Is reflection padding always better than zero padding or does it depend on the image type e.g. natural images vs hand drawn?
3 Likes
what does 16 represent in [16,128,512,512]?
1 Like
Batch size x channels x height x width
ended up ignoring those columns in lesson6 notebook as well. it would still be nice to know where to get the additional tables for store_states, state_names, googletrend, weather
Just a general reminder that there is no need to @ mention admins in the thread.
6 Likes
why divide by 6 in expand?
8 Likes
@KarlH Of course, the final model will be trained with all the data, not a subset. I am suggesting that applying the “dropout” method to the input data for each layer allows you to speed up the model building phase.
Do you have a suggestion if our question isn’t answered?
Think about kernels as weights. So how do we initialize weights? We use random initialization except if we have a pretrained model and we want to use it, then we use the pretrained weights.
1 Like
I can guarantee you that @ mention people (especially admins) won’t result in your question being answered. You can try asking it again, just in case people didn’t notice.
2 Likes
but Jeremy told that he intentionally created 1’s in the tensor. so they are not random in this case
Ask in the after-class discussion, or other branch of forum, maybe 
3 Likes
I believe you might have to normalize it (I guess)
We aren’t able to answer all questions in real time (it is a lot to keep up with the lesson, the discussion, and to formulate answers). If you are still stumped after the lesson, please be sure to search for an answer on your own and then you can create a topic with the question (possibly adding more detail) or ask on the relevant thread.
Also, feel free to @ mention me to point out questions with >6 votes that I have missed.
i was looking to know
if kernel
has postive 1st row,zero middle and -ve last row
what type of edges is it trying to find…
2 Likes
I guess someone has already asked but what the Lambda layer does? Is it something like Flatten() from Keras?
Not here, in the library. This is an example.
whats the intuition behind doing the avg pooling?
1 Like
That’s what the “Adaptive” part of Adaptive pooling does. It means if you send in different sized images, it still gets you down to the same size before your linear layers.