si_lab09

from keras.datasets import mnist
import numpy as np
import pandas as pd
data = mnist.load_data()
X_train, y_train = data[0][0], data[0][1]
X_test, y_test = data[1][0], data[1][1]
X_train = np.expand_dims(X_train, axis = -1)
X_test = np.expand_dims(X_test, axis = -1)
y_train = pd.get_dummies(pd.Categorical(y_train)).values
y_test = pd.get_dummies(pd.Categorical(y_test)).values
class_cnt = 64

from keras.layers import Dense, Conv2D, MaxPooling2D, GlobalAveragePooling2D,Input
from keras.models import Model
import numpy as np
filter_cnt = 32
kernel_size = (3,3)
act_func = 'selu'
layers = [Conv2D(filter_cnt, kernel_size,
 activation = act_func),
 MaxPooling2D(2,2),
 Conv2D(filter_cnt, kernel_size,
 activation = act_func),
 MaxPooling2D(2,2),
 Conv2D(filter_cnt, kernel_size,
 activation = act_func),
 GlobalAveragePooling2D(),
 Dense(class_cnt, activation = 'softmax')]
output_tensor = input_tensor = Input(X_train.shape[1:])
for layer in layers:
 output_tensor = layer(output_tensor)
ANN = Model(inputs = input_tensor, outputs = output_tensor)
ANN.compile(loss = 'categorical_crossentropy',
 metrics = 'accuracy', optimizer = 'adam')

from keras.layers import Conv2D, MaxPooling2D
from keras.layers import concatenate
def add_inseption_module(input_tensor):
     act_func = 'relu'
     paths = [
        [Conv2D(filters = 64, kernel_size=(1,1),
        padding='same', activation=act_func)
        ],
        [Conv2D(filters = 96, kernel_size=(1,1),
        padding='same', activation=act_func),
        Conv2D(filters = 128, kernel_size=(3,3),
        padding='same', activation=act_func)
        ],
        [Conv2D(filters = 16, kernel_size=(1,1),
        padding='same', activation=act_func),
        Conv2D(filters = 32, kernel_size=(5,5),
        padding='same', activation=act_func)
        ],
        [MaxPooling2D(pool_size=(3,3),
        strides = 1, padding='same'),
        Conv2D(filters = 32, kernel_size=(1,1),
        padding='same', activation=act_func)
        ]
     ]
     for_concat = []
     for path in paths:
        output_tensor = input_tensor
        for layer in path:
            output_tensor = layer(output_tensor)
        for_concat.append(output_tensor)
        return concatenate(for_concat)
output_tensor = input_tensor = Input(X_train.shape[1:])
insept_module_cnt = 2
for i in range(insept_module_cnt):
 output_tensor = add_inseption_module(output_tensor)
output_tensor = GlobalAveragePooling2D()(output_tensor)
output_tensor = Dense(class_cnt, activation='softmax')(output_tensor)
ANN = Model(inputs = input_tensor, outputs = output_tensor)
ANN.compile(loss = 'categorical_crossentropy', metrics = 'accuracy', optimizer = 'adam')
from keras.utils.vis_utils import plot_model
plot_model(ANN, show_shapes=True)