-
Notifications
You must be signed in to change notification settings - Fork 3
/
ner.py
200 lines (157 loc) · 6.58 KB
/
ner.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
import transformers as ts
from transformers import DataCollatorForTokenClassification
from datasets import Dataset
from datasets import load_metric
import numpy as np
import pandas as pd
import csv
datasetName = "NCBI-disease"
modelPath = "nlpie/distil-biobert"
tokenizerPath = "nlpie/distil-biobert"
datasetPath = f"biobert-datasets/datasets/NER/{datasetName}/"
logsPath = f"ner_logs/{modelPath}-{datasetName}-logs.txt"
def load_and_preprocess_dataset(datasetPath, tokenizer):
def load_ner_dataset(folder):
allLabels = set(pd.read_csv(folder + "train.tsv", sep="\t",
header=None, quoting=csv.QUOTE_NONE, encoding='utf-8')[1])
label_to_index = {label: index for index,
label in enumerate(allLabels)}
index_to_label = {index: label for index,
label in enumerate(allLabels)}
def load_subset(subset):
lines = []
with open(folder + subset, mode="r") as f:
lines = f.readlines()
sentences = []
labels = []
currentSampleTokens = []
currentSampleLabels = []
for line in lines:
if line.strip() == "":
sentences.append(currentSampleTokens)
labels.append(currentSampleLabels)
currentSampleTokens = []
currentSampleLabels = []
else:
cleanedLine = line.replace("\n", "")
token, label = cleanedLine.split(
"\t")[0].strip(), cleanedLine.split("\t")[1].strip()
currentSampleTokens.append(token)
currentSampleLabels.append(label_to_index[label])
dataDict = {
"tokens": sentences,
"ner_tags": labels,
}
return Dataset.from_dict(dataDict)
trainingDataset = load_subset("train.tsv")
validationDataset = Dataset.from_dict(
load_subset("train_dev.tsv")[len(trainingDataset):])
testDataset = load_subset("test.tsv")
return {
"train": trainingDataset,
"validation": validationDataset,
"test": testDataset,
"all_ner_tags": list(allLabels),
}
dataset = load_ner_dataset(datasetPath)
label_names = dataset["all_ner_tags"]
# Get the values for input_ids, token_type_ids, attention_mask
def tokenize_adjust_labels(all_samples_per_split):
tokenized_samples = tokenizer.batch_encode_plus(
all_samples_per_split["tokens"], is_split_into_words=True, max_length=512)
total_adjusted_labels = []
for k in range(0, len(tokenized_samples["input_ids"])):
prev_wid = -1
word_ids_list = tokenized_samples.word_ids(batch_index=k)
existing_label_ids = all_samples_per_split["ner_tags"][k]
i = -1
adjusted_label_ids = []
for wid in word_ids_list:
if(wid is None):
adjusted_label_ids.append(-100)
elif(wid != prev_wid):
i = i + 1
adjusted_label_ids.append(existing_label_ids[i])
prev_wid = wid
else:
adjusted_label_ids.append(existing_label_ids[i])
total_adjusted_labels.append(adjusted_label_ids)
tokenized_samples["labels"] = total_adjusted_labels
return tokenized_samples
tokenizedTrainDataset = dataset["train"].map(
tokenize_adjust_labels, batched=True)
tokenizedValDataset = dataset["validation"].map(
tokenize_adjust_labels, batched=True)
tokenizedTestDataset = dataset["test"].map(
tokenize_adjust_labels, batched=True)
metric = load_metric("seqeval")
def compute_metrics(p):
predictions, labels = p
predictions = np.argmax(predictions, axis=2)
# Remove ignored index (special tokens)
true_predictions = [
[label_names[p] for (p, l) in zip(prediction, label) if l != -100]
for prediction, label in zip(predictions, labels)
]
true_labels = [
[label_names[l] for (p, l) in zip(prediction, label) if l != -100]
for prediction, label in zip(predictions, labels)
]
results = metric.compute(
predictions=true_predictions, references=true_labels)
flattened_results = {
"overall_precision": results["overall_precision"],
"overall_recall": results["overall_recall"],
"overall_f1": results["overall_f1"],
"overall_accuracy": results["overall_accuracy"],
}
return flattened_results
return tokenizedTrainDataset, tokenizedValDataset, tokenizedTestDataset, compute_metrics, label_names
def train_and_evaluate(lr,
batchsize,
epochs,
tokenizer,
tokenizedTrainDataset,
tokenizedValDataset,
tokenizedTestDataset,
compute_metrics,
label_names,
logsPath=None,
trainingArgs=None):
model = ts.AutoModelForTokenClassification.from_pretrained(
modelPath, num_labels=len(label_names))
data_collator = DataCollatorForTokenClassification(tokenizer)
model.train()
if trainingArgs == None:
trainingArguments = ts.TrainingArguments(
"output/",
seed=42,
logging_steps=250,
save_steps=2500,
num_train_epochs=epochs,
learning_rate=lr,
lr_scheduler_type="cosine",
per_device_train_batch_size=batchsize,
per_device_eval_batch_size=batchsize,
weight_decay=0.01,
)
else:
trainingArguments = trainingArgs
trainer = ts.Trainer(
model=model,
args=trainingArguments,
train_dataset=tokenizedTrainDataset,
eval_dataset=tokenizedValDataset,
data_collator=data_collator,
compute_metrics=compute_metrics,
)
trainer.train()
model.eval()
evaluationResult = trainer.evaluate()
trainer.eval_dataset = tokenizedTestDataset
testResult = trainer.evaluate()
if logsPath != None:
with open(logsPath, mode="a+") as f:
f.write(
f"---HyperParams---\nBatchsize= {batchsize} Lr= {lr}\n---Val Results---\n{str(evaluationResult)}\n---Test Results---\n{str(testResult)}\n\n")
return model, evaluationResult, testResult