search

changun / CollMetric

A Tensorflow implementation of Collaborative Metric Learning (CML)
GNU General Public License v3.0
160 stars 61 forks source link

Validation recall is not improving during training #13

Closed goohanjun closed 5 years ago

goohanjun commented 6 years ago

I just ran a baseline code by typing python CML.py I just get validation recall 0.001952619578277473 every epoch. Although training loss is slightly decreasing, I cannot see any changes in validation recall. Is the baseline code wrong? How can i modify it to get the right result?

changun commented 6 years ago

Hi Hanjun,

I think it is because the optimizer is not working? Try run it on a GPU (the tensorflow-gpu somehow has more stable optimizer) or change the optimizer to more basic GradientDescent optimizer?

On Wed, Aug 8, 2018 at 12:53 AM HanjunKoo notifications@github.com wrote:

I just ran a baseline code by typing python CML.py I just get validation recall 0.001952619578277473 every epoch. Although training loss is slightly decreasing, I cannot see any changes in validation recall. Is the baseline code wrong? How can i modify it to get the right result?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/changun/CollMetric/issues/13, or mute the thread https://github.com/notifications/unsubscribe-auth/AA1Ff_YcvbaB2mJ9k64F2MjpvFVTIj_Gks5uOpjsgaJpZM4Vzcmq .

-- Cheng-Kang (Andy) Hsieh UCLA Computer Science Ph.D. Student M: (310) 990-4297

ObiJiang commented 5 years ago

Same problem. Tried both modifications. Still not working.

changun commented 5 years ago

Hmm let me see if I can fix it. Are you running the code on the master branch?

On Wed, Nov 7, 2018 at 4:42 PM ObiJiang notifications@github.com wrote:

Same problem. Tried both modifications. Still not working.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/changun/CollMetric/issues/13#issuecomment-436831184, or mute the thread https://github.com/notifications/unsubscribe-auth/AA1Ff0Jmbx6i5tgMx5fNFHbpXl2ZSeitks5us33hgaJpZM4Vzcmq .

-- Cheng-Kang Hsieh UCLA Computer Science PhD Student M: (310) 800-2321 A: 740 Weyburn Terrace Apt. 55, Los Angeles, CA 90024

ObiJiang commented 5 years ago

Sorry. Forget to update. I tried Adam optimizer. And now it's giving me ~24% recall rate.

changun commented 5 years ago

i see. thanks for the feedback. Let me make Adam as the default.

On Fri, Nov 9, 2018 at 7:08 PM ObiJiang notifications@github.com wrote:

Sorry. Forget to update. I tried Adam optimizer. And now it's giving me ~24% recall rate.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/changun/CollMetric/issues/13#issuecomment-437553730, or mute the thread https://github.com/notifications/unsubscribe-auth/AA1Ff4NPvVXc76Nb3oMGDu_4S01LPxpBks5utkNCgaJpZM4Vzcmq .

-- Cheng-Kang Hsieh UCLA Computer Science PhD Student M: (310) 800-2321 A: 740 Weyburn Terrace Apt. 55, Los Angeles, CA 90024