev_id, sentence))
samples.append({
'text': sentence,
'tokens': s_words,
'labels': s_labels
})
#print("s_labels: {}".format(s_labels))
s_words = []
s_labels = []
prev_id = s_id
s_words.append(words[i])
s_labels.append(labels[i])
#print("i={}, word={}, label={}".format(s_id, word, label))
sentence = " ".join(s_words)
#print("sentence id={}: {}".format(prev_id, sentence))
samples.append({
'text': sentence,
'tokens': s_words,
'labels': s_labels
})
texts = [sample['text'] for sample in samples]
predictions, raw_outputs = self.model.predict(texts)
#print(predictions)
acc_list = []
success_list = []
# More detailed preditctions
for i, (preds, raw_outs) in enumerate(zip(predictions, raw_outputs)):
print()
print("text: ", texts[i])
#print("\npreds: ", preds)
pred_labels = [list(t.values())[0] for t in preds]
print("pred_labels: ", pred_labels)
true_labels = samples[i]['labels']
print("true_labels: ", true_labels)
#print("raw_outs: ", raw_outs)
if len(true_labels) != len(pred_labels):
raise Exception("len(true_labels) != len(pred_labels)")
comp = [
true_labels[i] == pred_labels[i]
for i in range(len(pred_labels))
]
acc1sentence = np.mean(comp)
print("acc={:.3f}".format(acc1sentence))
acc_list.append(acc1sentence)
success = 1 if acc1sentence == 1.0 else 0
success_list.append(success)
avg_acc = np.mean(acc_list)
avg_success = np.mean(success_list)
return {'avg_acc': avg_acc, 'avg_success': avg_success}
#for pred, out in zip(preds, outs):
#print("pred:", pred)
#print("out:", out)
#key = list(pred.keys())[0]
#new_out = out[key]
#preds = list(softmax(np.mean(new_out, axis=0)))
#print(key, pred[key], preds[np.argmax(preds)], preds)
def simple_test(self):
# Predictions on arbitary text strings
sentences = ["Some arbitary sentence", "Simple Transformers sentence"]
predictions, raw_outputs = self.model.predict(sentences)
print(predictions)
# More detailed preditctions
for n, (preds, outs) in enumerate(zip(predictions, raw_outputs)):
print("\n___________________________")
print("Sentence: ", sentences[n])
for pred, out in zip(preds, outs):
key = list(pred.keys())[0]
new_out = out[key]
preds = list(softmax(np.mean(new_out, axis=0)))
print(key, pred[key], preds[np.argmax(preds)], preds)
def predict(self, sentences):
predictions, raw_outputs = self.model.predict(sentences)
#tokenized_sentences = [self.tokenizer.tokenize(sentence) for sentence in sentences]
#predictions, raw_outputs = self.model.predict(tokenized_sentences, split_on_space=False)
return predictions
def raw_predict(self, sentences):
predictions, raw_outputs = self.model.predict(sentences)
#print("raw_outputs:", raw_outputs)
#print(self.model.args.labels_list)
labels_list = self.model.args.labels_list
confidences = [
calc_confidence(raw_output, labels_list)
for raw_output in raw_outputs
]
#print("confidence:", confidence)
return {
'predictions': predictions,
'raw_outputs': raw_outputs,
'confidences': confidences
}
"""
model_args = NERArgs(
labels_list = ['B','I', 'O'],
manual_seed = 2021,
num_train_epochs = 4,
max_seq_length = 512,
use_early_stopping = True,
overwrite_output_dir = True,
train_batch_size = 8
)
model = NERModel(
"roberta", "roberta-base", args=model_args, weight = [2.52853895, 12.18978944, 0.39643544], use_cuda=True, cuda_device=-1
)
model.train_model(train_df)
result, model_outputs, wrong_predictions = model.eval_model(
eval_df
)
terms = []
preds = []
for lines_ in lines:
sentences = [[token.text for token in nlp1(line.strip())] for line in lines_]
predictions, raw_outputs = model.predict(sentences, split_on_space=False)
preds.extend(predictions)
with open('ann_weighted_roberta.pkl', 'wb') as f:
pkl.dump(preds, f)
[1, "perform", "O"],
[1, "NER", "B-MISC"],
]
eval_df = pd.DataFrame(eval_data, columns=["sentence_id", "words", "labels"])
# Create a NERModel
model = NERModel("bert", "bert-base-cased", args={"overwrite_output_dir": True, "reprocess_input_data": True})
# # Train the model
# model.train_model(train_df)
# # Evaluate the model
# result, model_outputs, predictions = model.eval_model(eval_df)
# Predictions on arbitary text strings
sentences = ["Some arbitary sentence", "Simple Transformers sentence"]
predictions, raw_outputs = model.predict(sentences)
print(predictions)
# More detailed preditctions
for n, (preds, outs) in enumerate(zip(predictions, raw_outputs)):
print("\n___________________________")
print("Sentence: ", sentences[n])
for pred, out in zip(preds, outs):
key = list(pred.keys())[0]
new_out = out[key]
preds = list(softmax(np.mean(new_out, axis=0)))
print(key, pred[key], preds[np.argmax(preds)], preds)
[0, "text", "O"],
[0, "classification", "B-MISC"],
[1, "Simple", "B-MISC"],
[1, "Transformers", "I-MISC"],
[1, "then", "O"],
[1, "expanded", "O"],
[1, "to", "O"],
[1, "perform", "O"],
[1, "NER", "B-MISC"],
]
eval_df = pd.DataFrame(eval_data, columns=["sentence_id", "words", "labels"])
# Create a NERModel
model = NERModel("bert",
"bert-base-cased",
args={
"overwrite_output_dir": True,
"reprocess_input_data": True
})
# Train the model
model.train_model(train_df)
# Evaluate the model
result, model_outputs, predictions = model.eval_model(eval_df)
# Predictions on arbitary text strings
predictions, raw_outputs = model.predict(["Some arbitary sentence"])
print(predictions)
def main(sentence):
"""Predicts NER labels"""
model = NERModel('bert', 'outputs/', use_cuda=False) #
predictions, raw_outputs = model.predict([sentence])
print(predictions)
