《昇思25天学习打卡营第06天|网络构建》

网络构建

神经网络模型由神经网络层和Tensor操作构成

#实验环境已经预装了mindspore==2.2.14，如需更换mindspore版本，可更改下面mindspore的版本号
!pip uninstall mindspore -y
!pip install -i https://pypi.mirrors.ustc.edu.cn/simple mindspore==2.2.14

定义模型类

import mindspore
from mindspore import nn, ops

class Network(nn.Cell):
    def __init__(self):
        super().__init__()
        self.flatten = nn.Flatten()
        self.dense_relu_sequential = nn.SequentialCell(
            nn.Dense(28*28, 512, weight_init="normal", bias_init="zeros"),
            nn.ReLU(),
            nn.Dense(512, 512, weight_init="normal", bias_init="zeros"),
            nn.ReLU(),
            nn.Dense(512, 10, weight_init="normal", bias_init="zeros")
        )

    def construct(self, x):
        x = self.flatten(x)
        logits = self.dense_relu_sequential(x)
        return logits
        
model = Network() #打印模型结构
print(model)
#Output
	#Network<
	  #(flatten): Flatten<>
	  #(dense_relu_sequential): SequentialCell<
	    #(0): Dense<input_channels=784, output_channels=512, has_bias=True>
	    #(1): ReLU<>
	    #(2): Dense<input_channels=512, output_channels=512, has_bias=True>
	    #(3): ReLU<>
	    #(4): Dense<input_channels=512, output_channels=10, has_bias=True>
	    #>
	  #>
X = ops.ones((1, 28, 28), mindspore.float32)
logits = model(X)
# print logits
logits
#Output 
#Tensor(shape=[1, 10], dtype=Float32, value= [[-2.40761833e-03,  2.76332069e-03,  4.36006673e-03 ... -2.03372864e-03,  2.23693671e-04,  5.74092008e-03]])
# 过Softmax获得预测概率
pred_prob = nn.Softmax(axis=1)(logits)
y_pred = pred_prob.argmax(1) #给出预测结果
print(f"Predicted class: {y_pred}")

模型层

#构建3个28*28的图像
input_image = ops.ones((3,28,28), mindspore.float32)
print(input_image.shape)
#Output (3, 28, 28)

nn.Flatten

flatten = nn.Flatten()
flat_image = flatten(input_image)
print(flat_image.shape()
#Output(3, 728) 把图像碾平成一维 满足模型输入要求

nn.dense

layer1 = nn.Dense(in_channels = 28 * 28, out_channels = 20)
hidden1 = layer1(flat_image)
print(hidden1.shape)
#Output(3, 20)

nn.ReLu

#加入非线性激活函数，增加模型复杂度
hidden1 = nn.ReLU()(hidden1)

nn.Sequential

seq_modules = nn.SequentialCell(
    flatten,
    layer1,
    nn.ReLU(),
    nn.Dense(20, 10)
)

logits = seq_modules(input_image)
print(logits.shape)

nn.Softmax

#将神经网络最后一个全连接层返回的logits值缩放至[0,1], 表示每个类别的概率预测。axis指定的维度数值和为1
softmax = nn.Softmax(axis=1)
pred_probab = softmax(logits)

模型参数

通过model.parameters_and_names()获取

for name, param in model.parameters_and_names():
    print(f"Layer: {name}\nSize: {param.shape}\nValues : {param[:2]} \n")

#Layer: dense_relu_sequential.0.weight
#Size: (512, 784)
#Values : [[ 0.01388695 -0.00604919 -0.00993734 ...  0.00366266  0.00065028 0.00334988] [ 0.01483851 -0.00953137  0.01422684 ...  0.01928892  0.00024049 -0.00365605]] 

#Layer: dense_relu_sequential.0.bias
#Size: (512,)
#Values : [0. 0.] 

#Layer: dense_relu_sequential.2.weight
#Size: (512, 512)
#Values : [[ 0.00495729 -0.01029267 -0.00672846 ...  0.02216997 -0.00423945 0.00603404] [-0.02003012  0.00643059  0.0076612  ... -0.0097923  -0.01475079 0.00485153]] 

#Layer: dense_relu_sequential.2.bias
#Size: (512,)
#Values : [0. 0.] 

#Layer: dense_relu_sequential.4.weight
#Size: (10, 512)
#Values : [[-0.00212924  0.0067424   0.00244794 ... -0.00193389 -0.01660973 -0.00875264] [-0.01889533  0.01057486 -0.0233639  ... -0.00306869 -0.007126 -0.00609088]] 

#Layer: dense_relu_sequential.4.bias
#Size: (10,)
#Values : [0. 0.]