|
|
|
|
|
import numpy as np
|
|
import tensorflow.keras as tfk
|
|
import tensorflow.keras.layers as tfl
|
|
|
|
def lenet_model(input_shape):
|
|
'''
|
|
Implement Functional Keras API to train model. The model has 5 layers
|
|
-----------------------------------------------------------------------------------
|
|
1. conv layer:
|
|
k_size: (5, 5)
|
|
stride: (1, 1)
|
|
n_filters: 6
|
|
padding: 'same'
|
|
activation: relu
|
|
|
|
2. avg pooling:
|
|
pool_size: (2, 2)
|
|
|
|
3. conv layer:
|
|
k_size: (5, 5)
|
|
stride: (1, 1)
|
|
n_filters: 16
|
|
padding: 'valid'
|
|
activation: relu
|
|
|
|
4. avg pooling
|
|
pool_size: (2, 2)
|
|
strides: (2, 2)
|
|
|
|
5. conv layer:
|
|
k_size: (5, 5)
|
|
stride: (1, 1)
|
|
n_filters: 120
|
|
padding: 'valid'
|
|
activation: relu
|
|
|
|
5. flatten
|
|
6. classifier
|
|
activation: softmax
|
|
classes: 10
|
|
-----------------------------------------------------------------------------------------------------------------------
|
|
'''
|
|
input_img = tfk.Input(shape = input_shape)
|
|
|
|
z1 = tfl.Conv2D(kernel_size = (5, 5), strides = (1, 1), filters = 6, padding = 'same', activation = 'tanh')(input_img)
|
|
p1 = tfl.AveragePooling2D(pool_size = (2, 2))(z1)
|
|
z2 = tfl.Conv2D(kernel_size = (5, 5), strides = (1, 1), filters = 16, padding = 'valid', activation = 'tanh')(p1)
|
|
p2 = tfl.AveragePooling2D(pool_size = (2, 2), strides = (2, 2))(z2)
|
|
z3 = tfl.Conv2D(kernel_size = (5, 5), strides = (1, 1), filters = 120, padding = 'valid', activation = 'tanh')(p2)
|
|
flat = tfl.Flatten()(z3)
|
|
tanh = tfl.Dense(units = 84, activation = 'tanh')(flat)
|
|
output = tfl.Dense(units = 10, activation = 'softmax')(tanh)
|
|
|
|
model = tfk.Model(inputs = input_img, outputs = output)
|
|
return model |
