def train(training_data, word_to_ix, pretrained_weight):
tag_to_ix = {"K": 0, "o": 1}
model = LSTMW2VTagger(EMBEDDING_DIM, HIDDEN_DIM, len(word_to_ix),
len(tag_to_ix), pretrained_weight)
loss_function = nn.NLLLoss()
optimizer = optim.SGD(model.parameters(), lr=0.1)
for epoch in range(EPOCHS): # 我们要训练300次，可以根据任务量的大小酌情修改次数。
i = 0
for sentence, tags in training_data:
i += 1
model.zero_grad()
model.hidden = model.init_hidden()
sentence_in = lstm.prepare_sequence(sentence, word_to_ix)
targets = lstm.prepare_sequence(tags, tag_to_ix)
tag_scores = model(sentence_in)
loss = loss_function(tag_scores, targets)
loss.backward()
optimizer.step()
if (i % 100 == 0):
now = datetime.strftime(datetime.now(), "%m-%d %H:%M:%S")
print(now, "epoch:", epoch, "num:", i)
path_name = "./model/lstm_w2v" + str(epoch) + ".pkl"
print(path_name)
torch.save(model, path_name)
print("model has been saved")
def report_accuracy(self,
data,
word_to_ix,
tag_to_ix,
ix_to_tag,
print_data=False):
""" Reports accuracy with respect to exact match (all tags correct per sentence)
and total matches (all correctly classified tags).
"""
# Here we don't need to train, so the code is wrapped in torch.no_grad()
with torch.no_grad():
total = 0
total_correct = 0
total_exact_correct = 0
for sentence, tags in data:
scores = self(prepare_sequence(sentence, word_to_ix))
out = torch.argmax(scores, dim=1)
out_tags = [ix_to_tag[ix] for ix in out]
targets = prepare_sequence(tags, tag_to_ix)
correct = 0
length = len(tags)
for i in range(length):
if out[i] == targets[i]:
correct += 1
total += length
total_correct += correct
n = len(data)
print('Accuracy: %d / %d, %0.4f' %
(total_correct, total, total_correct / total))
def main(train_file):
# Load the data.
training_data = read_data(train_file)
n = len(training_data)
########
#print(n)
# Store word -> word_index mapping.
word_to_ix = {}
for sent, tags in training_data:
for word in sent:
if word not in word_to_ix:
word_to_ix[word] = len(word_to_ix)
##print(word_to_ix)
# Store tag -> tag_index mapping.
tag_to_ix = {tag: ix for ix, tag in enumerate(TAGS)}
##################
#print(tag_to_ix)
##################
# Initialize the model.
model = LSTMTagger(EMBEDDING_DIM, HIDDEN_DIM, len(word_to_ix),
len(tag_to_ix))
loss_function = nn.NLLLoss()
optimizer = optim.SGD(model.parameters(), lr=0.1)
for epoch in range(EPOCHS):
for i, (sentence, tags) in enumerate(training_data):
# Step 1. Pytorch accumulates gradients.
# We need to clear them out before each instance
model.zero_grad()
# Also, we need to clear out the hidden state of the LSTM,
# detaching it from its history on the last instance.
model.hidden = model.init_hidden()
# Step 2. Get our inputs ready for the network, that is, turn them into
# Tensors of word indices.
sentence_in = prepare_sequence(sentence, word_to_ix)
targets = prepare_sequence(tags, tag_to_ix)
# Step 3. Run our forward pass.
tag_scores = model(sentence_in)
# Step 4. Compute the loss, gradients, and update the parameters by
# calling optimizer.step()
loss = loss_function(tag_scores, targets)
loss.backward()
optimizer.step()
if i % 90 == 0:
print('Epoch %d, sentence %d/%d, loss: %0.4f' %
(epoch + 1, i + 1, n, loss))
# Report training accuracy
model.report_accuracy(training_data, word_to_ix, tag_to_ix, TAGS)
def test_weight(model, data, word_to_ix):
result = []
tmp = []
for i, _ in enumerate(data):
inputs = lstm.prepare_sequence(data[i][0], word_to_ix)
tag_scores = model(inputs)
tmp.append(tag_scores)
result.append(data[i][0])
return tmp
def train(model, training_data, word_to_ix, pretrained_weight, start_epoch=0):
print("load data finished,begin to train")
tag_to_ix = {"K": 0, "o": 1}
loss_function = nn.NLLLoss()
optimizer = optim.SGD(model.parameters(), lr=0.1)
if start_epoch == 0:
write_path = "w"
else:
write_path = "a"
f = open("./output/bilstmw2vresult.txt", write_path, encoding="utf-8")
for epoch in range(start_epoch, EPOCHS):
i = 0
for sentence, tags in training_data:
i += 1
model.zero_grad()
model.hidden = model.init_hidden()
sentence_in = lstm.prepare_sequence(sentence, word_to_ix)
targets = lstm.prepare_sequence(tags, tag_to_ix)
tag_scores = model(sentence_in)
loss = loss_function(tag_scores, targets)
loss.backward()
optimizer.step()
if (i % 100 == 0):
now = datetime.strftime(datetime.now(), "%m-%d %H:%M:%S")
print(now, "epoch:", epoch, "num:", i)
path_name = "./model/bilstm_w2v" + str(epoch) + ".pkl"
torch.save(model, path_name)
yp_train = test(model, X_train, word_to_ix)
yp_test = test(model, X_test, word_to_ix)
p_t, r_t, f1_t = lstm.calculate(y_train, yp_train)
p, r, f1 = lstm.calculate(y_test, yp_test)
train_result = [p_t, r_t, f1_t]
test_result = [p, r, f1]
f.write(str(epoch) + "\t")
f.write(" ".join(map(str, train_result)))
f.write("\t")
f.write(" ".join(map(str, test_result)))
f.write("\n")
f.flush()
print("model has been saved")
sys.stdout.flush()
f.close()
def test(model, data, word_to_ix):
result = []
tmp = []
for i, _ in enumerate(data):
inputs = lstm.prepare_sequence(data[i][0], word_to_ix)
tag_scores = model(inputs)
tmp.append(tag_scores)
for item in tmp:
tmpp = []
for i, j in item:
if (i > j):
tmpp.append("K")
else:
tmpp.append("o")
result.append(tmpp)
return result
