VennData/Utils/NetworkFunctions.py at master · IQTLabs/VennData · 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
154
155
156
157
158
159
import torch
import torchvision

import torch.nn as nn
import torch.nn.functional as F

import numpy as np

from torch.utils.data import TensorDataset, DataLoader
from torch import optim

import math

import os

def getAndSetDevice():

    # Function: getAndSetDevice
    # Inputs:   none
    # Process: creates pytorch device
    # Output:   device (pytorch device)

    device = torch.device("cuda:1")

    torch.cuda.set_device(device)

    return device

def getDetector(save_path):

    # Function: getDetector
    # Inputs:   none
    # Process: creates pytorch model for detection
    # Output:   detector (pytorch model)

    if os.path.exists(save_path):
        detector.load_state_dict(torch.load(save_path))
        detector.train()
    else:
        detector = math.nan

    return detector

def loadCifar10Data(batch_size):

    # Function: loadData
    # Inputs:   none
    # Process: returns Cifar10 dataset
    # Output:   training_loader (pytorch data loader)
    #           testing_loader (pytorch data loader)
    #           len(training_labels) (int)
    #           len(testing_labels) (int)

    training_images, training_labels = tfds.as_numpy(tfds.load(
                                            'cifar10',
                                            split='train',
                                            batch_size=-1,
                                            as_supervised=True,
                                        ))

    testing_images, testing_labels = tfds.as_numpy(tfds.load(
                                            'cifar10',
                                            split='test',
                                            batch_size=-1,
                                            as_supervised=True,
                                        ))

    training_images_pytorch = torch.Tensor(training_images).transpose(1,3)
    training_labels_pytorch = torch.Tensor(training_labels).type(torch.LongTensor)

    testing_images_pytorch = torch.Tensor(testing_images).transpose(1,3)
    testing_labels_pytorch = torch.Tensor(testing_labels).type(torch.LongTensor)

    training_dataset = TensorDataset(training_images_pytorch,training_labels_pytorch)
    testing_dataset  = TensorDataset(testing_images_pytorch,testing_labels_pytorch)

    training_loader = DataLoader(training_dataset, batch_size=batch_size, shuffle=True, pin_memory=True,drop_last=True)
    testing_loader  = DataLoader(testing_dataset, batch_size=batch_size, shuffle=True, pin_memory=True,drop_last=True)

    return training_loader, testing_loader, len(training_labels), len(testing_labels)

def trainNetwork(model,device,training_loader,testing_loader,num_training_examples,num_testing_examples,numIter,save_path):

    # Function: trainNetwork
    # Inputs:   model (pytorch model)
    #           device (pytorch device)
    #           training_loader (pytorch data loader)
    #           testing_loader (pytorch data loader)
    #           num_training_examples (int)
    #           num_testing_examples (int)
    #           numIter (int)
    #           save_path (file path for saving)
    # Process: trains pytorch model, saves model every 10 epochs
    # Output:   none

    model.to(device)

    loss_function = nn.CrossEntropyLoss()
    optimizer     = optim.Adam(model.parameters(),lr=0.0001)

    loss_array     = np.zeros([numIter,2])
    accuracy_array = np.zeros([numIter,2])

    for epoch in range(numIter):

        training_loss_array = []
        testing_loss_array  = []

        training_correct = 0
        testing_correct  = 0

        model.train()

        for image_batch, label_batch in training_loader:

            optimizer.zero_grad()

            output            = model(image_batch.to(device))

            training_loss = loss_function(output,label_batch.to(device))

            training_loss.backward()
            optimizer.step()

            training_loss_array.append(training_loss.cpu().item())

            _, predicted      = torch.max(output.data, 1)
            training_correct += (predicted.cpu() == label_batch).sum().item()

        model.eval()

        for image_batch, label_batch in testing_loader:

            output            = model(image_batch.to(device))

            testing_loss = loss_function(model(image_batch.to(device)),label_batch.to(device))

            testing_loss_array.append(testing_loss.cpu().item())

            _, predicted     = torch.max(output.data, 1)
            testing_correct += (predicted.cpu() == label_batch).sum().item()

        print('Epoch: ' + str(epoch))
        print()
        print("Training Loss: " + '\t\t' + str(np.mean(training_loss_array)))
        print("Testing Loss: " + '\t\t' + str(np.mean(testing_loss_array)))
        print("Training Accuracy: " + '\t' + str(100.0*training_correct/num_training_examples) + '%')
        print("Testing Accuracy: " + '\t' + str(100.0*testing_correct/num_testing_examples) + '%')

        loss_array[epoch,0] = np.mean(training_loss_array)
        loss_array[epoch,1] = np.mean(testing_loss_array)

        accuracy_array[epoch,0] = np.mean(100.0*training_correct/num_training_examples)
        accuracy_array[epoch,1] = np.mean(100.0*testing_correct/num_testing_examples)

        print()

    if epoch % 10 == 0:
        torch.save(detector.state_dict(),save_path)