在模型训练过程中，一般分为四个步骤。·定义神经网络。·构建数据集。·定义超参、损失函数及优化器。·输入训练轮次和数据集进行训练。
工具/原料
MindSpore1.2
Windows10
方法/步骤
1、导入需要的模块并传入数据集importmindspore.datasetasdsimportmindspore.dataset.transforms.c_transformsasCimportmindspore.dataset.vision.c_transformsasCVfrommindsporeimportnn,Tensor,ModelfrommindsporeimportdtypeasmstypeDATA_DIR="./datasets/cifar-10-batches-bin/train"
2、定义神经网络classNet(nn.Cell):蟠校盯昂def__init__(self,num_cla霜杼厮贿ss=10,num_channel=3):super(Net,self).__init__()self.conv1=nn.Conv2d(num_channel,6,5,pad_mode='valid')self.conv2=nn.Conv2d(6,16,5,pad_mode='valid')self.fc1=nn.Dense(16*5*5,120)self.fc2=nn.Dense(120,84)self.fc3=nn.Dense(84,num_class)self.relu=nn.ReLU()self.max_pool2d=nn.MaxPool2d(kernel_size=2,stride=2)self.flatten=nn.Flatten()defconstruct(self,x):x=self.conv1(x)x=self.relu(x)x=self.max_pool2d(x)x=self.conv2(x)x=self.relu(x)x=self.max_pool2d(x)x=self.flatten(x)x=self.fc1(x)x=self.relu(x)x=self.fc2(x)x=self.relu(x)x=self.fc3(x)returnxnet=Net()epochs=5batch_size=64learning_rate=1e-3
3、构建数据集衡痕贤伎sampler=ds.SequentialSampler(num_samples=128)dataset=ds.Cif锾攒揉敫ar10Dataset(DATA_DIR,sampler=sampler)数据类型转换type_cast_op_image=C.TypeCast(mstype.float32)type_cast_op_label=C.TypeCast(mstype.int32)HWC2CHW=CV.HWC2CHW()dataset=dataset.map(operations=[type_cast_op_image,HWC2CHW],input_columns="image")dataset=dataset.map(operations=type_cast_op_label,input_columns="label")dataset=dataset.batch(batch_size)
4、定义超参、损失函数及优化器optim=nn.SGD(params=net.trainable_params(),learning_rate=learning_rate)loss=nn.SoftmaxCrossEntropyWithLogits(sparse=True,reduction='mean')
5、输入训练轮次和数据集进行训练model=Model(net,loss_fn=loss,optimizer=optim)model.train(epoch=epochs,train_dataset=dataset)