def prepare_data_and_model(Model, args, using_gpu=True):
if args.test:
## # narvi
#train_path = "/home/zhouy/thesis/data/text_classification_data/train_try.csv"
#test_path = "/home/zhouy/thesis/data/text_classification_data/test_try.csv"
# tut thinkstation
# train_path = "/media/yi/harddrive/codes/thesis_sentimentAnalysis/data/text_classification_data/train_try.csv"
# test_path = "/media/yi/harddrive/codes/thesis_sentimentAnalysis/data/text_classification_data/test_try.csv"
# # tripadvisor dataset
# # xps
test_path = "D:\\sentimentAnalysis\\data\\text_classification_data\\test_model_data\\rev_sent_5_score_train_test\\tripadvisor\\test_try.csv"
train_path = "D:\\sentimentAnalysis\\data\\text_classification_data\\test_model_data\\rev_sent_5_score_train_test\\tripadvisor\\train_try.csv"
else:
# original dataset
# # narvi
#train_path = "/home/zhouy/thesis/data/text_classification_data/tripadvisor_train_dataset.csv"
#test_path = "/home/zhouy/thesis/data/text_classification_data/tripadvisor_test_dataset.csv"
# # tut thinkstation
# train_path = "/home/yi/sentimentAnalysis/algos/5_ToxicCommentClassification-pytorch/data/train.csv"
# test_path = "/home/yi/sentimentAnalysis/algos/5_ToxicCommentClassification-pytorch/data/test.csv"
# # xps
# train_path = "D:/sentimentAnalysis/algos/5_ToxicCommentClassification-pytorch/data/train.csv"
# test_path = "D:/sentimentAnalysis/algos/5_ToxicCommentClassification-pytorch/data/test.csv"
# tripadvisor dataset
# xps
train_path = "D:/sentimentAnalysis/data/text_classification_data/tripadvisor_train_dataset.csv"
test_path = "D:/sentimentAnalysis/data/text_classification_data/tripadvisor_test_dataset.csv"
def tokenize(text):
fileters = '!"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\t\n'
trans_map = str.maketrans(fileters, " " * len(fileters))
text = text.translate(trans_map)
text = [
tok.text for tok in spacy_en.tokenizer(text) if tok.text != ' '
]
tokenized_text = []
auxiliary_verbs = ['am', 'is', 'are', 'was', 'were', "'s"]
for token in text:
if token == "n't":
tmp = 'not'
elif token == "'ll":
tmp = 'will'
elif token in auxiliary_verbs:
tmp = 'be'
else:
tmp = token
tokenized_text.append(tmp)
return tokenized_text
if args.dataset == 'tripadvisor':
TEXT = data.Field(tokenize=tokenize,
lower=True,
batch_first=True,
truncate_first=True)
LABEL = data.Field(sequential=False, use_vocab=False, batch_first=True)
test = CustomDataset(test_path,
text_field=TEXT,
label_field=LABEL,
test=True)
train = CustomDataset(train_path, text_field=TEXT, label_field=LABEL)
# should save the above train, test, these two variables.
if args.wordembedding == "glove-6b":
vectors = GloVe(name='6B', dim=args.embed_dim)
elif args.wordembedding == "FastText":
vectors = FastText(language='en')
else:
NotImplementedError
# # FastText
# vectors = FastText(name='6B', dim=args.embed_dim)
vectors.unk_init = init.xavier_uniform
# 下面这行代码报错
# TEXT.build_vocab(train, vectors=vectors, max_size=30000)
TEXT.build_vocab(train, vectors=vectors, max_size=10000, min_freq=10)
LABEL.build_vocab(train)
print('train.fields', train.fields)
print('train.name', getattr(train, 'text'))
print('len(train)', len(train))
print('vars(train[0])', vars(train[0]))
# using the training corpus to create the vocabulary
train_iter = data.Iterator(dataset=train,
batch_size=args.batch_size,
train=True,
repeat=False,
device=0 if using_gpu else -1)
test_iter = data.Iterator(dataset=test,
batch_size=args.batch_size,
train=False,
sort=False,
device=0 if using_gpu else -1)
# the number of unique words
num_tokens = len(TEXT.vocab.itos)
args.num_tokens = num_tokens
dev_iter = test_iter
elif args.dataset == 'SST':
text_field = data.Field(batch_first=True,
lower=True,
tokenize=tokenize)
label_field = data.Field(sequential=False, batch_first=True)
train_data, dev_data, test_data = datasets.SST.splits(
text_field, label_field, fine_grained=True)
vectors = GloVe(name='6B', dim=args.embed_dim)
text_field.build_vocab(train_data, vectors=vectors, min_freq=1)
label_field.build_vocab(train_data)
train_iter = data.Iterator(train_data,
batch_size=args.batch_size,
device=0 if using_gpu else -1,
train=True,
repeat=False,
sort=False,
shuffle=True)
dev_iter = data.Iterator(dev_data,
batch_size=args.batch_size,
device=0 if using_gpu else -1,
train=False,
repeat=False,
sort=False,
shuffle=False)
test_iter = data.Iterator(test_data,
batch_size=args.batch_size,
device=0 if using_gpu else -1,
train=False,
repeat=False,
sort=False,
shuffle=False)
# train_iter, dev_iter, test_iter = sst(text_field, label_field)
# train_iter, dev_iter, test_iter = SST.iters(batch_size=16, device=0 if using_gpu else -1, vectors="glove.6B.300d")
# config.target_class = train_iter.dataset.NUM_CLASSES
args.num_tokens = len(text_field.vocab)
args.num_classes = len(label_field.vocab) - 1
print("num_classes: ", args.num_classes)
if args.model == "VDCNN":
net = Model(depth=29,
vocabulary_size=args.num_tokens,
embed_size=16,
n_classes=args.num_classes,
k=2,
optional_shortcut=True)
else:
net = Model(args)
# # copy pretrained glove word embedding into the model
# net.embedding.weight.data.copy_(TEXT.vocab.vectors)
if using_gpu:
net.cuda()
return train_iter, test_iter, net
def __init__(self, batch_size=30, device=-1):
self.batch_size = batch_size
self.device = device
#Define fields
TEXT = data.Field(lower=True, include_lengths=False, batch_first=True)
CHAR = data.Field(lower=True,
include_lengths=False,
batch_first=True,
tokenize=list)
TEXT_C = data.NestedField(CHAR)
LABEL = data.Field(sequential=False, use_vocab=False, batch_first=True)
INDEX = data.Field(sequential=False, use_vocab=False, batch_first=True)
ID = data.RawField()
fields = [("context", TEXT), ("query", TEXT), ("label", LABEL),
("context_c", TEXT_C), ("query_c", TEXT_C), ("index", INDEX)]
train_data = []
val_data = []
dev_data = []
#Generate examples
print("Loading datasets...")
print("Loading training set...")
try:
with open("./data/processed/train_set.data", 'rb') as f:
train_data = pickle.load(f)
print("Loaded training set from file.")
except:
print(
"Failed to loaded training set from file. Processing training data..."
)
with open("./data/squad/train.context") as f:
context = list(f)
with open("./data/squad/train.question") as f:
query = list(f)
with open("./data/squad/train.span") as f:
label = list(f)
for i in range(len(label)):
splited = list(map(int, label[i].split()))
label[i] = splited
for i in tqdm(range(len(context)), ascii=True):
list_content = [
context[i], query[i], label[i], context[i], query[i], i
]
train_ex = data.Example.fromlist(list_content, fields)
train_data.append(train_ex)
with open("./data/processed/train_set.data", 'wb') as f:
pickle.dump(train_data, f)
train_set = data.Dataset(train_data, fields)
print("Loading dev set...")
try:
with open("./data/processed/dev_set.data", 'rb') as f:
dev_data = pickle.load(f)
print("Loaded dev set from file.")
except:
print("Failed to loaded dev set from file. Processing dev data...")
with open("./data/squad/dev.context") as f:
context = list(f)
with open("./data/squad/dev.question") as f:
query = list(f)
with open("./data/squad/dev.span") as f:
label = list(f)
for i in range(len(label)):
splited = list(map(int, label[i].split()))
label[i] = splited
for i in tqdm(range(len(context)), ascii=True):
list_content = [
context[i], query[i], label[i], context[i], query[i], i
]
dev_ex = data.Example.fromlist(list_content, fields)
dev_data.append(dev_ex)
with open("./data/processed/dev_set.data", 'wb') as f:
pickle.dump(dev_data, f)
dev_set = data.Dataset(dev_data, fields)
print("Loading validation set...")
try:
with open("./data/processed/val_set.data", 'rb') as f:
val_data = pickle.load(f)
print("Loaded validation set from file.")
except:
print(
"Failed to loaded validation set from file. Processing validation data..."
)
with open("./data/squad/val.context") as f:
context = list(f)
with open("./data/squad/val.question") as f:
query = list(f)
with open("./data/squad/val.span") as f:
label = list(f)
for i in range(len(label)):
splited = list(map(int, label[i].split()))
label[i] = splited
for i in tqdm(range(len(context)), ascii=True):
list_content = [
context[i], query[i], label[i], context[i], query[i], i
]
val_ex = data.Example.fromlist(list_content, fields)
val_data.append(val_ex)
with open("./data/processed/val_set.data", 'wb') as f:
pickle.dump(val_data, f)
val_set = data.Dataset(val_data, fields)
print("Loading word embeddings...")
glove_vecs = GloVe(name='6B', dim=100)
glove_vecs.unk_init = nn.init.xavier_uniform_
print("Building vocabulary...")
TEXT.build_vocab(train_set, vectors=glove_vecs)
TEXT_C.build_vocab(train_set, min_freq=20)
self.vocab_vec = TEXT.vocab.vectors
print(len(self.vocab_vec), " words in word vocabulary.")
self.char_size = len(TEXT_C.vocab)
print(len(TEXT_C.vocab), " tokens in char vocabulary.")
print("Generating iterator...")
self.train_iter = iter(
data.Iterator(train_set,
batch_size=self.batch_size,
device=self.device,
sort_key=lambda x: len(x.context),
repeat=True,
sort=True))
self.dev_iter = iter(
data.Iterator(dev_set,
batch_size=self.batch_size,
device=self.device,
sort_key=lambda x: len(x.context),
repeat=False,
sort=True))
self.val_iter = iter(
data.Iterator(val_set,
batch_size=self.batch_size,
device=self.device,
sort_key=lambda x: len(x.context),
repeat=True,
sort=True))
print("DataLoader initiated.")
