[MXNet]Lecture02卷积神经网络的从零开始的实现

import mxnet as mx
from mxnet import ndarray as nd
from mxnet import autograd
from mxnet import gluon

batch_size=256
ctx=mx.gpu()#将数组开在gpu上

def transform(data,label):
	return data.astype('float32')/255, label.astype('float32')#不是float_32哦

train_mnist=gluon.data.vision.FashionMNIST(train=True,transform=transform)
test_mnist=gluon.data.vision.FashionMNIST(train=False,transform=transform)

train_data=gluon.data.DataLoader(train_mnist,batch_size,shuffle=True)
test_data=gluon.data.DataLoader(test_mnist,batch_size,shuffle=False)

weight_scale=0.01
w1=nd.random_normal(shape=(20,1,5,5),scale=weight_scale,ctx=ctx)
b1=nd.zeros(w1.shape[0],ctx=ctx)

w2=nd.random_normal(shape=(50,20,3,3),scale=weight_scale,ctx=ctx)
b2=nd.zeros(w2.shape[0],ctx=ctx)

w3=nd.random_normal(shape=(1250,128),scale=weight_scale,ctx=ctx)#1250这个数据需要print一下h2的输出大小。
b3=nd.zeros(w3.shape[1],ctx=ctx)

w4=nd.random_normal(shape=(w3.shape[1],10),scale=weight_scale,ctx=ctx)
b4=nd.zeros(w4.shape[1],ctx=ctx)
#卷积层的权重和全联接层的权重，输入输出是反着的。
params=[w1,b1,w2,b2,w3,b3,w4,b4]
for param in params:
	param.attach_grad()

def net(X):
	X=X.as_in_context(w1.context)
	h1_conv=nd.Convolution(data=X.reshape((-1,1,28,28)),weight=w1,bias=b1,kernel=w1.shape[2:],num_filter=w1.shape[0])#输入的data 需要reshape一下
	h1_activation=nd.relu(h1_conv)
	h1=nd.Pooling(data=h1_activation,pool_type='max',kernel=(2,2),stride=(2,2))

	h2_conv=nd.Convolution(data=h1,weight=w2,bias=b2,kernel=w2.shape[2:],num_filter=w2.shape[0])
	h2_activation=nd.relu(h2_conv)
	h2=nd.Pooling(data=h2_activation,pool_type='max',kernel=(2,2),stride=(2,2))
	h2=nd.Flatten(h2)
	
	h3_linear=nd.dot(h2,w3)+b3
	h3=nd.relu(h3_linear)
	h4=nd.dot(h3,w4)+b4
	return h4

softmax_crossentropy=gluon.loss.SoftmaxCrossEntropyLoss()

def sgd(params,lr):
	for param in params:
		param[:]=param-lr*param.grad

def accuracy(output,label):
	return nd.mean(output.argmax(axis=1)==label).asscalar()

def evaluate_accuracy(X):
	acc=0.0
	for data,label in X:
		label=label.as_in_context(ctx)
		output=net(data)
		acc+=accuracy(output,label)
	return acc/len(X)		

learning_rate=0.2
for epoch in range(10):
	train_loss=0.0
	train_acc=0.0
	for data,label in train_data:
		with autograd.record():
			output=net(data)
			label=label.as_in_context(ctx)#这里不可以用data.context因为data并未被更改到gpu上
			loss=softmax_crossentropy(output,label)	
		loss.backward()
		sgd(params,learning_rate/batch_size)
		train_loss+=nd.mean(loss).asscalar()
		train_acc+=accuracy(output,label)
	test_acc=evaluate_accuracy(test_data)
	print("epoch: %d, train_loss:%f, train_acc:%f, test_acc:%f"%(epoch,train_loss/len(train_data),train_acc/len(train_data),test_acc))