how to debug torch lua program

320 views
Skip to first unread message

fireman

unread,
Oct 12, 2016, 8:51:56 PM10/12/16
to torch7
Hi, I am new to lua and help you can help me. I downloaded a torch lua convolutional neural network program from online and would like to study it. The network has many layers in it. I would like to step through each layer of them and exam the input and output of each layer. How can I do it? Currently I am using ZeroBrane Studio. 

Thanks a lot

Shuaib Ahmed S

unread,
Oct 12, 2016, 11:30:46 PM10/12/16
to torch7 on behalf of fireman
Hi, 

You can configure eclipse with torch for debugging.

On Thu, Oct 13, 2016 at 6:21 AM, fireman via torch7 <torch7+APn2wQeW3esQD-dFTzB55wMyB...@googlegroups.com> wrote:
Hi, I am new to lua and help you can help me. I downloaded a torch lua convolutional neural network program from online and would like to study it. The network has many layers in it. I would like to step through each layer of them and exam the input and output of each layer. How can I do it? Currently I am using ZeroBrane Studio. 

Thanks a lot

--
You received this message because you are subscribed to the Google Groups "torch7" group.
To unsubscribe from this group and stop receiving emails from it, send an email to torch7+unsubscribe@googlegroups.com.
To post to this group, send email to tor...@googlegroups.com.
Visit this group at https://groups.google.com/group/torch7.
For more options, visit https://groups.google.com/d/optout.

Vislab

unread,
Oct 13, 2016, 7:17:31 AM10/13/16
to torch7
To visualize the contents of a network you simply need to execute a forward pass of your input through the network and then you can access each individual layer's contents such as output, weights, bias, etc. You can do this using breakpoints in zbstudio or just simply launch a terminal, type "th" or "qlua", load your network, generate an input and do a :forward(). To access any layer in your network, you can simply do "net.modules[id]".

Sangram Gupta

unread,
Oct 13, 2016, 9:50:14 AM10/13/16
to torch7 on behalf of fireman

Follow Andrej Karpathay's lecture on calculating convoluting steps. I was on a same situation as you, debugging helped little to understand. That lecture gave a more meaningful understanding. CS231n and lecture might be the 6th one.

Just a suggestion.


On 13 Oct 2016 6:22 am, "fireman via torch7" <torch7+APn2wQeW3esQD-dFTzB55wMyB...@googlegroups.com> wrote:
Hi, I am new to lua and help you can help me. I downloaded a torch lua convolutional neural network program from online and would like to study it. The network has many layers in it. I would like to step through each layer of them and exam the input and output of each layer. How can I do it? Currently I am using ZeroBrane Studio. 

Thanks a lot

--
Reply all
Reply to author
Forward
0 new messages