-
Notifications
You must be signed in to change notification settings - Fork 0
/
2_train_main_task_model.py
180 lines (150 loc) · 6.14 KB
/
2_train_main_task_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
from statistics import mode
from numpy import argmax
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.utils import shuffle
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
import torch.nn.functional as F
from torch.optim import SGD
from opacus import PrivacyEngine
import data
from utils import fix_randomness
from models import Covid19_MainTaskModel, Adults_MainTaskModel, Fivethirtyeight_MainTaskModel, GSS_MainTaskModel
import yaml
import sys
fix_randomness()
# read config file
with open("./config/config.yaml", "r") as ymlfile:
try:
cfg = yaml.safe_load(ymlfile)
except yaml.YAMLError as exc:
print(exc)
# Step 1: load data from .csv file
problem = cfg["problem"]
datapath = cfg["dataset"][problem]["path_to_data"]
if problem == 'covid19':
X, y = data.covid19_load_data(datapath)
elif problem == 'adults':
X, y, scaler = data.adults_load_data(datapath)
elif problem == 'fivethirtyeight':
X, y, scaler = data.fivethirtyeight_load_data(datapath)
elif problem == 'gss':
X, y, scaler = data.gss_load_data(datapath)
else:
print("The problem was not supported!!!")
sys.exit()
# Step 2: shuffle data
sklearn_random_state = cfg["random"]["random_state_sklearn"]
X, y = shuffle(X, y, random_state=sklearn_random_state)
# Step 3: Take a part of data to train TARGET model
propotion_to_train_target = cfg["target_model"][problem]["propotion_to_train_target_model"]
X = X[:int(X.shape[0]*propotion_to_train_target),:]
y = y[:int(y.shape[0]*propotion_to_train_target)]
# Step 4: Split data into training set and test set (to train TARGET model)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=cfg["target_model"][problem]["test_size"], random_state=sklearn_random_state)
X_train = torch.from_numpy(X_train).type(torch.FloatTensor)
X_test = torch.from_numpy(X_test).type(torch.FloatTensor)
y_train = torch.from_numpy(y_train)
y_train = F.one_hot(y_train).type(torch.FloatTensor)
y_test = torch.from_numpy(y_test)
y_test = F.one_hot(y_test).type(torch.FloatTensor)
# parameters to train target model
BATCH_SIZE = cfg["target_model"][problem]["batch_size"]
EPOCHS = cfg["target_model"][problem]["epochs"]
LR = cfg["target_model"][problem]["learning_rate"]
MOMENTUM = cfg["target_model"][problem]["momentum"]
# use DataSet and DataLoader class in Pytorch
# NOTE: dont shuffle the data, we already shuffled them before
data_train = data.TabularData(X_train, y_train)
data_train_loader = DataLoader(data_train, batch_size=BATCH_SIZE, shuffle=False)
data_test = data.TabularData(X_test, y_test)
data_test_loader = DataLoader(data_test, batch_size=BATCH_SIZE, shuffle=False)
# Step 5: Define model to train
if problem == 'covid19':
model = Covid19_MainTaskModel()
elif problem == 'adults':
model = Adults_MainTaskModel()
elif problem == 'fivethirtyeight':
model = Fivethirtyeight_MainTaskModel()
elif problem == 'gss':
model = GSS_MainTaskModel()
else:
print("Can not define target model!!!")
sys.exit()
# Step 6: define optimizer used to train model
# optimizer = SGD(model.parameters(), lr=LR, momentum=MOMENTUM)
optimizer = SGD(model.parameters(), lr=LR)
# Step 7: define loss function
# lossFunc = nn.MSELoss()
lossFunc = nn.CrossEntropyLoss()
# Step 8: define training with DP
# Attaching a Differential Privacy Engine to the Optimizer
if cfg['target_model'][problem]['DP']['isDP'] == True:
privacy_engine = PrivacyEngine()
model, optimizer, data_loader = privacy_engine.make_private(
module=model,
optimizer=optimizer,
data_loader=data_train_loader,
noise_multiplier=1.1,
max_grad_norm=1.0,
)
delta=cfg['target_model'][problem]['DP']['delta']
for epoch in range(0, EPOCHS):
print("[INFO] epoch: {}...".format(epoch + 1))
# Step 1: Train on training set
trainLoss = 0
samples = 0
model.train()
train_preds = []
for (batchX, batchY) in data_train_loader:
outputs = model(batchX)
loss = lossFunc(outputs, batchY)
optimizer.zero_grad()
loss.backward()
optimizer.step()
trainLoss += loss.item() * batchY.size(0)
for output in outputs:
train_preds.append(argmax(output.detach().numpy()))
# if output[0] > output[1]:
# train_preds.append(0)
# else:
# train_preds.append(1)
samples += batchY.size(0)
y_true = argmax(y_train, axis=1)
trainAcc = accuracy_score(y_true, train_preds)
if cfg['target_model'][problem]['DP']['isDP'] == True:
print(delta)
print(type(delta))
epsilon = privacy_engine.get_epsilon(delta)
trainTemplate = "epoch: {} train loss: {:.3f} train accuracy: {:.3f} (ε = {:.2f}, δ = {})"
print(trainTemplate.format(epoch + 1, (trainLoss / samples), trainAcc, epsilon, delta))
# trainTemplate = "epoch: {} train loss: {:.3f} train accuracy: {:.3f}"
# print(trainTemplate.format(epoch + 1, (trainLoss / samples), trainAcc))
else:
trainTemplate = "epoch: {} train loss: {:.3f} train accuracy: {:.3f}"
print(trainTemplate.format(epoch + 1, (trainLoss / samples), trainAcc))
# Step 2: Evaluate on test set
testLoss = 0
samples = 0
model.eval()
test_preds = []
with torch.no_grad():
for (batchX, batchY) in data_test_loader:
outputs = model(batchX)
loss = lossFunc(outputs, batchY)
testLoss += loss.item() * batchY.size(0)
for output in outputs:
test_preds.append(argmax(output))
samples += batchY.size(0)
y_true = argmax(y_test, axis=1)
testAcc = accuracy_score(y_true, test_preds)
# display model progress on the current training batch
trainTemplate = "epoch: {} test loss: {:.3f} test accuracy: {:.3f}"
print(trainTemplate.format(epoch + 1, (testLoss / samples), testAcc))
# Save trained model
if cfg['target_model'][problem]['DP']['isDP'] == True:
torch.save(model, cfg["target_model"][problem]["path_to_target_model_with_DP"])
else:
torch.save(model, cfg["target_model"][problem]["path_to_target_model"])