[Convolutions on Medium] Used in Medium blog post series #python #convolutions

conv.py

def apply_conv(data, kernel, conv):   
    """
    Args:
        data (NDArray): input data.
        kernel (NDArray): convolution's kernel parameters.
        conv (Block): convolutional layer.
    Returns:
        NDArray: output data (after applying convolution).
    """
    # add dimensions for batch and channels if necessary
    while data.ndim < len(conv.weight.shape):
        data = data.expand_dims(0)
    # add dimensions for channels and in_channels if necessary
    while kernel.ndim < len(conv.weight.shape):
        kernel = kernel.expand_dims(0)
    # check if transpose convolution
    if type(conv).__name__.endswith("Transpose"):
        in_channel_idx = 0
    else:
        in_channel_idx = 1
    # initialize and set weight
    conv._in_channels = kernel.shape[in_channel_idx]
    conv.initialize()
    conv.weight.set_data(kernel)
    return conv(data)