Comments (5)
Thanks for the suggestion. Indeed, epochs
could be moved to the fit method, as you might want to call fit
multiple times with different epoch numbers.
However, I would not move the other arguments to the fit
method, as they are values inherently related to the model instance, in the sense that you don't want to change them between fit
calls.
Can we restrict this issue to epochs
only?
Opinions @ManuelAlvarezC @leix28 ?
from tgan.
Hi, I opened this from #22, where @ManuelAlvarezC asked me to create a new issue.
I agree that batch_size is not the best example, cause it changes the model. However, I do think there are more parameters than just epochs that are best suited in the fit
call, take for example learning rate, batches per epoch, and whether to save and/or restore checkpoints. I think that might also be all of them. Thoughts?
from tgan.
Hi, I opened this from #22, where @ManuelAlvarezC asked me to create a new issue.
Yes, I saw it! It was just to comment a bit and enrich the discussion.
I agree that we could put more things, not only epochs. But the things we put should be only those that you might want to change. Perhaps we can list candidates here?
About the ones you said, the batches per epoch and the and the learning rate are things that might change the overall convergence of the model, so I don't see them as things that you change in the middle of a fitting process. You start with one value and keep it until the end.
Saving and restoring could be there, I see no harm on that, but I do not see why this would be useful either, especially restore. You tell TGAN to either start from scratch or recover from a checkpoint when you create the TGAN instance, and from there, if you call fit multiple times you should always restore and continue from the previous call.
If you want to start from scratch again, you just create a new instance, as you want it to be completely unrelated to the previous one.
from tgan.
Hi,
Thanks for open the issue @Baukebrenninkmeijer.
About the subject of discussion:
I think that moving max_epochs
and steps_per_epoch
to the fit method could make sense, however, I have to agree with @csala about restore_session
and save_checkpoints
that even if it can be done it doesn't seem specially useful.
Any other candidate to be moved to .fit()
?
from tgan.
Hmm, I would like to add my perspective on not including the restore_checkpoints
, lr
and save_checkpoints
. I see your point on them, which makes sense in classic development methods, but as someone who works almost exclusively in notebooks, having the .fit
call have the flexibility to experiment with different settings at run time is very useful. Most notably, the lr
parameters is very useful to quickly try out multiple different learning rates.
from tgan.
Related Issues (20)
- How to save model during experiment?
- How to deal with discrete (integer) variables? HOT 1
- Random search crash at the end? due to NaN ? on classifier.fit(..) HOT 1
- vecs_denorm in model.py
- Handling the target variable in data generation (question)
- RuntimeError: dictionary changed size during iteration HOT 1
- tarfile.ReadError: unexpected end of data
- How to pass TGANModel hyper parameters as a dictionary
- I have a problem when I try to use 'Random hyperparameter search' HOT 1
- Fixing random seed to reproduce the same results HOT 2
- Issue with tgan installation in mac HOT 1
- AttributeError: can't set attribute HOT 1
- Error on changing parameters regarding discriminator and generator
- Usage of discriminator as classifier
- creating samples is stuck HOT 2
- How to work with GPU ?
- Increase code style lint
- self.training = training assignment is failing HOT 2
- importing TGAN HOT 2
- HELP ModuleNotFoundError (maybe version issue) HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from tgan.