Text this: An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets