from NeuralNetworks.convolutionNN import createConvolutionnalSharedWeightLayer, createSubSamplingLayerSharedWeight, \
createConvolutionnalLayer, createSubSamplingLayer
sys.path.append('../')
__author__ = 'Gabriel'
'''
Create a neural network then train in for the character recognition
Use pybrain library
Try to work on convolution and connect the layer in a meaningful way
'''
start = time.time()
# Last row is the label (64 features + 1 label)
############################LOAD DATA
(X, Y, Xtest, Ytest) = opticalCharacterManipulation.loadTrainAndTestFeaturesData(True, True)
############################CREATE NETWORK
# Generate the neural network model
nn = FeedForwardNetwork()
# Input Layer
feature_size = len(X[0])
print("Input size : " + str(feature_size) + " features")
inputLayer = LinearLayer(feature_size, name="Input")
nn.addInputModule(inputLayer)
convLayer = createConvolutionnalLayer(nn, inputLayer, sizewindow=2)
import opticalCharacterManipulation
import matplotlib.pyplot as plt
import time
from features import *
start = time.time()
#Last row is the label (64 features + 1 label)
'''
Improvement :
Layer join on subcells above
Then second hidden layer
Then result
'''
(X,Y,Xtest,Ytest) = opticalCharacterManipulation.loadTrainAndTestFeaturesData(sidePoints)
feature_size = len(X[0])
# Normalize
features_training = range(0,len(X[i]))
for i in range(0,len(X)):
for j in range(0,len(X[i])):
features_training[j] += [X[i][j]]
# Generate the neural network model
nn = FeedForwardNetwork()
inLayer = LinearLayer(feature_size, name="Input")
hiddenLayer = SigmoidLayer(50, name="First hidden layer")
outLayer = LinearLayer(10, name="Output")
nn.addInputModule(inLayer)