def run_keras():
pretreatment = Pretreatment()
train_x, test_x, train_y, test_y = pretreatment.train_test_split(c=3, test_size=0.6)
embedding_matrix = pretreatment.create_embedding_matrix(15000)
textrnn = TextRNN(pretreatment.nnparam)
textrnn.train(train_x, train_y, embedding_matrix)
textrnn.predict(test_x, test_y)
def run_keras():
pretreatment = Pretreatment()
train_x, test_x, train_y, test_y = pretreatment.train_test_split(c=2, test_size=0.1)
# embedding_matrix = pretreatment.create_embedding_matrix(30000)
textrnn = FastText(pretreatment.nnparam)
# textrnn.train(train_x, train_y, embedding_matrix) # 精度 0.9323043484250149 损失 0.270193725742771
textrnn.train(train_x, train_y, '') # 精度 0.9353858005601531 损失 0.2599002837189978
textrnn.predict(test_x, test_y)
def run_keras():
pretreatment = Pretreatment()
train_x, test_x, train_y, test_y = pretreatment.train_test_split(c=5, y_one_hot=False, test_size=0.6)
embedding_matrix = pretreatment.create_embedding_matrix(20000)
textrnn = TextCNN(pretreatment.nnparam)
textrnn.train(train_x, train_y, embedding_matrix)
textrnn.predict(test_x, test_y)
"""
def run_pytorch():
pretreatment = Pretreatment()
train_x, test_x, train_y, test_y = pretreatment.train_test_split(c=3, y_one_hot=False, test_size=0.1)
train_x, train_y, test_x, test_y = data2tensor(train_x, train_y, test_x, test_y)
m = TextRNN2(pretreatment.nnparam)
# m = BiGRUAttention(pretreatment.nnparam)
print(m)
get_parameter_number(m)
for epoch in range(epochs):
train_model(net=m, train_x=train_x, train_y=train_y, epoch=epoch, lr=0.0001)
test_model(m, test_x, test_y)
torch.save(m, os.path.join(corpus_root_path, 'torch_demo', 'textrnn_attention.pkl'))
def run_pytorch():
pretreatment = Pretreatment()
train_x, test_x, train_y, test_y = pretreatment.train_test_split(c=3, y_one_hot=False, test_size=0.1) # 0.9785
train_x, train_y, test_x, test_y = data2tensor(train_x, train_y, test_x, test_y)
fasttext = FastText2(pretreatment.nnparam)
p = 0
for k, v in fasttext.named_parameters():
print('name', k, 'param', v.size())
p += reduce(lambda x, y: x * y, list(v.size()))
print(fasttext)
print('参数量', p)
for epoch in range(epochs):
train_model(net=fasttext, train_x=train_x, train_y=train_y, epoch=epoch, lr=0.0001)
test_model(fasttext, test_x, test_y)
torch.save(fasttext, os.path.join(corpus_root_path, 'torch_demo', 'fasttext.pkl'))
def run_pytorch():
pretreatment = Pretreatment()
train_x, test_x, train_y, test_y = pretreatment.train_test_split(
c=2, test_size=0.1)
train_x, train_y, test_x, test_y = data2tensor(train_x, train_y, test_x,
test_y)
m = BiGRUAttention(pretreatment.nnparam)
print(m)
get_parameter_number(m)
train_model = TrainTorch(model=m)
train_model.train(train_set=(train_x[:1000], train_y[:1000]),
test_set=(test_x, test_y),
batch_size=batch_size,
epochs=epochs,
lr=0.0001,
use_test=True)
torch.save(
m,
os.path.join(corpus_root_path, 'torch_demo',
'text_{}.pkl'.format(m.instance_name)))