[MXNet逐梦之旅]练习三·使用MXNetFashionMNIST数据集分类手动实现

#%%
import sys
import time
from mxnet import gluon as gl
import mxnet as mx
from matplotlib import pyplot as plt
from mxnet import autograd, nd
import numpy as np

mnist_train = gl.data.vision.FashionMNIST(root="fashion-mnist/",train=True)
mnist_test = gl.data.vision.FashionMNIST(root="fashion-mnist/",train=False)
print(len(mnist_train), len(mnist_test))

"""我也遇到这个问题，想用之前下载好的数据集直接使用。但是由于MXNet的MNIST数据集读取的机制会去验证数据集的正确性，所以我使用的用keras下载的数据集放到相关位置，MXNet还是还是会去下载。
解决方法，我用MXNet下载了好了fashion-mnist数据集，上传了，大家可以直接下载到相应位置解压即可。
也可以从采用自定义位置的方式 ：_mnist_train = gl.data.vision.FashionMNIST(root="fashion-mnist/",train=True)_
只需放到编写代码处解压即可。。"""

#%%

batch_size = 100
transformer = gl.data.vision.transforms.ToTensor()
if sys.platform.startswith('win'):
    num_workers = 0  # 0表示不用额外的进程来加速读取数据
else:
    num_workers = 4


train_iter = gl.data.DataLoader(mnist_train.transform_first(transformer),
                              batch_size, shuffle=True,
                              num_workers=num_workers)
test_iter = gl.data.DataLoader(mnist_test.transform_first(transformer),
                             1000, shuffle=False,
                             num_workers=num_workers)



def get_fashion_mnist_labels(labels):
    text_labels = ['t-shirt', 'trouser', 'pullover', 'dress', 'coat',
                   'sandal', 'shirt', 'sneaker', 'bag', 'ankle boot']
    return([text_labels[int(i.asnumpy())] for i in labels])

def show_img(x,y):
    plt.figure("Image") # 图像窗口名称
    plt.imshow(x.asnumpy()*255)
    plt.title(get_fashion_mnist_labels(y)) # 图像题目
    plt.show()


"""start = time.time()
for X, y in train_iter:
    break
print('%.2f sec' % (time.time() - start))"""

num_inputs, num_outputs, num_hiddens = 784, 10, 256

W1 = nd.random.normal(scale=0.01, shape=(num_inputs, num_hiddens))
b1 = nd.zeros(num_hiddens)
W2 = nd.random.normal(scale=0.01, shape=(num_hiddens, num_outputs))
b2 = nd.zeros(num_outputs)
params = [W1, b1, W2, b2]

for param in params:
    param.attach_grad()

def sgd(params, lr, batch_size):  # 本函数已保存在d2lzh包中方便以后使用
    for param in params:
        param[:] = param - lr * param.grad / batch_size


def relu(X):
    return nd.maximum(X, 0)

def model(X):
    X = X.reshape((-1, num_inputs))
    H = relu(nd.dot(X, W1) + b1)
    return nd.dot(H, W2) + b2

loss = gl.loss.SoftmaxCrossEntropyLoss()

def accuracy(y_hat, y):
    return (y_hat.argmax(axis=1) == y.astype('float32')).mean().asscalar()

lr = 0.1
num_epochs = 10
for epoch in range(1, num_epochs + 1):
    losses = []
    for X, y in train_iter:
        with autograd.record():
            l = loss(model(X), y)
            losses.append(l.asnumpy())
        l.backward()
        sgd(params,lr,batch_size)

    loss_sum = np.mean(losses)
    for Xt, yt in test_iter:
        accy = accuracy(model(Xt),yt)
        break

    print('epoch %d, loss: %f, accy: %f' % (epoch,loss_sum,accy))

60000 10000
epoch 1, loss: 0.776444, accy: 0.827000
epoch 2, loss: 0.490361, accy: 0.855000
epoch 3, loss: 0.433773, accy: 0.855000
epoch 4, loss: 0.404391, accy: 0.833000
epoch 5, loss: 0.380910, accy: 0.856000
epoch 6, loss: 0.363898, accy: 0.873000
epoch 7, loss: 0.351115, accy: 0.873000
epoch 8, loss: 0.340445, accy: 0.875000
epoch 9, loss: 0.328219, accy: 0.876000
epoch 10, loss: 0.318909, accy: 0.872000