1:13 in the video
The conv2d_blocks have two convolutional layers which are created by for loops. Is this possible because the model is stored as the Model object separately from the conv2d_block code? Doesn’t the garbage collector delete the memory allocated to the for loop after the loop is finished?
As far as I understand, when the model is called, it calls the conv2d_block function, and those conv2d blocks are created at that moment, part of the model.
if you notice the for loop indentation is inside the def statement allowing the iterative custom computations.
When a for loop is placed inside a def statement, which defines a function, it operates on the input tensor passed to that function.
The for loop iterates over the tensor’s elements or dimensions, based on how it’s structured within the function.