import pandas, numpy

from keras.utils import np_utils
from keras.models import Sequential
from keras.layers import *
from keras.preprocessing.image import ImageDataGenerator

#=======
train_raw = numpy.array(pandas.read_csv('../input/train.csv'))

train_x = train_raw[:,1:].reshape(train_raw.shape[0], 28, 28, 1).astype('float32') / 255.
train_y = np_utils.to_categorical(train_raw[:,0], 10)


test_raw = numpy.array(pandas.read_csv('../input/test.csv'))
test_x = test_raw.reshape(test_raw.shape[0], 28, 28, 1).astype('float32') / 255.

#=======
model = Sequential()

model.add( Convolution2D(36, 7, 7, input_shape=(28, 28, 1)) )
model.add( BatchNormalization() )
model.add( Activation('relu') )
model.add( MaxPooling2D() )

model.add( Convolution2D(36, 4, 4) )
model.add( BatchNormalization() )
model.add( Activation('relu') )
model.add( MaxPooling2D() )


model.add( Flatten() ) 

model.add( Dense(output_dim=128) )
model.add( BatchNormalization() )
model.add( Activation('relu') )

model.add( Dense(output_dim=10, activation='softmax') )

model.compile(loss='categorical_crossentropy', optimizer='adamax', metrics=['accuracy'])

#=======
datagen = ImageDataGenerator(
		width_shift_range=0.1,
		height_shift_range=0.1,
		zoom_range=0.1
	)

#=======
model.fit_generator( datagen.flow(train_x, train_y, batch_size=64), len(train_x), nb_epoch=7 )

y = model.predict_classes(test_x)
numpy.savetxt('mnist.csv', numpy.c_[range(1,len(y)+1),y], delimiter=',', header = 'ImageId,Label', comments = '', fmt='%d')