Implementing a CNN

So far, we have implemented convolution and pooling layers. Now, we will combine these layers to create a CNN that recognizes handwritten digits and implement it, as shown in Figure 7.23.

As shown in Figure 7.23, the network consists of "Convolution – ReLU – Pooling – Affine – ReLU – Affine – Softmax" layers. We will implement this as a class named SimpleConvNet:

Figure 7.23: Network configuration of a simple CNN

Now, let's look at the initialization of SimpleConvNet (__init__). It takes the following arguments:

• input_dim: Dimensions of the input data (channel, height, width).
• conv_param: Hyperparameters of the convolution layer (dictionary). The following are the dictionary keys:
• filter_num: Number of filters
• filter_size: Size of the filter
• stride: Stride
• pad: Padding
• hidden_size: Number of neurons in the hidden layer (fully connected)
• output_size...