《昇思25天学习打卡营第02天|快速入门》

快速入门

1. 准备MindSpore环境
   • 安装环境

      !pip uninstall mindspore -y
      !pip install -i https://pypi.mirrors.ustc.edu.cn/simple mindspore==2.2.14
   

   • 环境导入

   import mindspore
   from mindspore import nn #神经网络模块
   from mindspore.dataset import vision, transforms #图像模块(vision)&数据增强+预处理(transforms)
   from mindspore.dataset import MnistDataset #数据集模块
   

2. 数据集处理

   #数据集为开放数据集，下载完成后解压
   from download import download
   url = "https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/" \
         "notebook/datasets/MNIST_Data.zip"
   path = download(url, "./", kind="zip", replace=True)
   MNIST数据集目录结构如下：
   	MNIST_Data
   	└── train
   	    ├── train-images-idx3-ubyte (60000个训练图片)
   	    ├── train-labels-idx1-ubyte (60000个训练标签)
   	└── test
   	    ├── t10k-images-idx3-ubyte (10000个测试图片)
   	    ├── t10k-labels-idx1-ubyte (10000个测试标签)
   	数据下载完成后，获得数据集对象。
   
   #获取训练&测试数据集
   train_dataset = MnistDataset('MNIST_Data/train')
   test_dataset = MnistDataset('MNIST_Data/test')
   
   #数据集预处理
   def datapipe(dataset, batch_size):
   	#定义图像转换操作
       image_transforms = [
       	#图片缩放至范围(0，1)
           vision.Rescale(1.0 / 255.0, 0), 
           #将图像数据的分布调整到均值为0，标准差为1的分布
           vision.Normalize(mean=(0.1307,), std=(0.3081,)), 
         	#HWC(Height, Width, Channel) -> (Channel, Height, Width)
           vision.HWC2CHW()
       ]
       #将标签的数据类型转化成int32，便于模型进行处理，目的是确保计算过程中的精度&性能
       label_transform = transforms.TypeCast(mindspore.int32)
   	
   	#对图像&标签进行数据转化
       dataset = dataset.map(image_transforms, 'image')
       dataset = dataset.map(label_transform, 'label')
       
       #分batch，便于后续训练
       dataset = dataset.batch(batch_size)
       return dataset
   

3. 网络模型构建

   #定义模型
   class Network(nn.Cell):
      def __init__(self):
          super().__init__()
          self.flatten = nn.Flatten()
          self.dense_relu_sequential = nn.SequentialCell(
              nn.Dense(28*28, 512),
              nn.ReLU(),
              nn.Dense(512, 512),
              nn.ReLU(),
              nn.Dense(512, 10)
          )
   
      def construct(self, x):
          x = self.flatten(x)
          logits = self.dense_relu_sequential(x)
          return logits
   
   model = Network()
   

4. 模型训练

   #定义损失函数&优化器
   loss_fn = nn.CrossEntropyLoss()
   optimizer = nn.SGD(model.trainable_params(), 1e-2)
   
   #正向计算
   def forward_fn(data, label):
      logits = model(data)
      loss = loss_fn(logits, label)
      return loss, logits
   
   #反向传播
   grad_fn = mindspore.value_and_grad(forward_fn, None, optimizer.parameters, has_aux=True)
   
   #单步训练
   def train_step(data, label):
      (loss, _), grads = grad_fn(data, label)
      optimizer(grads)
      return loss
      
   #训练过程
   def train(model, dataset):
      size = dataset.get_dataset_size()
      model.set_train()
      for batch, (data, label) in enumerate(dataset.create_tuple_iterator()):
          loss = train_step(data, label)
   
          if batch % 100 == 0:
              loss, current = loss.asnumpy(), batch
              print(f"loss: {loss:>7f}  [{current:>3d}/{size:>3d}]")
   

5. 测试函数(模型评估)

   def test(model, dataset, loss_fn):
       num_batches = dataset.get_dataset_size()
       model.set_train(False)
       total, test_loss, correct = 0, 0, 0
       for data, label in dataset.create_tuple_iterator():
           pred = model(data)
           total += len(data)
           test_loss += loss_fn(pred, label).asnumpy()
           correct += (pred.argmax(1) == label).asnumpy().sum()
       test_loss /= num_batches
       correct /= total
       print(f"Test: \n Accuracy: {(100*correct):>0.1f}%, Avg loss: {test_loss:>8f} \n")
   

6. 模型保存&加载&推理

   #保存训练完后模型
   mindspore.save_checkpoint(model, "model.ckpt")
   
   #实例模型
   model = Network()
   #加载模型
   param_dict = mindspore.load_checkpoint("model.ckpt")
   param_not_load, _ = mindspore.load_param_into_net(model, param_dict)
   #推理
   model.set_train(False)
   for data, label in test_dataset:
       pred = model(data)
       predicted = pred.argmax(1)
       print(f'Predicted: "{predicted[:10]}", Actual: "{label[:10]}"')
       break