第六次作业

import os
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"

import torch

import torch.nn as nn

import numpy as np

import matplotlib.pyplot as plt

# torch.manual_seed(1)    # reproducible

# np.random.seed(1)

# Hyper Parameters

BATCH_SIZE = 64

LR_G = 0.0001  # learning rate for generator

LR_D = 0.0001  # learning rate for discriminator

N_IDEAS = 5  # think of this as number of ideas for generating an art work (Generator)

ART_COMPONENTS = 15  # it could be total point G can draw in the canvas

PAINT_POINTS = np.vstack([np.linspace(-1, 1, ART_COMPONENTS) for _ in range(BATCH_SIZE)])

# show our beautiful painting range

# plt.plot(PAINT_POINTS[0], 2 * np.power(PAINT_POINTS[0], 2) + 1, c='#74BCFF', lw=3, label='upper bound')

# plt.plot(PAINT_POINTS[0], 1 * np.power(PAINT_POINTS[0], 2) + 0, c='#FF9359', lw=3, label='lower bound')

# plt.legend(loc='upper right')

# plt.show()

def artist_works():  # painting from the famous artist (real target)

    a = np.random.uniform(1, 2, size=BATCH_SIZE)[:, np.newaxis]

    paintings = a * np.power(PAINT_POINTS, 2) + (a - 1)

    paintings = torch.from_numpy(paintings).float()

    return paintings

G = nn.Sequential(  # Generator

    nn.Linear(N_IDEAS, 128),  # random ideas (could from normal distribution)

    nn.ReLU(),

    nn.Linear(128, ART_COMPONENTS),  # making a painting from these random ideas

)

D = nn.Sequential(  # Discriminator

    nn.Linear(ART_COMPONENTS, 128),  # receive art work either from the famous artist or a newbie like G

    nn.ReLU(),

    nn.Linear(128, 1),

    nn.Sigmoid(),  # tell the probability that the art work is made by artist

)

opt_D = torch.optim.Adam(D.parameters(), lr=LR_D)

opt_G = torch.optim.Adam(G.parameters(), lr=LR_G)

plt.ion()  # something about continuous plotting

for step in range(10000):

    artist_paintings = artist_works()  # real painting from artist

    G_ideas = torch.randn(BATCH_SIZE, N_IDEAS, requires_grad=True)  # random ideas\n

    G_paintings = G(G_ideas)  # fake painting from G (random ideas)

    prob_artist1 = D(G_paintings)  # D try to reduce this prob

    G_loss = torch.mean(torch.log(1. - prob_artist1))

    opt_G.zero_grad()

    G_loss.backward()

    opt_G.step()

    prob_artist0 = D(artist_paintings)  # D try to increase this prob

    prob_artist1 = D(G_paintings.detach())  # D try to reduce this prob

    D_loss = - torch.mean(torch.log(prob_artist0) + torch.log(1. - prob_artist1))

    opt_D.zero_grad()

    D_loss.backward(retain_graph=True)  # reusing computational graph

    opt_D.step()

    if step % 50 == 0:  # plotting

        plt.cla()

        plt.plot(PAINT_POINTS[0], G_paintings.data.numpy()[0], c='#4AD631', lw=3, label='Generated painting', )

        plt.plot(PAINT_POINTS[0], 2 * np.power(PAINT_POINTS[0], 2) + 1, c='#74BCFF', lw=3, label='upper bound')

        plt.plot(PAINT_POINTS[0], 1 * np.power(PAINT_POINTS[0], 2) + 0, c='#FF9359', lw=3, label='lower bound')

        plt.text(-.5, 2.3, 'D accuracy=%.2f (0.5 for D to converge)' % prob_artist0.data.numpy().mean(),
                 fontdict={'size': 13})

        plt.text(-.5, 2, 'D score= %.2f (-1.38 for G to converge)' % -D_loss.data.numpy(), fontdict={'size': 13})

        plt.ylim((0, 3));
        plt.legend(loc='upper right', fontsize=10);
        plt.draw();
        plt.pause(0.01)

plt.ioff()

plt.show()