def train():
# Config Loader
train_args = ConfigSection()
test_args = ConfigSection()
ConfigLoader().load_config(cfgfile, {
"train": train_args,
"test": test_args
})
print("loading data set...")
data = SeqLabelDataSet(load_func=TokenizeDataSetLoader.load)
data.load(cws_data_path)
data_train, data_dev = data.split(ratio=0.3)
train_args["vocab_size"] = len(data.word_vocab)
train_args["num_classes"] = len(data.label_vocab)
print("vocab size={}, num_classes={}".format(len(data.word_vocab),
len(data.label_vocab)))
change_field_is_target(data_dev, "truth", True)
save_pickle(data_dev, "./save/", "data_dev.pkl")
save_pickle(data.word_vocab, "./save/", "word2id.pkl")
save_pickle(data.label_vocab, "./save/", "label2id.pkl")
# Trainer
trainer = SeqLabelTrainer(epochs=train_args["epochs"],
batch_size=train_args["batch_size"],
validate=train_args["validate"],
use_cuda=train_args["use_cuda"],
pickle_path=train_args["pickle_path"],
save_best_dev=True,
print_every_step=10,
model_name="trained_model.pkl",
evaluator=SeqLabelEvaluator())
# Model
model = AdvSeqLabel(train_args)
try:
ModelLoader.load_pytorch(model, "./save/saved_model.pkl")
print('model parameter loaded!')
except Exception as e:
print("No saved model. Continue.")
pass
# Start training
trainer.train(model, data_train, data_dev)
print("Training finished!")
# Saver
saver = ModelSaver("./save/trained_model.pkl")
saver.save_pytorch(model)
print("Model saved!")
def predict():
# Config Loader
test_args = ConfigSection()
ConfigLoader().load_config(cfgfile, {"POS_test": test_args})
# fetch dictionary size and number of labels from pickle files
word2index = load_pickle(pickle_path, "word2id.pkl")
test_args["vocab_size"] = len(word2index)
index2label = load_pickle(pickle_path, "label2id.pkl")
test_args["num_classes"] = len(index2label)
# load dev data
dev_data = load_pickle(pickle_path, "data_dev.pkl")
# Define the same model
model = AdvSeqLabel(test_args)
# Dump trained parameters into the model
ModelLoader.load_pytorch(model, "./save/trained_model.pkl")
print("model loaded!")
# Tester
test_args["evaluator"] = SeqLabelEvaluator()
tester = SeqLabelTester(**test_args.data)
# Start testing
tester.test(model, dev_data)
def infer():
# Config Loader
test_args = ConfigSection()
ConfigLoader().load_config(cfgfile, {"POS_test": test_args})
# fetch dictionary size and number of labels from pickle files
word2index = load_pickle(pickle_path, "word2id.pkl")
test_args["vocab_size"] = len(word2index)
index2label = load_pickle(pickle_path, "label2id.pkl")
test_args["num_classes"] = len(index2label)
# Define the same model
model = AdvSeqLabel(test_args)
try:
ModelLoader.load_pytorch(model, "./save/trained_model.pkl")
print('model loaded!')
except Exception as e:
print('cannot load model!')
raise
# Data Loader
infer_data = SeqLabelDataSet(load_func=BaseLoader.load_lines)
infer_data.load(data_infer_path,
vocabs={"word_vocab": word2index},
infer=True)
print('data loaded')
# Inference interface
infer = SeqLabelInfer(pickle_path)
results = infer.predict(model, infer_data)
print(results)
print("Inference finished!")
def test_case_2(self):
config = "[section_A]\n[section_B]\n"
with open("./test.cfg", "w", encoding="utf-8") as f:
f.write(config)
saver = ConfigSaver("./test.cfg")
section = ConfigSection()
section["doubles"] = 0.8
section["tt"] = [1, 2, 3]
section["test"] = 105
section["str"] = "this is a str"
saver.save_config_file("section_A", section)
os.system("rm ./test.cfg")
def _load_all(src):
model_path = src
src = os.path.dirname(src)
word_v = _load(src + '/word_v.pkl')
pos_v = _load(src + '/pos_v.pkl')
tag_v = _load(src + '/tag_v.pkl')
pos_pp = torch.load(src + '/pos_pp.pkl')['pipeline']
model_args = ConfigSection()
ConfigLoader.load_config('cfg.cfg', {'model': model_args})
model_args['word_vocab_size'] = len(word_v)
model_args['pos_vocab_size'] = len(pos_v)
model_args['num_label'] = len(tag_v)
model = BiaffineParser(**model_args.data)
model.load_state_dict(torch.load(model_path))
return {
'word_v': word_v,
'pos_v': pos_v,
'tag_v': tag_v,
'model': model,
'pos_pp': pos_pp,
}
# emb_file_name = '/home/yfshao/glove.6B.100d.txt'
# loader = ConlluDataLoader()
datadir = '/home/yfshao/workdir/parser-data/'
train_data_name = "train_ctb5.txt"
dev_data_name = "dev_ctb5.txt"
test_data_name = "test_ctb5.txt"
emb_file_name = "/home/yfshao/workdir/parser-data/word_OOVthr_30_100v.txt"
# emb_file_name = "/home/yfshao/workdir/word_vector/cc.zh.300.vec"
loader = CTBDataLoader()
cfgfile = './cfg.cfg'
processed_datadir = './save'
# Config Loader
train_args = ConfigSection()
test_args = ConfigSection()
model_args = ConfigSection()
optim_args = ConfigSection()
ConfigLoader.load_config(
cfgfile, {
"train": train_args,
"test": test_args,
"model": model_args,
"optim": optim_args
})
print('trainre Args:', train_args.data)
print('test Args:', test_args.data)
print('optim Args:', optim_args.data)
import os
import torch
from fastNLP import Trainer
from fastNLP import Tester
from fastNLP import CrossEntropyLoss
from fastNLP import Adam
from fastNLP import AccuracyMetric
from fastNLP.io.config_io import ConfigSection, ConfigLoader
import fastNLP.core.utils as util
from model import myESIM
args = ConfigSection()
ConfigLoader().load_config("../data/config.json", {"train": args})
# 加载训练、验证数据集和词向量
print("\t* Loading train data...")
train_data = util.load_pickle(os.path.normpath(args["data_dir"]),
args["train_file"])
print("\t* Loading dev data...")
dev_data = util.load_pickle(os.path.normpath(args["data_dir"]),
args["dev_file"])
print("\t* Loading word embeddings...")
embeddings = util.load_pickle(os.path.normpath(args["data_dir"]),
args["embeddings_file"])
embeddings = torch.Tensor(embeddings)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = myESIM(embeddings.shape[0],
embeddings.shape[1],
300,
embeddings=embeddings,
dropout=0.5,
def train(train_data_path, dev_data_path, checkpoint=None, save=None):
# load config
train_param = ConfigSection()
model_param = ConfigSection()
ConfigLoader().load_config(cfgfile, {
"train": train_param,
"model": model_param
})
print("config loaded")
# Data Loader
print("loading training set...")
dataset = ConllxDataLoader().load(train_data_path, return_dataset=True)
print("loading dev set...")
dev_data = ConllxDataLoader().load(dev_data_path, return_dataset=True)
print(dataset)
print("================= dataset ready =====================")
dataset.rename_field("tag", "truth")
dev_data.rename_field("tag", "truth")
vocab_proc = VocabIndexerProcessor("words",
new_added_filed_name="word_seq")
tag_proc = VocabIndexerProcessor("truth", is_input=True)
seq_len_proc = SeqLenProcessor(field_name="word_seq",
new_added_field_name="word_seq_origin_len",
is_input=True)
set_input_proc = SetInputProcessor("word_seq", "word_seq_origin_len")
vocab_proc(dataset)
tag_proc(dataset)
seq_len_proc(dataset)
# index dev set
word_vocab, tag_vocab = vocab_proc.vocab, tag_proc.vocab
dev_data.apply(lambda ins: [word_vocab.to_index(w) for w in ins["words"]],
new_field_name="word_seq")
dev_data.apply(lambda ins: [tag_vocab.to_index(w) for w in ins["truth"]],
new_field_name="truth")
dev_data.apply(lambda ins: len(ins["word_seq"]),
new_field_name="word_seq_origin_len")
# set input & target
dataset.set_input("word_seq", "word_seq_origin_len", "truth")
dev_data.set_input("word_seq", "word_seq_origin_len", "truth")
dataset.set_target("truth", "word_seq_origin_len")
dev_data.set_target("truth", "word_seq_origin_len")
# dataset.set_is_target(tag_ids=True)
model_param["vocab_size"] = vocab_proc.get_vocab_size()
model_param["num_classes"] = tag_proc.get_vocab_size()
print("vocab_size={} num_classes={}".format(model_param["vocab_size"],
model_param["num_classes"]))
# define a model
if checkpoint is None:
# pre_trained = load_tencent_embed("/home/zyfeng/data/char_tencent_embedding.pkl", vocab_proc.vocab.word2idx)
pre_trained = None
model = AdvSeqLabel(model_param, id2words=None, emb=pre_trained)
print(model)
else:
model = torch.load(checkpoint)
# call trainer to train
trainer = Trainer(dataset,
model,
loss=None,
metrics=SpanFPreRecMetric(
tag_proc.vocab,
pred="predict",
target="truth",
seq_lens="word_seq_origin_len"),
dev_data=dev_data,
metric_key="f",
use_tqdm=True,
use_cuda=True,
print_every=10,
n_epochs=20,
save_path=save)
trainer.train(load_best_model=True)
# save model & pipeline
model_proc = ModelProcessor(model,
seq_len_field_name="word_seq_origin_len")
id2tag = Index2WordProcessor(tag_proc.vocab, "predict", "tag")
pp = Pipeline(
[vocab_proc, seq_len_proc, set_input_proc, model_proc, id2tag])
save_dict = {"pipeline": pp, "model": model, "tag_vocab": tag_proc.vocab}
torch.save(save_dict, os.path.join(save, "model_pp.pkl"))
print("pipeline saved")
def train(checkpoint=None):
# load config
train_param = ConfigSection()
model_param = ConfigSection()
ConfigLoader().load_config(cfgfile, {
"train": train_param,
"model": model_param
})
print("config loaded")
# Data Loader
dataset = ZhConllPOSReader().load("/home/hyan/train.conllx")
print(dataset)
print("dataset transformed")
dataset.rename_field("tag", "truth")
vocab_proc = VocabIndexerProcessor("words",
new_added_filed_name="word_seq")
tag_proc = VocabIndexerProcessor("truth")
seq_len_proc = SeqLenProcessor(field_name="word_seq",
new_added_field_name="word_seq_origin_len",
is_input=True)
vocab_proc(dataset)
tag_proc(dataset)
seq_len_proc(dataset)
dataset.set_input("word_seq", "word_seq_origin_len", "truth")
dataset.set_target("truth", "word_seq_origin_len")
print("processors defined")
# dataset.set_is_target(tag_ids=True)
model_param["vocab_size"] = vocab_proc.get_vocab_size()
model_param["num_classes"] = tag_proc.get_vocab_size()
print("vocab_size={} num_classes={}".format(model_param["vocab_size"],
model_param["num_classes"]))
# define a model
if checkpoint is None:
# pre_trained = load_tencent_embed("/home/zyfeng/data/char_tencent_embedding.pkl", vocab_proc.vocab.word2idx)
pre_trained = None
model = AdvSeqLabel(model_param,
id2words=tag_proc.vocab.idx2word,
emb=pre_trained)
print(model)
else:
model = torch.load(checkpoint)
# call trainer to train
trainer = Trainer(dataset,
model,
loss=None,
metrics=SpanFPreRecMetric(
tag_proc.vocab,
pred="predict",
target="truth",
seq_lens="word_seq_origin_len"),
dev_data=dataset,
metric_key="f",
use_tqdm=True,
use_cuda=True,
print_every=5,
n_epochs=6,
save_path="./save")
trainer.train(load_best_model=True)
# save model & pipeline
model_proc = ModelProcessor(model,
seq_len_field_name="word_seq_origin_len")
id2tag = Index2WordProcessor(tag_proc.vocab, "predict", "tag")
pp = Pipeline([vocab_proc, seq_len_proc, model_proc, id2tag])
save_dict = {"pipeline": pp, "model": model, "tag_vocab": tag_proc.vocab}
torch.save(save_dict, "model_pp.pkl")
print("pipeline saved")
torch.save(model, "./save/best_model.pkl")
dim=attention_unit,
num_vec=attention_hops)
self.mlp = MLP(
size_layer=[lstm_hidden_size * 2 * attention_hops, nfc, class_num])
def forward(self, x):
x_emb = self.embedding(x)
output = self.lstm(x_emb)
after_attention, penalty = self.attention(output, x)
after_attention = after_attention.view(after_attention.size(0), -1)
output = self.mlp(after_attention)
return output
def loss(self, predict, ground_truth):
print("predict:%s; g:%s" %
(str(predict.size()), str(ground_truth.size())))
print(ground_truth)
return F.cross_entropy(predict, ground_truth)
train_args = ConfigSection()
ConfigLoader("good path").load_config('config.cfg', {"train": train_args})
train_args['vocab'] = len(word2index)
trainer = ClassificationTrainer(**train_args.data)
# for k in train_args.__dict__.keys():
# print(k, train_args[k])
model = SELF_ATTENTION_YELP_CLASSIFICATION(train_args)
trainer.train(model, train_data, dev_data)
def test_fastnlp_advanced_tutorial(self):
import os
os.chdir("tutorials/fastnlp_advanced_tutorial")
from fastNLP import DataSet
from fastNLP import Instance
from fastNLP import Vocabulary
from fastNLP import Trainer
from fastNLP import Tester
# ### Instance
# Instance表示一个样本，由一个或者多个field（域、属性、特征）组成，每个field具有自己的名字以及值
# 在初始化Instance的时候可以定义它包含的field，使用"field_name=field_value"的写法
# In[2]:
# 组织一个Instance，这个Instance由premise、hypothesis、label三个field组成
instance = Instance(premise='an premise example .',
hypothesis='an hypothesis example.',
label=1)
instance
# In[3]:
data_set = DataSet([instance] * 5)
data_set.append(instance)
data_set[-2:]
# In[4]:
# 如果某一个field的类型与dataset对应的field类型不一样仍可被加入dataset中
instance2 = Instance(premise='the second premise example .',
hypothesis='the second hypothesis example.',
label='1')
try:
data_set.append(instance2)
except:
pass
data_set[-2:]
# In[5]:
# 如果某一个field的名字不对，则该instance不能被append到dataset中
instance3 = Instance(premises='the third premise example .',
hypothesis='the third hypothesis example.',
label=1)
try:
data_set.append(instance3)
except:
print('cannot append instance')
pass
data_set[-2:]
# In[6]:
# 除了文本以外，还可以将tensor作为其中一个field的value
import torch
tensor_ins = Instance(image=torch.randn(5, 5), label=0)
ds = DataSet()
ds.append(tensor_ins)
ds
from fastNLP import DataSet
from fastNLP import Instance
# 从csv读取数据到DataSet
# 类csv文件，即每一行为一个example的文件，都可以使用这种方法进行数据读取
dataset = DataSet.read_csv('tutorial_sample_dataset.csv',
headers=('raw_sentence', 'label'),
sep='\t')
# 查看DataSet的大小
len(dataset)
# In[8]:
# 使用数字索引[k]，获取第k个样本
dataset[0]
# In[9]:
# 获取的样本是一个Instance
type(dataset[0])
# In[10]:
# 使用数字索引[a: b]，获取第a到第b个样本
dataset[0:3]
# In[11]:
# 索引也可以是负数
dataset[-1]
data_path = ['premise', 'hypothesis', 'label']
# 读入文件
with open(data_path[0]) as f:
premise = f.readlines()
with open(data_path[1]) as f:
hypothesis = f.readlines()
with open(data_path[2]) as f:
label = f.readlines()
assert len(premise) == len(hypothesis) and len(hypothesis) == len(
label)
# 组织DataSet
data_set = DataSet()
for p, h, l in zip(premise, hypothesis, label):
p = p.strip() # 将行末空格去除
h = h.strip() # 将行末空格去除
data_set.append(Instance(premise=p, hypothesis=h, truth=l))
data_set[0]
# ### DataSet的其他操作
# 在构建完毕DataSet后，仍然可以对DataSet的内容进行操作，函数接口为DataSet.apply()
# In[13]:
# 将premise域的所有文本转成小写
data_set.apply(lambda x: x['premise'].lower(),
new_field_name='premise')
data_set[-2:]
# In[14]:
# label转int
data_set.apply(lambda x: int(x['truth']), new_field_name='truth')
data_set[-2:]
# In[15]:
# 使用空格分割句子
def split_sent(ins):
return ins['premise'].split()
data_set.apply(split_sent, new_field_name='premise')
data_set.apply(lambda x: x['hypothesis'].split(),
new_field_name='hypothesis')
data_set[-2:]
# In[16]:
# 筛选数据
origin_data_set_len = len(data_set)
data_set.drop(lambda x: len(x['premise']) <= 6)
origin_data_set_len, len(data_set)
# In[17]:
# 增加长度信息
data_set.apply(lambda x: [1] * len(x['premise']),
new_field_name='premise_len')
data_set.apply(lambda x: [1] * len(x['hypothesis']),
new_field_name='hypothesis_len')
data_set[-1]
# In[18]:
# 设定特征域、标签域
data_set.set_input("premise", "premise_len", "hypothesis",
"hypothesis_len")
data_set.set_target("truth")
# In[19]:
# 重命名field
data_set.rename_field('truth', 'label')
data_set[-1]
# In[20]:
# 切分训练、验证集、测试集
train_data, vad_data = data_set.split(0.5)
dev_data, test_data = vad_data.split(0.4)
len(train_data), len(dev_data), len(test_data)
# In[21]:
# 深拷贝一个数据集
import copy
train_data_2, dev_data_2 = copy.deepcopy(train_data), copy.deepcopy(
dev_data)
del copy
# 初始化词表，该词表最大的vocab_size为10000，词表中每个词出现的最低频率为2，'<unk>'表示未知词语，'<pad>'表示padding词语
# Vocabulary默认初始化参数为max_size=None, min_freq=None, unknown='<unk>', padding='<pad>'
vocab = Vocabulary(max_size=10000,
min_freq=2,
unknown='<unk>',
padding='<pad>')
# 构建词表
train_data.apply(lambda x: [vocab.add(word) for word in x['premise']])
train_data.apply(
lambda x: [vocab.add(word) for word in x['hypothesis']])
vocab.build_vocab()
# In[23]:
# 根据词表index句子
train_data.apply(
lambda x: [vocab.to_index(word) for word in x['premise']],
new_field_name='premise')
train_data.apply(
lambda x: [vocab.to_index(word) for word in x['hypothesis']],
new_field_name='hypothesis')
dev_data.apply(
lambda x: [vocab.to_index(word) for word in x['premise']],
new_field_name='premise')
dev_data.apply(
lambda x: [vocab.to_index(word) for word in x['hypothesis']],
new_field_name='hypothesis')
test_data.apply(
lambda x: [vocab.to_index(word) for word in x['premise']],
new_field_name='premise')
test_data.apply(
lambda x: [vocab.to_index(word) for word in x['hypothesis']],
new_field_name='hypothesis')
train_data[-1], dev_data[-1], test_data[-1]
# 读入vocab文件
with open('vocab.txt') as f:
lines = f.readlines()
vocabs = []
for line in lines:
vocabs.append(line.strip())
# 实例化Vocabulary
vocab_bert = Vocabulary(unknown=None, padding=None)
# 将vocabs列表加入Vocabulary
vocab_bert.add_word_lst(vocabs)
# 构建词表
vocab_bert.build_vocab()
# 更新unknown与padding的token文本
vocab_bert.unknown = '[UNK]'
vocab_bert.padding = '[PAD]'
# In[25]:
# 根据词表index句子
train_data_2.apply(
lambda x: [vocab_bert.to_index(word) for word in x['premise']],
new_field_name='premise')
train_data_2.apply(
lambda x: [vocab_bert.to_index(word) for word in x['hypothesis']],
new_field_name='hypothesis')
dev_data_2.apply(
lambda x: [vocab_bert.to_index(word) for word in x['premise']],
new_field_name='premise')
dev_data_2.apply(
lambda x: [vocab_bert.to_index(word) for word in x['hypothesis']],
new_field_name='hypothesis')
train_data_2[-1], dev_data_2[-1]
# step 1：加载模型参数（非必选）
from fastNLP.io.config_io import ConfigSection, ConfigLoader
args = ConfigSection()
ConfigLoader().load_config("./data/config", {"esim_model": args})
args["vocab_size"] = len(vocab)
args.data
# In[27]:
# step 2：加载ESIM模型
from fastNLP.models import ESIM
model = ESIM(**args.data)
model
# In[28]:
# 另一个例子：加载CNN文本分类模型
from fastNLP.models import CNNText
cnn_text_model = CNNText(embed_num=len(vocab),
embed_dim=50,
num_classes=5,
padding=2,
dropout=0.1)
cnn_text_model
from fastNLP import CrossEntropyLoss
from fastNLP import Adam
from fastNLP import AccuracyMetric
trainer = Trainer(
train_data=train_data,
model=model,
loss=CrossEntropyLoss(pred='pred', target='label'),
metrics=AccuracyMetric(),
n_epochs=3,
batch_size=16,
print_every=-1,
validate_every=-1,
dev_data=dev_data,
use_cuda=False,
optimizer=Adam(lr=1e-3, weight_decay=0),
check_code_level=-1,
metric_key='acc',
use_tqdm=False,
)
trainer.train()
tester = Tester(
data=test_data,
model=model,
metrics=AccuracyMetric(),
batch_size=args["batch_size"],
)
tester.test()
os.chdir("../..")
def test_case_1(self):
config_file_dir = "test/io/"
config_file_name = "config"
config_file_path = os.path.join(config_file_dir, config_file_name)
tmp_config_file_path = os.path.join(config_file_dir, "tmp_config")
with open(config_file_path, "r") as f:
lines = f.readlines()
standard_section = ConfigSection()
t_section = ConfigSection()
ConfigLoader().load_config(config_file_path, {
"test": standard_section,
"t": t_section
})
config_saver = ConfigSaver(config_file_path)
section = ConfigSection()
section["doubles"] = 0.8
section["tt"] = 0.5
section["test"] = 105
section["str"] = "this is a str"
test_case_2_section = section
test_case_2_section["double"] = 0.5
for k in section.__dict__.keys():
standard_section[k] = section[k]
config_saver.save_config_file("test", section)
config_saver.save_config_file("another-test", section)
config_saver.save_config_file("one-another-test", section)
config_saver.save_config_file("test-case-2", section)
test_section = ConfigSection()
at_section = ConfigSection()
another_test_section = ConfigSection()
one_another_test_section = ConfigSection()
a_test_case_2_section = ConfigSection()
ConfigLoader().load_config(
config_file_path, {
"test": test_section,
"another-test": another_test_section,
"t": at_section,
"one-another-test": one_another_test_section,
"test-case-2": a_test_case_2_section
})
assert test_section == standard_section
assert at_section == t_section
assert another_test_section == section
assert one_another_test_section == section
assert a_test_case_2_section == test_case_2_section
config_saver.save_config_file("test", section)
with open(config_file_path, "w") as f:
f.writelines(lines)
with open(tmp_config_file_path, "w") as f:
f.write('[test]\n')
f.write('this is an fault example\n')
tmp_config_saver = ConfigSaver(tmp_config_file_path)
try:
tmp_config_saver._read_section()
except Exception as e:
pass
os.remove(tmp_config_file_path)
try:
tmp_config_saver = ConfigSaver("file-NOT-exist")
except Exception as e:
pass