Hi,
A very basic question:
I liked the idea of modular programming in TensorFlow, wherein if you repeat the name_scopes it would automatically append “_1”, “_2” and so on. This keeps the code small and clean. It worked for simple variable and constants; but when using complex functions like,
# default 3x3 kernel with no skips and ReLU activations
def create_convolution_layer(source, fmaps, kernel_size=3, strides=1, padding="SAME", activation=tf.nn.relu):
with tf.name_scope("convolution"):
return tf.layers.conv2d(source, filters=fmaps, kernel_size=kernel_size, strides=strides, padding=padding, activation=activation, name="layer")
It simply throws an error on calling the above function twice, stating some inner variable of layer (layer/kernel) is already defined.
Am I doing something wrong?
Note: It works if I remove the inner
name=“layer”
part from the above code. Maybe that’s the correct way to do it!?
Regards,
Saurabh Singh Parihar
