Python: Advanced Guide to Artificial Intelligence

LeNet CNN for MNIST with Keras

Let us revisit the same LeNet architecture with the same dataset to build and train the CNN model in Keras:

1. Import the required Keras modules:

import kerasfrom keras.models import Sequentialfrom keras.layers import Conv2D,MaxPooling2D, Dense, Flatten, Reshapefrom keras.optimizers import SGD

2. Define the number of filters for each layer:

n_filters=[32,64]

3. Define other hyper-parameters:

learning_rate = 0.01n_epochs = 10batch_size = 100

4. Define the sequential model and add the layer to reshape the input data to shape (n_width,n_height,n_depth):

model = Sequential()model.add(Reshape(target_shape=(n_width,n_height,n_depth),                   input_shape=(n_inputs,))         )

5. Add the first convolutional layer with 4 x 4 kernel filter, ...