testing.py

from src.training_data import Data
from src.cnn import CNN
from src.model import Model

## TODO: Add unit testing using pytest

def loadData_testing(testingData):
    print('TESTING DATA LOADING:')
    for i, dataset in enumerate(testingData['dataset']):
        data = Data(dataset)
        data.load()
        checkScale(data, [0, 255])
        if not testingData['colour'][i]:
            data.rgb2greyScale()
            data.blackAndWhite()
            checkScale(data, [0,1])
        checkDim(data, testingData['resolution'][i])
        checkChannels(data, testingData['resolution'][i])
        
        print('Data loaded correctly for "{}" dataset'.format(dataset))
    print('\n')


def train_testing(testingData):
    print('TESTING CNN TRAINING:')
    for i, dataset in enumerate(testingData['dataset']):
        data = Data(dataset)
        data.load()
        data = getOneDataPoint(data)
        if not testingData['colour'][i]:
            data.rgb2greyScale()

        try:
            cnn = CNN(data)
        except Exception as e:
            raise ValueError('Failed when creating model. {}'.format(e))
        try:
            cnn.build(nConvBlocks=1,nFilters=1, kernelSize=1, stride=2)
        except Exception as e:
            raise ValueError('Failed when building model. {}'.format(e))
        try:
            model = Model(cnn)
            model.compile(optimizer='adam',loss='mean_squared_error')
        except Exception as e:
            raise ValueError('Failed when compiling model. {}'.format(e))
        try:
            model.train(epochs=1, nBatch=1, earlyStopPatience=1)
        except Exception as e:
            raise ValueError('Failed when training model. {}'.format(e))
        try:
            model.predict()
        except Exception as e:
            raise ValueError('Failed when predicting. {}'.format(e))

        print('Training successfully tested for "{}" dataset'.format(dataset))
        del model; del data


def checkDim(data, refResolution):
    assert data.x_train.shape[1:3] == refResolution[0:2], 'Training dataset dimension is incorrect'
    assert data.x_val.shape[1:3] == refResolution[0:2], 'Validation dataset dimension is incorrect'
    assert data.x_test.shape[1:3] == refResolution[0:2], 'Testing dataset dimension is incorrect'


def checkChannels(data, refResolution):
    assert data.resolution[2] == refResolution[2], 'Data has an incorrect number of channels'


def checkScale(data, lim):
    assert data.x_train.min() >= lim[0] and data.x_train.max() <= lim[1],\
        'Training dataset not scaled between {} and {}'.format(lim[0],lim[1])
    assert data.x_val.min() >= lim[0] and data.x_val.max() <= lim[1], \
        'Validation dataset not scaled between {} and {}'.format(lim[0], lim[1])
    assert data.x_test.min() >= lim[0] and data.x_test.max() <= lim[1],\
        'Testing dataset not scaled between {} and {}'.format(lim[0],lim[1])

def getOneDataPoint(data):
    data.x_train = data.x_train[0].reshape(1, data.resolution[0], data.resolution[1], data.resolution[2])
    data.x_val = data.x_val[0].reshape(1, data.resolution[0], data.resolution[1], data.resolution[2])
    data.x_test = data.x_test[0].reshape(1, data.resolution[0], data.resolution[1], data.resolution[2])
    return data


if __name__ == '__main__':
    testingData = {
        'dataset' : ['mnist', 'afreightdata_test', 'afreightdata_test', 'beam_test', 'beam_homog_test'],
        'colour' : [False, False, True, False, False, False],
        'resolution' : [(28, 28, 1), (120, 160, 1), (120, 160, 3), (100,300,1), (80,160,1), (80,160,1)]
    }
    loadData_testing(testingData)
    train_testing(testingData)