#!/usr/bin/python2
# coding: utf-8
import numpy as np
import neural_network as network
import mnist_loader_with_pickle as loader
import sys
if __name__ == '__main__':
print 'network training'
datapath = 'parameter/mnist_dropout/'
#datapath = 'parameter/init_params/'
training_data, validation_data, test_data = loader.load_data_wrapper()
epochs = 50
mini_batch_size = 1
learning_rate = 0.01
dropout_rate = (0.8, 0.9)
net = network.Neural_Network([784, 30, 10], dropout_rate)
net.set_test(test_data)
net.set_validation(validation_data)
train = True
#train = False
if train:
#net.load_parameter(path=datapath)
net.train(training_data, epochs, mini_batch_size, learning_rate)
print 'save parameter? (y/n)'
import network
# Activation functions
import logistic
import rectified_linear
# Layers
import fullyconnected_layer
import convolution_layer
import maxpooling_layer
import input_layer
import mnist_loader_with_pickle as loader
if __name__ == '__main__':
# Load training data and test data
training_data, validation_data, test_data = \
loader.load_data_wrapper()
# Create layers
input_layer = input_layer.InputLayer(784)
input_layer.setData(training_data[:1000])
conv1 = convolution_layer.ConvolutionLayer( \
(28, 28, 1), (4, 4, 6), (24, 24, 6), logistic.LogisticFunction(), 0.1)
pool1 = maxpooling_layer.MaxPoolingLayer( \
(24, 24, 6), (2, 2), (12, 12, 6))
conv2 = convolution_layer.ConvolutionLayer( \
(12, 12, 6), (4, 4, 14), (8, 8, 14), logistic.LogisticFunction(), 0.1)
pool2 = maxpooling_layer.MaxPoolingLayer( \
(8, 8, 14), (2, 2), (4, 4, 14))
full1 = fullyconnected_layer.FullyConnectedLayer( \
224, 100, logistic.LogisticFunction(), 0.1, 1.0, 0.0)
full2 = fullyconnected_layer.FullyConnectedLayer( \
#!/usr/bin/python2
# coding: utf-8
# import library
import sys
import numpy as np
# import my library
import network
import mnist_loader_with_pickle
import read
if __name__ == '__main__':
## read mnist data
training_data, validation_data, test_data = \
mnist_loader_with_pickle.load_data_wrapper()
ball_data = read.load_ball()
training_data = training_data + ball_data
## create network
## input layer: 784 (28 * 28)
## hidden layer: 30 * 1
## output layer: 11 (digits and ball)
net = network.Network([784, 30, 11])
## execute training
net.SGD(training_data, 30, 10, 3.0, test_data=test_data)
net.save_data()
#net.evaluate(test_data)
#!/usr/bin/python2
# coding: utf-8
# import library
# import sys
# import numpy as np
# import my library
# import network
import mnist_loader_with_pickle
# import read
import cnn
if __name__ == "__main__":
## read mnist data
training_data, validation_data, test_data = mnist_loader_with_pickle.load_data_wrapper()
# ball_data = read.load_ball()
# training_data = training_data + ball_data
## create cnn
nn = cnn.CNN((4, 20, 1), (28, 28), output=False)
nn.process(training_data[0:11])
## create network
## input layer: 784 (28 * 28)
## hidden layer: 30 * 1
## output layer: 11 (digits and ball)
# net = network.Network([784, 30, 11])
## execute training
# net.SGD(training_data, 30, 10, 3.0, test_data=test_data)