Janghoon Park

Janghoon Park

Deeplearning Researcher

Mlp codes

7 minute read

Library import

# library import
from keras.datasets import mnist
from keras.utils.np_utils import to_categorical
import numpy as np
from sklearn.preprocessing import minmax_scale
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.optimizers import Adam

Data import

#load data to use Keras.dataset.mnist
(X_train, y_train), (X_test, y_test) = mnist.load_data()


# Format transformation
samples_num = X_train.shape[0]  
width = X_train.shape[1]
height = X_train.shape[2]
X_train = X_train.reshape(samples_num, width * height)
num_of_test_samples = X_test.shape[0]
X_test = X_test.reshape(num_of_test_samples, width * height)

# Convert int to float
X_train = X_train.astype(np.float64)
X_test = X_test.astype(np.float64)


# Regulization
X_train = minmax_scale(X_train, feature_range=(0, 1), axis=0)
X_test = minmax_scale(X_test, feature_range=(0, 1), axis=0)

# Data Division
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)

MLP

# Multilayer Perceptron (MLP) Create
model = Sequential()

# Input-Layer
model.add(Dense(256, input_dim=width * height,  activation='elu'))
model.add(Dropout(0.3))

# 2nd-Layer
model.add(Dense(256,  activation='elu'))
model.add(Dropout(0.3))

# 3rd-Layer
model.add(Dense(256, activation='elu'))
model.add(Dropout(0.3))

# 4th-Layer
model.add(Dense(256, activation='elu'))
model.add(Dropout(0.3))

# 5th-Layer
model.add(Dense(256, activation='elu'))
model.add(Dropout(0.3))


# 6th-Layer
number_of_class = 10 
model.add(Dense(number_of_class, activation='softmax'))  

# Cost Function & Optimizer Setting
# CE & Adam
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

Traing & Evaluation

# model training
training_epochs = 100
batch_size = 200
model.fit(X_train, y_train,batch_size,training_epochs)

Epoch 1/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0260 - accuracy: 0.9931
Epoch 2/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0280 - accuracy: 0.9924
Epoch 3/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0308 - accuracy: 0.9918
Epoch 4/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0303 - accuracy: 0.9919
Epoch 5/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0275 - accuracy: 0.9923
Epoch 6/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0281 - accuracy: 0.9925
Epoch 7/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0298 - accuracy: 0.9927
Epoch 8/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0257 - accuracy: 0.9935
Epoch 9/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0244 - accuracy: 0.9936
Epoch 10/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0271 - accuracy: 0.9933
Epoch 11/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0265 - accuracy: 0.9930
Epoch 12/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0270 - accuracy: 0.9930
Epoch 13/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0269 - accuracy: 0.9930
Epoch 14/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0241 - accuracy: 0.9938
Epoch 15/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0341 - accuracy: 0.9916
Epoch 16/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0294 - accuracy: 0.9928
Epoch 17/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0281 - accuracy: 0.9928
Epoch 18/100
300/300 [==============================] - 1s 4ms/step - loss: 0.0258 - accuracy: 0.9928
Epoch 19/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0262 - accuracy: 0.9928
Epoch 20/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0224 - accuracy: 0.9937
Epoch 21/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0273 - accuracy: 0.9929
Epoch 22/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0277 - accuracy: 0.9931
Epoch 23/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0248 - accuracy: 0.9936
Epoch 24/100
300/300 [==============================] - 1s 5ms/step - loss: 0.0249 - accuracy: 0.9934
Epoch 25/100
300/300 [==============================] - 2s 5ms/step - loss: 0.0236 - accuracy: 0.9938
Epoch 26/100
300/300 [==============================] - 2s 5ms/step - loss: 0.0282 - accuracy: 0.9928
Epoch 27/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0248 - accuracy: 0.9938
Epoch 28/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0255 - accuracy: 0.9931
Epoch 29/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0257 - accuracy: 0.9934
Epoch 30/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0280 - accuracy: 0.9928
Epoch 31/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0260 - accuracy: 0.9934
Epoch 32/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0232 - accuracy: 0.9937
Epoch 33/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0239 - accuracy: 0.9940
Epoch 34/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0242 - accuracy: 0.9942
Epoch 35/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0260 - accuracy: 0.9934
Epoch 36/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0267 - accuracy: 0.9936
Epoch 37/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0223 - accuracy: 0.9947
Epoch 38/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0321 - accuracy: 0.9929
Epoch 39/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0230 - accuracy: 0.9941
Epoch 40/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0240 - accuracy: 0.9941
Epoch 41/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0246 - accuracy: 0.9936
Epoch 42/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0260 - accuracy: 0.9936
Epoch 43/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0356 - accuracy: 0.9915
Epoch 44/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0253 - accuracy: 0.9934
Epoch 45/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0254 - accuracy: 0.9934
Epoch 46/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0197 - accuracy: 0.9944
Epoch 47/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0184 - accuracy: 0.9947
Epoch 48/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0211 - accuracy: 0.9947
Epoch 49/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0256 - accuracy: 0.9940
Epoch 50/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0276 - accuracy: 0.9932
Epoch 51/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0241 - accuracy: 0.9941
Epoch 52/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0241 - accuracy: 0.9941
Epoch 53/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0223 - accuracy: 0.9945
Epoch 54/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0217 - accuracy: 0.9941
Epoch 55/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0241 - accuracy: 0.9941
Epoch 56/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0221 - accuracy: 0.9945
Epoch 57/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0238 - accuracy: 0.9943
Epoch 58/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0277 - accuracy: 0.9937
Epoch 59/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0203 - accuracy: 0.9951
Epoch 60/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0230 - accuracy: 0.9942
Epoch 61/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0237 - accuracy: 0.9946
Epoch 62/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0249 - accuracy: 0.9941
Epoch 63/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0196 - accuracy: 0.9949
Epoch 64/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0295 - accuracy: 0.9933
Epoch 65/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0303 - accuracy: 0.9935
Epoch 66/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0241 - accuracy: 0.9944
Epoch 67/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0376 - accuracy: 0.9926
Epoch 68/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0334 - accuracy: 0.9924
Epoch 69/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0250 - accuracy: 0.9937
Epoch 70/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0227 - accuracy: 0.9951
Epoch 71/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0211 - accuracy: 0.9947
Epoch 72/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0215 - accuracy: 0.9944
Epoch 73/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0232 - accuracy: 0.9946
Epoch 74/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0281 - accuracy: 0.9934
Epoch 75/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0228 - accuracy: 0.9946
Epoch 76/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0218 - accuracy: 0.9952
Epoch 77/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0176 - accuracy: 0.9952
Epoch 78/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0190 - accuracy: 0.9953
Epoch 79/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0215 - accuracy: 0.9953
Epoch 80/100
300/300 [==============================] - 2s 6ms/step - loss: 0.0257 - accuracy: 0.9939
Epoch 81/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0222 - accuracy: 0.9944
Epoch 82/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0224 - accuracy: 0.9945
Epoch 83/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0187 - accuracy: 0.9952
Epoch 84/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0202 - accuracy: 0.9944
Epoch 85/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0231 - accuracy: 0.9952
Epoch 86/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0239 - accuracy: 0.9948
Epoch 87/100
300/300 [==============================] - 2s 8ms/step - loss: 0.0317 - accuracy: 0.9941
Epoch 88/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0387 - accuracy: 0.9923
Epoch 89/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0303 - accuracy: 0.9930
Epoch 90/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0648 - accuracy: 0.9925
Epoch 91/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0296 - accuracy: 0.9925
Epoch 92/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0467 - accuracy: 0.9919
Epoch 93/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0268 - accuracy: 0.9932
Epoch 94/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0238 - accuracy: 0.9940
Epoch 95/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0212 - accuracy: 0.9945
Epoch 96/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0450 - accuracy: 0.9934
Epoch 97/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0290 - accuracy: 0.9934
Epoch 98/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0357 - accuracy: 0.9920
Epoch 99/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0216 - accuracy: 0.9944
Epoch 100/100
300/300 [==============================] - 2s 7ms/step - loss: 0.0232 - accuracy: 0.9945





<tensorflow.python.keras.callbacks.History at 0x7faa2a454e50>

# Model evaluation using test set
print('Model evaluation')
evaluation = model.evaluate(X_test, y_test, batch_size=batch_size)
print('Accuracy: ' + str(evaluation[1]))

Model evaluation
50/50 [==============================] - 0s 2ms/step - loss: 0.1513 - accuracy: 0.9850
Accuracy: 0.9850000143051147

Updated:

You may also enjoy