Where to find log files?

6,192 views
Skip to first unread message

James Guo

unread,
Dec 16, 2014, 9:45:44 AM12/16/14
to caffe...@googlegroups.com
Hello guys,

I'm currently using Caffe to train CNN, after training I want to check training and validation loss, which were written into LOG(INFO). I've noticed that Caffe use Google glog library to log information but I couldn't find these files. Anyone can tell me where to find these logging files or how to turn on logging? Thank you very much!


Best,

James

Manuel Lopez Antequera

unread,
Dec 16, 2014, 10:11:41 AM12/16/14
to caffe...@googlegroups.com
Unless otherwise specified, glog writes to the filename "/tmp/<program name>.<hostname>.<user name>.log.<severity level>.<date>.<time>.<pid>" (e.g., "/tmp/hello_world.example.com.hamaji.log.INFO.20080709-222411.10474"). By default, glog copies the log messages of severity level ERROR or FATAL to standard error (stderr) in addition to log files.

James Guo

unread,
Dec 16, 2014, 10:21:13 AM12/16/14
to caffe...@googlegroups.com
Thanks Manuel, but still, I didn't see any tmp folder after I finish training with Caffe. So maybe Caffe has its own folder to store log file or has some specific flag to use?

Manuel Lopez Antequera

unread,
Dec 16, 2014, 11:05:07 AM12/16/14
to caffe...@googlegroups.com
You could call the caffe executable with the --log_dir=yourdesiredpath as well.

James Guo

unread,
Dec 16, 2014, 11:07:13 AM12/16/14
to caffe...@googlegroups.com
Problem solved by add "2>&1 | tee your_name.log" at the end of command, i.e. "
./build/tools/caffe train --solver=/path/to/my_solver.prototxt 2>&1 | tee your_name.log"

James Guo

unread,
Dec 16, 2014, 11:12:02 AM12/16/14
to caffe...@googlegroups.com
Problem solved, thanks a lot!!

Comma

unread,
Sep 10, 2015, 1:33:55 PM9/10/15
to Caffe Users
Very Helpful. thank you very much!

在 2014年12月16日星期二 UTC-5上午11:07:13,James Guo写道:

Zhao.Kai

unread,
Oct 13, 2015, 4:58:34 AM10/13/15
to Caffe Users
Hi:

But how to specify the log file DIR in python interface ?
thanks in advance !

James Guo

unread,
Oct 13, 2015, 5:15:37 AM10/13/15
to Caffe Users
You can call python subprocess module, something like this

subprocess.call(".build/tools/caffe train You/solver/path 2>&1 | tee You/log/path", shell=True)

Marianne

unread,
Apr 12, 2017, 10:04:57 AM4/12/17
to Caffe Users
When resuming training the log file is overwritten and information about the former training progress is lost.
You may want to use
... | tee -a your_name.log
instead.
Reply all
Reply to author
Forward
0 new messages