test['y'] = np.concatenate(test['y'], axis=0)
print(test['x'].shape, test['y'].shape)
trainDq = DataQueue(trainGen, childCount=3, qsize=8)
model = Vgg16()
model.noOfClasses = 2
model.hiddenUnits = 2048
model.imageShape = [imageSize, imageSize, 3]
model.addMetrics('Accuracy')
model.convActivation = "relu"
model.fcActivation = "leakyrelu"
model.regularizationCoefficient = 0.005
model.regularizationType = "l2"
model.build()
trainDq.start()
try:
trainer = Trainer(model, "adam")
trainer.batchSize = 32
trainer.steps = 1000
trainer.restoreState = True
trainer.maxKeep = 25
trainer.learningRate = 0.000001
trainer.trainPrintFreq = 1
trainer.workingDir = "vggDogCat"
trainer.train(trainDq, test, valPrintFreq=25)
finally:
trainDq.stop()
pass
