en-ner-conll-2003/sequence-labeling.py
2021-06-08 23:09:05 +02:00

124 lines
4.1 KiB
Python

import torch
import pandas as pd
from torchtext.vocab import Vocab
from collections import Counter
x_train = pd.read_table('train/train.tsv', sep='\t', header = None)
x_dev = pd.read_table('dev-0/in.tsv', sep='\t', header = None)
y_dev = pd.read_table('dev-0/expected.tsv', sep='\t', header = None)
x_test = pd.read_table('test-A/in.tsv', sep='\t', header = None)
label_list = ['O', 'B-LOC', 'I-LOC', 'B-MISC', 'I-MISC', 'B-ORG', 'I-ORG', 'B-PER', 'I-PER']
x_train[0] = x_train[0].apply(lambda x: [label_list.index(i) for i in x.split()])
x_train[1] = x_train[1].apply(lambda x: x.split())
x_dev[0] = x_dev[0].apply(lambda x: x.split())
x_test[0] = x_test[0].apply(lambda x: x.split())
def data_process(dt):
return [ torch.tensor([vocab['<bos>']] +[vocab[token] for token in document ] + [vocab['<eos>']], dtype = torch.long) for document in dt]
def build_vocab(dataset):
counter = Counter()
for document in dataset:
counter.update(document)
return Vocab(counter, specials=['<unk>', '<pad>', '<bos>', '<eos>'])
vocab = build_vocab(x_train[1])
def labels_process(dt):
return [ torch.tensor([0] + document + [0], dtype = torch.long) for document in dt]
class NERModel(torch.nn.Module):
def __init__(self,):
super(NERModel, self).__init__()
self.emb = torch.nn.Embedding(23627,200)
self.fc1 = torch.nn.Linear(600,9)
def forward(self, x):
x = self.emb(x)
x = x.reshape(600)
x = self.fc1(x)
return x
ner_model = NERModel()
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(ner_model.parameters())
train_labels = labels_process(x_train[0])
train_tokens_ids = data_process(x_train[1])
for epoch in range(2):
loss_score = 0
acc_score = 0
prec_score = 0
selected_items = 0
recall_score = 0
relevant_items = 0
items_total = 0
ner_model.train()
for i in range(len(train_labels)):
for j in range(1, len(train_labels[i])-1):
X = train_tokens_ids[i][j-1: j+2]
Y = train_labels[i][j: j+1]
Y_predictions = ner_model(X)
acc_score += int(torch.argmax(Y_predictions) == Y)
if torch.argmax(Y_predictions) != 0:
selected_items +=1
if torch.argmax(Y_predictions) != 0 and torch.argmax(Y_predictions) == Y.item():
prec_score +=1
if Y.item() != 0:
relevant_items += 1
if Y.item() != 0 and torch.argmax(Y_predictions) == Y.item():
recall_score += 1
items_total +=1
optimizer.zero_grad()
loss = criterion(Y_predictions.unsqueeze(0), Y)
loss.backward()
optimizer.step()
loss_score += loss.item()
precision = prec_score / selected_items
recall = recall_score / relevant_items
f1_score = (2*precision * recall) / (precision + recall)
print('epoch: ', epoch)
print('loss: ', loss_score / items_total)
print('acc: ', acc_score / items_total)
print('prec: ', precision)
print('recall: ', recall)
print('f1: ', f1_score)
dev_data_tokens_ids = data_process(x_dev[0])
dev_results = []
for i in range(len(dev_data_tokens_ids)):
line = []
for j in range(1, len(dev_data_tokens_ids[i]) - 1):
X = dev_data_tokens_ids[i][j-1: j+2]
Y_predictions = ner_model(X)
result = torch.argmax(Y_predictions)
label = label_list[result]
line.append(label)
dev_results.append(line)
test_data_tokens_ids = data_process(x_test[0])
test_results = []
for i in range(len(test_data_tokens_ids)):
line = []
for j in range(1, len(test_data_tokens_ids[i]) - 1):
X = test_data_tokens_ids[i][j-1: j+2]
Y_predictions = ner_model(X)
result = torch.argmax(Y_predictions)
label = label_list[result]
line.append(label)
test_results.append(line)
dev_results.to_csv(r'dev-0/out.tsv', sep='\t', index=False, header=False)
test_results.to_csv(r'test-A/out.tsv', sep='\t', index=False, header=False)