How to draw the learning curve?

6,492 views
Skip to first unread message

Phil Huang

unread,
Dec 4, 2014, 2:30:31 AM12/4/14
to caffe...@googlegroups.com
I'm wondering if there is any utility to draw the learning curve during or after training which will help to optimize the network.
I've search the issues, PR list and the source code but found nothing.

Yasser Souri

unread,
Dec 5, 2014, 3:39:15 AM12/5/14
to caffe...@googlegroups.com
I've written this small python script that draws the loss function and the accuracy on test set on a plot.

You have to first save the whole output that was printed to the terminal during training to a txt file. Then pass that txt file as an argument to this python script. You can also pass multiple txt files.

Dependencies are numpy, matplotlib and click.


I originally did this to visualise fine-tuning.

Evan Shelhamer

unread,
Dec 5, 2014, 8:42:49 PM12/5/14
to Phil Huang, caffe...@googlegroups.com

Evan Shelhamer

On Wed, Dec 3, 2014 at 11:30 PM, Phil Huang <phil...@gmail.com> wrote:
I'm wondering if there is any utility to draw the learning curve during or after training which will help to optimize the network.
I've search the issues, PR list and the source code but found nothing.

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/22ed2081-5ab8-4c14-b315-bb12c42ab538%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Phil Huang

unread,
Dec 8, 2014, 9:05:59 AM12/8/14
to caffe...@googlegroups.com
Thanks for the useful script!

在 2014年12月5日星期五UTC+8下午4时39分15秒,Yasser Souri写道:

Phil Huang

unread,
Dec 8, 2014, 9:07:31 AM12/8/14
to caffe...@googlegroups.com, phil...@gmail.com, shel...@eecs.berkeley.edu
Thanks for pointing out the undocumented tool in the caffe~~

在 2014年12月6日星期六UTC+8上午9时42分49秒,Evan Shelhamer写道:

Hossein Hasanpour

unread,
Apr 16, 2016, 4:23:05 PM4/16/16
to Caffe Users
Thanks alot, your script is actually way better than the Caffes script.
I just had to run your script and that was all.
Its really hard to get caffe's script to work.
Message has been deleted

Aisha urooj khan

unread,
Aug 6, 2016, 3:53:53 PM8/6/16
to Caffe Users
Thanks yasser, your script saved me from a great hassle.
Reply all
Reply to author
Forward
0 new messages