def createNeuralNetwork(self, inputFilePath, trainingPercent, maxItr,
nHiddenLayers, nNeurons):
"""
given function creates neural network,
and displays its waight at each level and accuracy of the network
0. read dataset
1. split dataset into training and testing datasets
2. initialize network
a. take weights randomly for hidden and output layers
3. forward propogation
a. neuron activation: sigma(wi*xi)
b. neuron transfer : sigmoid function 1/(1+e(-x))
4. back propogation
a. transfer derivative
b. error back propogation
5. train network
a. update weights
6. predict
7. find mean square errors
"""
#0. read dataset
myIO = MyIO()
inputDataFrame = myIO.inputProcessedCSV(filePath=inputFilePath)
headerList = inputDataFrame.columns.values
#1. split dataset into training and testing dataset\
myUtility = MyUtility()
trainingDataFrame, testingDataFrame = myUtility.splitDataset(\
inputDataFrame = inputDataFrame,\
trainingPercent = trainingPercent)
#2. initializeNeuralNetwork
uniqueClasses = inputDataFrame['class'].unique()
numOfUniqueClasses = uniqueClasses.size
trainingDataArr = trainingDataFrame.values
testingDataArr = testingDataFrame.values
trainingAtrArr, trainingClassArr, trainingAtrHeader = \
myUtility.segregateAttributesAndClass(\
inputArr = trainingDataArr,\
inputHeader = headerList)
testingAtrArr, testingClassArr, testingAtrHeader =\
myUtility.segregateAttributesAndClass(\
inputArr = testingDataArr,\
inputHeader = headerList)
nRows, nCols = trainingDataArr.shape
neuralNetwork = NeuralNetwork( nInputs = nCols-1,\
nHiddenLayers = nHiddenLayers, \
nNeurons = nNeurons, \
nOutputs = numOfUniqueClasses)
#4. back propogation
# neuralNetwork.findBackwardPropagationError(targetValue = [1,0,0])
trainingError = neuralNetwork.trainNetwork(\
trainingDataArr = trainingDataArr,\
nIteration = maxItr,\
numOfUniqueClasses=numOfUniqueClasses, \
learningRate=0.5)
trainingPredictedOPArr = neuralNetwork.predictDataset(\
testingDataSet = trainingDataArr)
testingPredictedOPArr = neuralNetwork.predictDataset(\
testingDataSet = testingDataArr)
trainingError = neuralNetwork.meanSquareError(\
targetArr = trainingClassArr,\
predictedOutputArr = trainingPredictedOPArr)
testingError = neuralNetwork.meanSquareError(\
targetArr = testingClassArr,\
predictedOutputArr = testingPredictedOPArr)
#debug
print("\nAfter training neural network:\n")
neuralNetwork.printNeuralNetworkWeights(headerList=trainingAtrHeader)
print('\ntrainingError = {}'.format(trainingError))
print('testingError = {} '.format(testingError))
