从零开始深度学习0613——pytorch入门之optimizer+CNN手写数字识别

------------------optimizer-----------------------------------------------

 

Python中Zip() 函数

 

>>>a = [1,2,3]

>>> b = [4,5,6]

>>> c = [4,5,6,7,8]

>>> zipped = zip(a,b) # 打包为元组的列表 [(1, 4), (2, 5), (3, 6)]

>>> zip(a,c) # 元素个数与最短的列表一致 [(1, 4), (2, 5), (3, 6)]

>>> zip(*zipped) # 与 zip 相反，*zipped 可理解为解压，返回二维矩阵式 [(1, 2, 3), (4, 5, 6)]

import torch
import torch.utils.data as Data
import torch.nn.functional as F
import matplotlib.pyplot as plt

# torch.manual_seed(1)    # reproducible

LR = 0.01
BATCH_SIZE = 32
EPOCH = 12

# fake dataset
x = torch.unsqueeze(torch.linspace(-1, 1, 1000), dim=1)
y = x.pow(2) + 0.1*torch.normal(torch.zeros(*x.size()))

# plot dataset
# plt.scatter(x.numpy(), y.numpy())
# plt.show()

# put dateset into torch dataset
torch_dataset = Data.TensorDataset(x, y)
loader = Data.DataLoader(dataset=torch_dataset, batch_size=BATCH_SIZE, shuffle=True, num_workers=2,)


# default network
class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.hidden = torch.nn.Linear(1, 20)   # hidden layer
        self.predict = torch.nn.Linear(20, 1)   # output layer

    def forward(self, x):
        x = F.relu(self.hidden(x))      # activation function for hidden layer
        x = self.predict(x)             # linear output
        return x

if __name__ == '__main__':
    # different nets
    net_SGD         = Net()
    net_Momentum    = Net()
    net_RMSprop     = Net()
    net_Adam        = Net()
    nets = [net_SGD, net_Momentum, net_RMSprop, net_Adam]

    # different optimizers
    opt_SGD         = torch.optim.SGD(net_SGD.parameters(), lr=LR)
    opt_Momentum    = torch.optim.SGD(net_Momentum.parameters(), lr=LR, momentum=0.8)
    opt_RMSprop     = torch.optim.RMSprop(net_RMSprop.parameters(), lr=LR, alpha=0.9)
    opt_Adam        = torch.optim.Adam(net_Adam.parameters(), lr=LR, betas=(0.9, 0.99))
    optimizers = [opt_SGD, opt_Momentum, opt_RMSprop, opt_Adam]

    loss_func = torch.nn.MSELoss()
    losses_his = [[], [], [], []]   # record loss

    # training
    for epoch in range(EPOCH):
        print('Epoch: ', epoch)
        for step, (b_x, b_y) in enumerate(loader):          # for each training step
            for net, opt, l_his in zip(nets, optimizers, losses_his):
                output = net(b_x)              # get output for every net
                loss = loss_func(output, b_y)  # compute loss for every net
                opt.zero_grad()                # clear gradients for next train
                loss.backward()                # backpropagation, compute gradients
                opt.step()                     # apply gradients
                l_his.append(loss.data.numpy())     # loss recoder

    labels = ['SGD', 'Momentum', 'RMSprop', 'Adam']
    for i, l_his in enumerate(losses_his):
        plt.plot(l_his, label=labels[i])
    plt.legend(loc='best')
    plt.xlabel('Steps')
    plt.ylabel('Loss')
    plt.ylim((0, 0.2))
    plt.show()

 

 

----------------------------CNN--------------------------------------------

################################################################

 

实现mnist 手写数字识别  及 预测

完整代码：

import os

os.environ["CUDA_VISIBLE_DEVICES"] = "1"  #设定GPU cuda运行 与下面选一即可



import torch



###########################检查pytorch是否正在使用cuda

# a = torch.cuda.is_available()

# print(a)

# ngpu= 1

# # Decide which device we want to run on

# device = torch.device("cuda:0" if (torch.cuda.is_available() and ngpu > 0) else "cpu")

# print(device)

# print(torch.cuda.get_device_name(0))

# print(torch.rand(3,3).cuda())

###################################################





# third-party library

#import torch

import torch.nn as nn

import torch.utils.data as Data

import torchvision

import matplotlib.pyplot as plt



# torch.manual_seed(1)    # reproducible



# Hyper Parameters

EPOCH = 1               # train the training data n times, to save time, we just train 1 epoch

BATCH_SIZE = 50

LR = 0.001              # learning rate

DOWNLOAD_MNIST = False





# Mnist digits dataset

if not(os.path.exists('./mnist/')) or not os.listdir('./mnist/'):

    # not mnist dir or mnist is empyt dir

    DOWNLOAD_MNIST = True



train_data = torchvision.datasets.MNIST(

    root='./mnist/',

    train=True,                                     # this is training data

    transform=torchvision.transforms.ToTensor(),    # Converts a PIL.Image or numpy.ndarray to

                                                    # torch.FloatTensor of shape (C x H x W) and normalize in the range [0.0, 1.0]

                                                    # 将图像的像素值（0,255）压缩为（0,1）区间

    download=DOWNLOAD_MNIST,

)



# plot one example

print(train_data.train_data.size())                 # (60000, 28, 28)

print(train_data.train_labels.size())               # (60000)

plt.imshow(train_data.train_data[0].numpy(), cmap='gray')  # 显示出第一张图片

plt.title('%i' % train_data.train_labels[0])

plt.show()



# Data Loader for easy mini-batch return in training, the image batch shape will be (50, 1, 28, 28)

train_loader = Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)



# pick 2000 samples to speed up testing

test_data = torchvision.datasets.MNIST(root='./mnist/', train=False)

test_x = torch.unsqueeze(test_data.test_data, dim=1).type(torch.FloatTensor)[:2000]/255. # 手动压缩   # shape from (2000, 28, 28) to (2000, 1, 28, 28), value in range(0,1)

test_y = test_data.test_labels[:2000]





class CNN(nn.Module):

    def __init__(self):

        super(CNN, self).__init__()

        self.conv1 = nn.Sequential(         # input shape (1, 28, 28)

            nn.Conv2d(

                in_channels=1,              # input height  灰度图只有一个通道

                out_channels=16,            # n_filters 卷积核的数量

                kernel_size=5,              # filter size  卷积核的大小

                stride=1,                   # filter movement/step

                padding=2,                  # if want same width and length of this image after Conv2d, padding=(kernel_size-1)/2 if stride=1

            ),                              # output shape (16, 28, 28)

            nn.ReLU(),                      # activation

            nn.MaxPool2d(kernel_size=2),    # choose max value in 2x2 area, output shape (16, 14, 14)  降维一半

        )

        self.conv2 = nn.Sequential(         # input shape (16, 14, 14)

            nn.Conv2d(16, 32, 5, 1, 2),     # output shape (32, 14, 14)

            nn.ReLU(),                      # activation

            nn.MaxPool2d(2),                # output shape (32, 7, 7)

        )

        self.out = nn.Linear(32 * 7 * 7, 10)   # fully connected layer, output 10 classes  全连接层，将上一步所有的feature map 铺平  将三维数据变为2维



    def forward(self, x):

        x = self.conv1(x)

        x = self.conv2(x)          # (batch,32, 7, 7)

        x = x.view(x.size(0), -1)   # 将batch维度保留，其他地方合并  flatten the output of conv2 to (batch_size, 32 * 7 * 7)

        output = self.out(x)

        return output, x    # return x for visualization





cnn = CNN()

print(cnn)  # net architecture





optimizer = torch.optim.Adam(cnn.parameters(), lr=LR)   # optimize all cnn parameters

loss_func = nn.CrossEntropyLoss()                       # the target label is not one-hotted



# following function (plot_with_labels) is for visualization, can be ignored if not interested

# from matplotlib import cm

# try: from sklearn.manifold import TSNE; HAS_SK = True

# except: HAS_SK = False; print('Please install sklearn for layer visualization')

# def plot_with_labels(lowDWeights, labels):

#     plt.cla()

#     X, Y = lowDWeights[:, 0], lowDWeights[:, 1]

#     for x, y, s in zip(X, Y, labels):

#         c = cm.rainbow(int(255 * s / 9)); plt.text(x, y, s, backgroundcolor=c, fontsize=9)

#     plt.xlim(X.min(), X.max()); plt.ylim(Y.min(), Y.max()); plt.title('Visualize last layer'); plt.show(); plt.pause(0.01)

#

# plt.ion()

# training and testing

for epoch in range(EPOCH):

    for step, (b_x, b_y) in enumerate(train_loader):   # gives batch data, normalize x when iterate train_loader



        output = cnn(b_x)[0]               # cnn output

        loss = loss_func(output, b_y)   # cross entropy loss

        optimizer.zero_grad()           # clear gradients for this training step

        loss.backward()                 # backpropagation, compute gradients

        optimizer.step()                # apply gradients



        if step % 50 == 0:

            test_output, last_layer = cnn(test_x)

            pred_y = torch.max(test_output, 1)[1].data.numpy()

            accuracy = float((pred_y == test_y.data.numpy()).astype(int).sum()) / float(test_y.size(0))

            print('Epoch: ', epoch, '| train loss: %.4f' % loss.data.numpy(), '| test accuracy: %.2f' % accuracy)

#             if HAS_SK:

#                 # Visualization of trained flatten layer (T-SNE)

#                 tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000)

#                 plot_only = 500

#                 low_dim_embs = tsne.fit_transform(last_layer.data.numpy()[:plot_only, :])

#                 labels = test_y.numpy()[:plot_only]

#                 plot_with_labels(low_dim_embs, labels)

# plt.ioff()



# print 10 predictions from test data

test_output, _ = cnn(test_x[:10])

pred_y = torch.max(test_output, 1)[1].data.numpy()

print(pred_y, 'prediction number')

print(test_y[:10].numpy(), 'real number')

 

pytorch 设置GPU 的三种方式

# 1：

torch.cuda.set_device(1)

 

# 2：

device = torch.device("cuda:1")

 

# 3：（官方推荐）<br>import os

os.environ["CUDA_VISIBLE_DEVICES"] = '1'

 

（同时调用两块GPU的话）

os.environ["CUDA_VISIBLE_DEVICES"] = '1,2'