-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
203 lines (164 loc) · 7.19 KB
/
model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
import pickle
import io
import torch
import torch.nn as nn
from tqdm import tqdm
from torch.utils.data import Dataset, DataLoader
from word_encoder import BasicWordEncoder
torch.manual_seed(1)
class LabeledDataset(Dataset):
def __init__(self, df):
x = df[df.columns[:-1]]
y = df[df.columns[-1:]]
self.x = torch.tensor(x.values, dtype=torch.long)
self.y = torch.tensor(y.values, dtype=torch.long)
def __len__(self):
return len(self.y)
def __getitem__(self, idx):
return self.x[idx], self.y[idx].squeeze()
class NeuralNetwork(torch.nn.Module):
def __init__(self, dictionary_size, embedding_dim,
num_of_tokens, num_transitions,
hidden_size,
embedding_weight=None) -> None:
super().__init__()
if embedding_weight is not None:
# normal
# embedding_weight = torch.randn((dictionary_size, embedding_dim))*embedding_weight
# uniform
embedding_weight = torch.rand((dictionary_size, embedding_dim))*embedding_weight
self.dictionary_size = dictionary_size
self.num_of_tokens = num_of_tokens
self.num_transitions = num_transitions
self.embeddings = nn.Embedding(
dictionary_size, embedding_dim, _weight=embedding_weight
)
self.linear1 = nn.Linear(num_of_tokens * embedding_dim, hidden_size)
self.activation = torch.nn.Tanh()
self.dropout = nn.Dropout(p=0.5)
self.linear2 = nn.Linear(hidden_size, num_transitions)
def forward(self, x):
embeds = self.embeddings(x).view((x.shape[0], -1))
out = self.activation(self.linear1(embeds))
out = self.linear2(out)
return out
#########################################################################
# CPU_Unpickler taken from:
# https://github.com/pytorch/pytorch/issues/16797#issuecomment-633423219
#########################################################################
class CPU_Unpickler(pickle.Unpickler):
def find_class(self, module, name):
if module == 'torch.storage' and name == '_load_from_bytes':
return lambda b: torch.load(io.BytesIO(b), map_location='cpu')
return super().find_class(module, name)
class Model:
def __init__(self,
word_encoder: BasicWordEncoder, embedding_dim, num_of_tokens,
hidden_size, learning_rate, regularization_rate,
embedding_weight=None,
use_gpu=True) -> None:
self.use_gpu = use_gpu
self.word_encoder = word_encoder
self.network = NeuralNetwork(
word_encoder.get_dictionary_size(),
embedding_dim,
num_of_tokens,
word_encoder.get_num_of_labels(),
hidden_size,
embedding_weight=embedding_weight
)
self.init_device()
self.optimizer = torch.optim.Adagrad(
self.network.parameters(), lr=learning_rate, lr_decay=0,
weight_decay=regularization_rate,
initial_accumulator_value=0, eps=1e-10
)
self.criterion = nn.CrossEntropyLoss()
print(self.network)
print(f'Emb_W_scaling = {embedding_weight}, lr = {learning_rate}, '
f'reg_rate = {regularization_rate}')
def init_device(self):
if self.use_gpu and torch.cuda.is_available():
print('Using GPU')
self.device = torch.device('cuda')
else:
print('Using CPU')
self.device = torch.device('cpu')
self.network.to(self.device)
def debug(self, statement='', end="\n", flush=True):
if self.debug_file:
self.debug_file.write(str(statement) + end)
print(statement, end=end, flush=flush)
def train_model(self,
train_loader: DataLoader, dev_loader: DataLoader,
epochs: int, verbose=1):
epochs_trange = tqdm(range(epochs), desc='Epochs')
for _ in epochs_trange:
train_loss, train_acc =\
self._train_epoch(train_loader, epochs_trange)
val_loss, val_acc = self.evaluate(dev_loader)
epochs_trange.set_postfix({
'loss': f'{train_loss:5.2f}',
'acc' : f'{train_acc: < .2f}',
'vloss': f'{val_loss:5.2f}',
'vacc' : f'{val_acc: < .2f}'
})
print()
def _train_epoch(self, train_loader: DataLoader, epochs_trange: tqdm):
num_batches = len(train_loader)
accumulated_training_loss = 0
training_corrects = 0
self.network.train() # turn on training mode
# loop over the data iterator, and feed the inputs to the network and adjust the weights.
for batch_idx, (X, y) in enumerate(train_loader):
X = X.to(self.device)
y = y.to(self.device)
preds = self.network(X)
loss = self.criterion(preds, y)
# Backpropagation
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
training_corrects += (preds.argmax(1) == y).sum().item()
accumulated_training_loss += loss.item()
epochs_trange.set_postfix({
'batch' : f'{batch_idx + 1:5d}/{num_batches:5d} ',
'loss': f'{accumulated_training_loss/(batch_idx+1): < .2f}',
'acc' : f'{training_corrects/((batch_idx+1)*X.shape[0]): < .2f}'
})
return accumulated_training_loss/(batch_idx+1), training_corrects/len(train_loader.dataset)
def evaluate(self, dev_loader: DataLoader) -> float:
self.network.eval() # turn on evaluation mode
total_loss = 0.
corrects = 0
with torch.no_grad():
for _, (X, y) in enumerate(dev_loader):
X = X.to(self.device)
y = y.to(self.device)
preds = self.network(X)
total_loss += self.criterion(preds, y).item()
corrects += (preds.argmax(1) == y).sum().item()
return total_loss / (len(dev_loader) - 1), corrects/len(dev_loader.dataset)
def save_model(self, model_filename):
with open(model_filename, "wb") as file:
self.device = torch.device('cpu')
self.network.to(self.device)
print('Transfered forcibly model to CPU')
pickle.dump(self, file)
@staticmethod
def load_model(model_file):
# static so can be called as Model.load_model('examplemodel.model')
with open(model_file, "rb") as file:
# model = pickle.load(file)
model = CPU_Unpickler(file).load()
model.init_device()
model.network.to(model.device)
return model
def classify(self, x, pre_encoded=False):
if not pre_encoded:
x = self.word_encoder.encode_features_vector(x)
x = torch.tensor(x, dtype=torch.long)
x = x.to(self.device)
pred = self.network(x).argmax(1).item()
decoded_pred = self.word_encoder.decode_label(pred)
return decoded_pred