ANVAE/train.py at master · oadonca/ANVAE · 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
# -*- coding: utf-8 -*-
"""
Created on Wed Oct  7 18:17:17 2020

@author: Octavian
"""
import tensorflow as tf
import numpy as np
from dataset import dataset
import anvae2
import matplotlib.pyplot as plt
import pandas as pd
from IPython import display

RUN_NAME = "1"
TRAIN_PATH = './Data/mnist_train.csv'
TEST_PATH = './Data/mnist_test.csv'
LATENT_SPACES = 3
TRAIN_BUF=60000
BATCH_SIZE=16
TEST_BUF=10000
DIMS = (32,32,1)
N_TRAIN_BATCHES =int(TRAIN_BUF/BATCH_SIZE)
N_TEST_BATCHES = int(TEST_BUF/BATCH_SIZE)
N_EPOCHS = 200

def plot_losses(losses, epoch=0):
    fig, axs =plt.subplots(ncols = 4, nrows = 1, figsize= (16,4))
    axs[0].plot(losses.encoder_loss.values, label = 'encoder_loss')
    axs[0].plot(losses.disc_loss1.values, label = 'disc_loss1')
    axs[0].plot(losses.disc_loss2.values, label = 'disc_loss2')
    axs[1].plot(losses.disc_loss3.values, label = 'disc_loss3')
    axs[2].plot(losses.disc_acc1.values, label = 'disc_acc1')
    axs[2].plot(losses.disc_acc2.values, label = 'disc_acc2')
    axs[2].plot(losses.disc_acc3.values, label = 'disc_acc3')
    axs[3].plot(losses.gen_loss.values, label = 'gen_loss')

    for ax in axs.flatten():
        ax.legend()
    plt.savefig('Output/plots/losses_plot_{}.png'.format(epoch))

def plot_reconstruction(model, data, epoch, step):
    _, _, out = model.encode(data, False)
    recon = model.decode(out)
    fig, axs = plt.subplots(ncols=2, nrows=1, figsize=(8,4))
    for axi, (dat, lab) in enumerate(
        zip(
            [data, recon],
            ["data", "data recon"],
        )
    ):
        for ex in range(1):

            if len(dat[ex].shape) != 3:
                axs[axi].matshow(
                    np.squeeze(dat[ex][0]), cmap=plt.cm.Greys, vmin=0, vmax=1
                )
            else:
                axs[axi].matshow(
                    np.squeeze(dat[ex]), cmap=plt.cm.Greys, vmin=0, vmax=1
                )
            axs[axi].axes.get_xaxis().set_ticks([])
            axs[axi].axes.get_yaxis().set_ticks([])
        axs[axi].set_ylabel(lab)

    plt.savefig('Output/images/image_{}_{}.png'.format(epoch, step))


train_data = dataset(data_path = TRAIN_PATH, batch_size=BATCH_SIZE)

train_model = anvae2.ANVAE(LATENT_SPACES, BATCH_SIZE)

losses = pd.DataFrame(columns=[
    'encoder_loss',
    'disc_loss1',
    'disc_loss2',
    'disc_loss3',
    'disc_acc1',
    'disc_acc2',
    'disc_acc3',
    'gen_loss',
])

for epoch in range(N_EPOCHS):
    loss = []
    iteration = 0
    ex_image = None
    epoch_completed = False
    while epoch_completed == False:
        debug = False
        if (iteration % 100 == 0):
            debug = True
        image_batch, label_batch, epoch_completed = train_data.next_batch()
        if iteration == 0:
            ex_image=image_batch
        enc_loss, dc_loss, dc_acc, gen_loss = train_model.train_step(image_batch, label_batch)

        temp2 = []

        temp2.append(enc_loss)
        for l in dc_loss:
            temp2.append(l)

        for acc in dc_acc:
            temp2.append(acc)

        temp2.append(gen_loss)

        loss.append(temp2)

        if (iteration % 100 == 0):
            for i, (l, label) in enumerate(zip(temp2, losses.columns)):
                print("{}: {}".format(label, l))
            print("Iteration: {}".format(iteration))
            print('x'*80)

        if (epoch_completed == True):
            break

        iteration += 1
        print(iteration)

        if (train_model.step_count % 200 == 0):
            print('1'*80)

            losses.loc[len(losses)] = np.mean(loss)

            print("Epoch: {}".format(epoch))

            display.clear_output()

            print('2'*80)

            plot_losses(losses, epoch)

            print('3'*80)

            plot_reconstruction(train_model, ex_image, epoch, train_model.step_count)

            print('4'*80)