NeuralNet_Classifier/model.py at master · AdamJedz/NeuralNet_Classifier · GitHub

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
from network import Network
import numpy as np
import torch
import torch.nn as nn
import time
import matplotlib.pyplot as plt

from load_data import download_data, create_loaders


class Model:

    def __init__(self, name):
        self.name = name

        torch.manual_seed(0)
        self.model = Network().cuda()
        print(self.model)

    def num_params(self):

        model = self.model
        num_el = 0

        for item in model.parameters():
            num_el += item.numel()
        print(f'Number of elements of model {self.name} is {num_el}')

    def train_model(self, train_loader, test_loader, epochs=50, early_stopping_rounds=10, plot=True):

        def save_model(model, name):
            torch.save(model.state_dict(), f'./Model/{name}_best.pt')

        def print_stats(epoch, train_loss, train_acc, test_loss, test_acc, st_time):
            print(f'Epoch: {epoch+1} Loss: {train_loss.item():10.3f} Accuracy: {train_acc.item()*100/50000:7.2f}% \
            Val-loss: {test_loss.item():10.3f} Val-accuracy: {test_acc.item()*100/10000:7.2f}% Epoch time: {time.time() - st_time:.2f} seconds.')

        def plot_loss(train_loss, test_loss):
            plt.figure(figsize=(20,4))
            plt.plot(train_loss, label='Train Loss')
            plt.plot(test_loss, label='Test Loss')
            plt.legend()
            plt.title('Model Loss')
            plt.xlabel('Epoch')
            plt.ylabel('Crossentropy Loss')
            plt.show()

        def plot_acc(train_acc, test_acc):
            plt.figure(figsize=(20,4))
            plt.plot([t*100/50000 for t in train_acc], label='Train Accuracy')
            plt.plot([t*100/10000 for t in test_acc], label='Test Accuracy')
            plt.legend()
            plt.title('Model Accuracy')
            plt.xlabel('Epoch')
            plt.ylabel('Accuracy')
            plt.show()

        model = self.model

        criterion = nn.CrossEntropyLoss()
        optimizer = torch.optim.Adam(model.parameters(), lr=.001)

        start_time = time.time()
        epochs = epochs
        train_losses = []
        test_losses = []
        train_correct = []
        test_correct = []
        best_accuracy = 0
        early_stopping = 0

        for e in range(epochs):
            epoch_time = time.time()
            trn_cor = 0
            tst_cor = 0
            batch_loss = 0
            batch_tst_loss = 0

            model.train()

            for b, (X_train, y_train) in enumerate(train_loader):
                X_train = X_train.cuda()
                y_train = y_train.cuda()

                y_pred = model(X_train)
                loss = criterion(y_pred, y_train)

                predicted = torch.max(y_pred.data, 1)[1]
                optimizer.zero_grad()
                loss.backward()
                optimizer.step()

                batch_cor = (predicted == y_train).sum()
                trn_cor += batch_cor
                batch_loss += loss

            batch_loss = batch_loss / (50000/10)
            train_losses.append(batch_loss)
            train_correct.append(trn_cor)

            model.eval()
            with torch.no_grad():
                for b, (X_test, y_test) in enumerate(test_loader):

                    X_test = X_test.cuda()
                    y_test = y_test.cuda()

                    y_val = model(X_test)
                    predicted = torch.max(y_val.data, 1)[1]
                    batch_cor = (predicted == y_test).sum()
                    tst_cor += batch_cor

                    tst_loss = criterion(y_val, y_test)
                    batch_tst_loss += tst_loss

                batch_tst_loss = batch_tst_loss / (10000/10)
                test_losses.append(batch_tst_loss)
                test_correct.append(tst_cor)

            test_accuracy = tst_cor.item()*100/10000
            print_stats(e, batch_loss, trn_cor, batch_tst_loss, tst_cor, epoch_time)

            if test_accuracy > best_accuracy:
                save_model(model, self.name)
                best_accuracy = test_accuracy
                early_stopping = 0
            else:
                early_stopping += 1

            if early_stopping >= early_stopping_rounds:
                print(f'Early stopping after {e+1} epochs.')
                break

        print(f'Duration of training: {(time.time() - start_time)/60:.2f} minutes.')

        if plot:
            plot_loss(train_losses, test_losses)
            plot_acc(train_correct, test_correct)

    def load_model(self):
        self.model.load_state_dict(torch.load(f'./Model/{self.name}_best.pt'))


if __name__ == "__main__":
    model = Model('test')
    model.num_params()

    train_data, test_data = download_data()
    train_loader, test_loader = create_loaders(train_data, test_data)
    model.train_model(train_loader, test_loader)