concept y_pred[:, 2:, :] and downsample_factor ?

[PythonImageDeveloper]

Hi
Q1 - I don't understand why don't use all of y_pred[ :,:,:]? why use y_pred[ :, 2:,:]? why don't use 0,1 dims?

def ctc_lambda_func(args):
    y_pred, labels, input_length, label_length = args
    # the 2 is critical here since the first couple outputs of the RNN
    # tend to be garbage:
    y_pred = y_pred[:, 2:, :]
    return K.ctc_batch_cost(labels, y_pred, input_length, label_length)

Q2 - What's meaning the downsampling parameter? if we change the input_size, is necessary to change the value of downsampling parameter? what's the principle of the parameter?
and I also don't understand, why use this line and why we multiply (self.img_w // self.downsample_factor - 2) to np.ones((self.batch_size, 1)) ? what's the advantage of this work? if we use only np.ones((self.batch_size, 1)) , Causing the problem?
and why use (self.downsample_factor - 2) ? why 2 ?