Classification of MNIST Digits dataset using CNN using Keras

keras_cnn_mnist.py

import numpy as np

from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense,Conv2D,MaxPooling2D
from keras.layers import Flatten,Dropout,Activation

#create a model
model = Sequential()
model.add(Conv2D(64,(5,5),input_shape=(28,28,1),data_format="channels_last",strides=(1,1),padding="valid",activation="relu"))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Conv2D(64,(3,3),activation="relu",padding="valid"))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(256,activation="relu"))
model.add(Dropout(0.2))
model.add(Dense(64,activation="relu"))
model.add(Dropout(0.2))
model.add(Dense(10,activation="softmax"))

from keras import losses
from keras import optimizers
sgd = optimizers.SGD(lr=0.01,decay=0.001,momentum=0.9,nesterov=True)
adam = optimizers.Adam(lr=0.01, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.001, amsgrad=True)
model.compile(loss=losses.categorical_crossentropy,optimizer=optimizers.Adadelta(),metrics=['accuracy'])
print(model.summary())

(trainData,trainLabels),(testData,testLabels)=mnist.load_data()

#normalizaing image pixels
trainData = trainData.reshape(len(trainData),28,28,1).astype('float32')/255
testData = testData.reshape(len(testData),28,28,1).astype('float32')/255

from keras.utils import np_utils
mTrainLabels = np_utils.to_categorical(trainLabels, 10)
mTestLabels = np_utils.to_categorical(testLabels, 10)

batch_size=128
epochs=10

model.fit(trainData,
        mTrainLabels,
        validation_data=(testData,mTestLabels),
        batch_size=batch_size,
        verbose=1,
        epochs=epochs
         )

score = model.evaluate(testData,mTestLabels,verbose=1)
print("Test Loss: {}".format(score[0]))
print("Test Accuracy: {}".format(score[1]))


#Best of 10 epochs
"""
all activations are relu and output layer has softmax/ loss is categorical_crossentropy
0.9894 using conv(64,(5,5)),conv(64,(3,3)),128 hidden layers and sgd(lr=0.01,decay=0.001), batch_size=150
0.9935 using conv(64,(5,5)),conv(64,(3,3)),128 hidden layers and adam(lr=0.015,decay=0.002), batch_size=200, amsgrad=True
0.9923 using conv(64,(5,5)),conv(64,(3,3)),256,64 hidden layers and adam(lr=0.01,decay=0.001), batch_size=128, amsgrad=True
Time = 200s*10 epochs = 2000s = 33 min approx
0.9918 using conv(64,(5,5)),conv(64,(3,3)),256,64 hidden layers and Adadelta, batch_size=128
"""