Predicting human behavior in size-variant repeated games through deep convolutional neural networks