#import stuff

# package for math operations
import numpy
# package for sigmoid function
import scipy.special
# package for graphics
import matplotlib.pyplot
get_ipython().magic('matplotlib inline')


# create the network


class neuralNetwork:
    def __init__(self, inputnodes, hiddennodes, outputnodes, learningrate):
        # create nodes
        self.inodes = inputnodes
        self.hnodes = hiddennodes
        self.onodes = outputnodes
        # use learningrate
        self.lr = learningrate
        # link weights
        self.wih = numpy.random.normal(0.0, pow(self.hnodes, -0.5), (self.hnodes, self.inodes))
        self.who = numpy.random.normal(0.0, pow(self.hnodes, -0.5), (self.onodes, self.hnodes))
        # create sigmoid function
        self.activation_function = lambda x: scipy.special.expit(x)

        pass

    def train(self, inputs_list, targets_list):
        # convert inputs into array
        inputs = numpy.array(inputs_list, ndmin=2).T
        targets = numpy.array(targets_list, ndmin=2).T
        # calculate inputs and outputs of the hidden layer
        hidden_inputs = numpy.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        # calculate inputs and outputs of output layer
        final_inputs = numpy.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)
        # calculate the error
        output_errors = targets - final_outputs
        hidden_errors = numpy.dot(self.who.T, output_errors)
        # adjust weights depending on error
        self.who += self.lr * numpy.dot((output_errors * final_outputs * (1.0 - final_outputs)), numpy.transpose(hidden_outputs))
        self.wih += self.lr * numpy.dot((hidden_errors * hidden_outputs * (1.0 - hidden_outputs)), numpy.transpose(inputs))

        pass

    def query(self, inputs_list):
        # convert inputs into array
        inputs = numpy.array(inputs_list, ndmin=2).T
        # calculate inputs and outputs of hidden layer
        hidden_inputs = numpy.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        # calculate inputs and outputs of output layer
        final_inputs = numpy.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)

        return final_outputs

        pass


# train network

# set number of nodes and inputs
input_nodes = 784
hidden_inputs = 100
output_nodes = 10

# set learningrate

learning_rate = 0.2

# create instance of neural network
n = neuralNetwork(input_nodes, hidden_inputs, output_nodes, learning_rate)

# load training data
training_data_file = open ("mnist_train.csv", 'r')
training_data_list = training_data_file.readlines()
training_data_file.close()

# split data records and train
epochs = 2
for e in range(epochs):
    for record in training_data_list:
        all_values = record.split(',')
        inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
        targets = numpy.zeros(output_nodes) + 0.01
        targets[int(all_values[0])] = 0.99
        n.train(inputs, targets)
        pass
    pass



# test neural network

# how well does the network perform?
scorecard = []
# open test data
test_data_file = open ("mnist_test.csv", 'r')
test_data_list = test_data_file.readlines()
test_data_file.close()
# iterate through the whole test set
for record in test_data_list:
    # prepare data
    all_values = record.split(',')
    # print the correct answer
    correct_label = int(all_values[0])
   # print(correct_label, "correct label")
    inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
    # calculate outputs
    outputs = n.query(inputs)
    # translate network's answer
    label = numpy.argmax(outputs)
   # print(label, "networks answer")
    # fill scorecard
    if (label == correct_label):
        scorecard.append(1)
    else:
        scorecard.append(0)
        pass

    pass





# calculate performance of network
scorecard_array = numpy.asarray(scorecard)
print ("Performance = ", scorecard_array.sum() / scorecard_array.size)