Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

example time_sequence_prediction does not produce expected output #247

Open
VeniVidiGavi opened this issue Oct 27, 2017 · 5 comments · May be fixed by #1191
Open

example time_sequence_prediction does not produce expected output #247

VeniVidiGavi opened this issue Oct 27, 2017 · 5 comments · May be fixed by #1191

Comments

@VeniVidiGavi
Copy link

https://github.com/pytorch/examples/tree/master/time_sequence_prediction

After a few steps, the loss suddenly jumps up and stays high:
loss: 0.00101216329969
loss: 0.195070432582

the pdf0 is OK, pdf1 is OK but from pdf2 the graph becomes an "L "

@philippslang
Copy link

#243

@barnabytprowe
Copy link

Yup, see #243

@rallen10
Copy link

FYI: 4 years later and I'm having the same problem with time_sequence_prediction not being able to reproduce the expected results. My loss explodes after the 4th or 5th epoch. Changing the output layer to Linear, as is posed as the solution to this problem in #250, does not seem to fix this.

I'm guessing it's still a problem with learning rate being too high, but as the example doesn't work as it is currently written.

@subramen
Copy link
Contributor

@rallen10 Thanks for raising this. I agree, this example's learning rate is (way) too high and the exploding loss resolves by setting it to something like 0.001.

I'll leave this open for a "good first issue"; along with fixing the learning rate, this example could also use a refactor ;)

@anikethc
Copy link

anikethc commented Sep 28, 2023

Hi @subramen and @msaroufim ,
I've made the changes and created a pull request. Could one of you please review it?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
6 participants