@@ -27,8 +27,8 @@ class Action(object): | |||
:return iteration:int, the number of step in each epoch | |||
generator:generator, to generate batch inputs | |||
""" | |||
n_samples = X.shape[0] | |||
num_iter = n_samples / batch_size | |||
n_samples = X.size()[0] | |||
num_iter = n_samples // batch_size | |||
if Y is None: | |||
generator = self._batch_generate(batch_size, num_iter, X) | |||
else: | |||
@@ -39,8 +39,8 @@ class Action(object): | |||
def _batch_generate(batch_size, num_iter, *data): | |||
for step in range(num_iter): | |||
start = batch_size * step | |||
end = (batch_size + 1) * step | |||
yield tuple([x[start:end, :] for x in data]) | |||
end = batch_size * (step + 1) | |||
yield tuple([x[start:end] for x in data]) | |||
def make_log(self, *args): | |||
return "log" |
@@ -27,17 +27,18 @@ class Tester(Action): | |||
self.batch_size = test_args.batch_size | |||
def test(self, network, data): | |||
print("testing") | |||
network.mode(test=True) # turn on the testing mode | |||
if not self.save_dev_input: | |||
# transform into network input and label | |||
valid_x, valid_y = network.prepare_input(data) | |||
if self.validate_in_training: | |||
if self.save_dev_input: | |||
if self.valid_x is None: | |||
valid_x, valid_y = network.prepare_input(data) | |||
self.valid_x = valid_x | |||
self.valid_y = valid_y | |||
else: | |||
valid_x = self.valid_x | |||
valid_y = self.valid_y | |||
else: | |||
valid_x = self.valid_x | |||
valid_y = self.valid_y | |||
valid_x, valid_y = network.prepare_input(data) | |||
# split into batches by self.batch_size | |||
iterations, test_batch_generator = self.batchify(self.batch_size, valid_x, valid_y) | |||
@@ -53,10 +54,10 @@ class Tester(Action): | |||
# forward pass from tests input to predicted output | |||
prediction = network.data_forward(batch_x) | |||
loss = network.loss(batch_y, prediction) | |||
loss = network.get_loss(prediction, batch_y) | |||
if self.save_output: | |||
batch_output.append(prediction) | |||
batch_output.append(prediction.data) | |||
if self.save_loss: | |||
loss_history.append(loss) | |||
self.log(self.make_log(step, loss)) | |||
@@ -74,9 +75,10 @@ class Tester(Action): | |||
def result(self): | |||
return self.output | |||
def make_output(self, batch_output): | |||
@staticmethod | |||
def make_output(batch_outputs): | |||
# construct full prediction with batch outputs | |||
return np.concatenate((batch_output[0], batch_output[1]), axis=0) | |||
return np.concatenate(batch_outputs, axis=0) | |||
def load_config(self, args): | |||
raise NotImplementedError | |||
@@ -8,7 +8,8 @@ class Trainer(Action): | |||
""" | |||
Trainer for common training logic of all models | |||
""" | |||
TrainConfig = namedtuple("config", ["epochs", "validate", "save_when_better", "log_per_step", "log_validation"]) | |||
TrainConfig = namedtuple("config", ["epochs", "validate", "save_when_better", | |||
"log_per_step", "log_validation", "batch_size"]) | |||
def __init__(self, train_args): | |||
""" | |||
@@ -20,6 +21,7 @@ class Trainer(Action): | |||
self.save_when_better = train_args.save_when_better | |||
self.log_per_step = train_args.log_per_step | |||
self.log_validation = train_args.log_validation | |||
self.batch_size = train_args.batch_size | |||
def train(self, network, train_data, dev_data): | |||
""" | |||
@@ -28,20 +30,19 @@ class Trainer(Action): | |||
:param dev_data: raw data for validation | |||
:return: | |||
""" | |||
train_x, train_y = network.prepare_input(train_data.train_set, train_data.train_label) | |||
train_x, train_y = network.prepare_input(train_data) | |||
network.mode(test=False) # turn on the train mode | |||
iterations, train_batch_generator = self.batchify(train_x, train_y) | |||
iterations, train_batch_generator = self.batchify(self.batch_size, train_x, train_y) | |||
test_args = Tester.TestConfig(save_output=True, validate_in_training=True, | |||
save_dev_input=True, save_loss=True, batch_size=16) | |||
save_dev_input=True, save_loss=True, batch_size=self.batch_size) | |||
evaluator = Tester(test_args) | |||
best_loss = 1e10 | |||
loss_history = list() | |||
for epoch in range(self.n_epochs): | |||
network.mode(test=False) # turn on the train mode | |||
network.define_optimizer() | |||
for step in range(iterations): | |||
@@ -49,10 +50,11 @@ class Trainer(Action): | |||
prediction = network.data_forward(batch_x) | |||
loss = network.loss(batch_y, prediction) | |||
loss = network.get_loss(prediction, batch_y) | |||
network.grad_backward() | |||
if step % self.log_per_step == 0: | |||
print("step ", step) | |||
loss_history.append(loss) | |||
self.log(self.make_log(epoch, step, loss)) | |||
@@ -24,7 +24,7 @@ class BaseModel(object): | |||
def grad_backward(self): | |||
raise NotImplementedError | |||
def loss(self, pred, truth): | |||
def get_loss(self, pred, truth): | |||
raise NotImplementedError | |||
@@ -50,7 +50,7 @@ class ToyModel(BaseModel): | |||
def grad_backward(self): | |||
print("loss gradient backward") | |||
def loss(self, pred, truth): | |||
def get_loss(self, pred, truth): | |||
self._loss = np.mean(np.square(pred - truth)) | |||
return self._loss | |||
@@ -10,6 +10,8 @@ from torch.autograd import Variable | |||
from model.base_model import BaseModel | |||
USE_GPU = True | |||
class CharLM(BaseModel): | |||
@@ -20,16 +22,16 @@ class CharLM(BaseModel): | |||
""" | |||
DataTuple = namedtuple("DataTuple", ["feature", "label"]) | |||
def __init__(self): | |||
def __init__(self, lstm_batch_size, lstm_seq_len): | |||
super(CharLM, self).__init__() | |||
""" | |||
Settings: should come from config loader or pre-processing | |||
""" | |||
self.word_embed_dim = 100 | |||
self.word_embed_dim = 300 | |||
self.char_embedding_dim = 15 | |||
self.cnn_batch_size = 40 | |||
self.lstm_seq_len = 10 | |||
self.lstm_batch_size = 4 | |||
self.cnn_batch_size = lstm_batch_size * lstm_seq_len | |||
self.lstm_seq_len = lstm_seq_len | |||
self.lstm_batch_size = lstm_batch_size | |||
self.num_epoch = 10 | |||
self.old_PPL = 100000 | |||
self.best_PPL = 100000 | |||
@@ -45,8 +47,9 @@ class CharLM(BaseModel): | |||
self.data = None # named tuple to store all data set | |||
self.data_ready = False | |||
self.criterion = nn.CrossEntropyLoss() | |||
self.loss = None | |||
self.use_gpu = False | |||
self._loss = None | |||
self.use_gpu = USE_GPU | |||
# word_emb_dim == hidden_size / num of hidden units | |||
self.hidden = (to_var(torch.zeros(2, self.lstm_batch_size, self.word_embed_dim)), | |||
to_var(torch.zeros(2, self.lstm_batch_size, self.word_embed_dim))) | |||
@@ -64,7 +67,7 @@ class CharLM(BaseModel): | |||
def prepare_input(self, raw_text): | |||
""" | |||
:param raw_text: raw input data | |||
:param raw_text: raw input text consisting of words | |||
:return: torch.Tensor, torch.Tensor | |||
feature matrix, label vector | |||
This function is only called once in Trainer.train, but may called multiple times in Tester.test | |||
@@ -78,17 +81,12 @@ class CharLM(BaseModel): | |||
max_word_len = self.max_word_len | |||
print("word/char dictionary built. Start making inputs.") | |||
input_vec = np.array(text2vec(raw_text, char_dict, max_word_len)) | |||
words = raw_text | |||
input_vec = np.array(text2vec(words, char_dict, max_word_len)) | |||
# Labels are next-word index in word_dict with the same length as inputs | |||
input_label = np.array([word_dict[w] for w in raw_text[1:]] + [word_dict[raw_text[-1]]]) | |||
data = self.DataTuple(feature=input_vec, label=input_label) | |||
feature_input = torch.from_numpy(data.feature) | |||
label_input = torch.from_numpy(data.label) | |||
num_seq = feature_input.size()[0] // self.lstm_seq_len | |||
feature_input = feature_input[:num_seq * self.lstm_seq_len, :] | |||
feature_input = feature_input.view(-1, self.lstm_seq_len, self.max_word_len + 2) | |||
input_label = np.array([word_dict[w] for w in words[1:]] + [word_dict[words[-1]]]) | |||
feature_input = torch.from_numpy(input_vec) | |||
label_input = torch.from_numpy(input_label) | |||
return feature_input, label_input | |||
def mode(self, test=False): | |||
@@ -98,6 +96,15 @@ class CharLM(BaseModel): | |||
self.model.train() | |||
def data_forward(self, x): | |||
""" | |||
:param x: Tensor of size [lstm_batch_size, lstm_seq_len, max_word_len+2] | |||
:return: Tensor of size [num_words, ?] | |||
""" | |||
# additional processing of inputs after batching | |||
num_seq = x.size()[0] // self.lstm_seq_len | |||
x = x[:num_seq * self.lstm_seq_len, :] | |||
x = x.view(-1, self.lstm_seq_len, self.max_word_len + 2) | |||
# detach hidden state of LSTM from last batch | |||
hidden = [state.detach() for state in self.hidden] | |||
output, self.hidden = self.model(to_var(x), hidden) | |||
@@ -105,13 +112,13 @@ class CharLM(BaseModel): | |||
def grad_backward(self): | |||
self.model.zero_grad() | |||
self.loss.backward() | |||
self._loss.backward() | |||
torch.nn.utils.clip_grad_norm(self.model.parameters(), 5, norm_type=2) | |||
self.optimizer.step() | |||
def loss(self, predict, truth): | |||
self.loss = self.criterion(predict, to_var(truth)) | |||
return self.loss | |||
def get_loss(self, predict, truth): | |||
self._loss = self.criterion(predict, to_var(truth)) | |||
return self._loss.data # No pytorch data structure exposed outsides | |||
def define_optimizer(self): | |||
# redefine optimizer for every new epoch | |||
@@ -123,12 +130,13 @@ class CharLM(BaseModel): | |||
def preprocess(self, all_text_files): | |||
word_dict, char_dict = create_word_char_dict(all_text_files) | |||
self.num_char = len(char_dict) | |||
num_char = len(char_dict) | |||
self.vocab_size = len(word_dict) | |||
char_dict["BOW"] = self.num_char + 1 | |||
char_dict["EOW"] = self.num_char + 2 | |||
char_dict["BOW"] = num_char + 1 | |||
char_dict["EOW"] = num_char + 2 | |||
char_dict["PAD"] = 0 | |||
# dict of (int, string) | |||
self.num_char = num_char + 3 | |||
# char_dict is a dict of (int, string), int counting from 0 to 47 | |||
reverse_word_dict = {value: key for key, value in word_dict.items()} | |||
self.max_word_len = max([len(word) for word in word_dict]) | |||
objects = { | |||
@@ -194,7 +202,7 @@ def create_word_char_dict(*file_name): | |||
def to_var(x): | |||
if torch.cuda.is_available(): | |||
if torch.cuda.is_available() and USE_GPU: | |||
x = x.cuda() | |||
return Variable(x) | |||
@@ -246,7 +254,8 @@ class charLM(nn.Module): | |||
self.convolutions = [] | |||
# list of tuples: (the number of filter, width) | |||
self.filter_num_width = [(25, 1), (50, 2), (75, 3), (100, 4), (125, 5), (150, 6)] | |||
# self.filter_num_width = [(25, 1), (50, 2), (75, 3), (100, 4), (125, 5), (150, 6)] | |||
self.filter_num_width = [(25, 1), (50, 2), (75, 3)] | |||
for out_channel, filter_width in self.filter_num_width: | |||
self.convolutions.append( | |||
@@ -304,7 +313,7 @@ class charLM(nn.Module): | |||
# [num_seq*seq_len, max_word_len+2, char_emb_dim] | |||
x = torch.transpose(x.view(x.size()[0], 1, x.size()[1], -1), 2, 3) | |||
# [num_seq*seq_len, 1, max_word_len+2, char_emb_dim] | |||
# [num_seq*seq_len, 1, char_emb_dim, max_word_len+2] | |||
x = self.conv_layers(x) | |||
# [num_seq*seq_len, total_num_filters] | |||
@@ -6,10 +6,11 @@ from model.char_language_model import CharLM | |||
def test_charlm(): | |||
train_config = Trainer.TrainConfig(epochs=1, validate=True, save_when_better=True, | |||
log_per_step=10, log_validation=True) | |||
log_per_step=10, log_validation=True, batch_size=160) | |||
trainer = Trainer(train_config) | |||
model = CharLM() | |||
model = CharLM(lstm_batch_size=16, lstm_seq_len=10) | |||
train_data = ToyLoader0("load_train", "./data_for_tests/charlm.txt").load() | |||
valid_data = ToyLoader0("load_valid", "./data_for_tests/charlm.txt").load() | |||
@@ -18,7 +19,7 @@ def test_charlm(): | |||
trainer.save_model(model) | |||
test_config = Tester.TestConfig(save_output=True, validate_in_training=True, | |||
save_dev_input=True, save_loss=True, batch_size=16) | |||
save_dev_input=True, save_loss=True, batch_size=160) | |||
tester = Tester(test_config) | |||
test_data = ToyLoader0("load_test", "./data_for_tests/charlm.txt").load() | |||