-
-
Notifications
You must be signed in to change notification settings - Fork 41
Conversation
@foldfelis should I update the readme and docs as well? |
Hi @Abhishek-1Bhatt it would be great to have the docs and readme updated 😊 |
An import of Random was placed, although there was no use for it. Since it isn't in the deps, it would be best to remove it from here for now, so that the tests don't fail.
Great 😊, I'll start working on it. |
Codecov Report
@@ Coverage Diff @@
## master #39 +/- ##
===========================================
- Coverage 100.00% 91.89% -8.11%
===========================================
Files 2 5 +3
Lines 52 74 +22
===========================================
+ Hits 52 68 +16
- Misses 0 6 +6
Continue to review full report at Codecov.
|
Hi @Abhishek-1Bhatt I didn't notice that there is a model training task inside the test 🤣 . I was wondering why the CI kept running for an hour. Is there another way to test DeepONet? |
Oh😅, this was intended to be a regression test, just to sanity check that our model's training does not get deteriorated if we try out something new in the future. Indeed it doesn't make sense to train for 400 epochs here, I see that now : ) . One thing we can do is to reduce the no. of epochs to 100 or even lesser and then check against the error bound or would you like to have a different way of testing for it? @ChrisRackauckas might also have some suggestions regarding regression testing here. |
Usually, I only do the unit test and put the training tasks into the example folder. I will manually test the examples locally to make sure that things won't go wrong. I am looking for a more efficient way to test the robustness of the models as well. |
That seems like a nice way to do it, we can transfer it to example/Burgers |
Feel free to add new examples and I will reorganize and refactor the project once the new featur is merged. |
Removed an import of MAT.jl, it was causing the tests to fail😅 |
Looks good to me 🎉 |
Resolves #37. I have added the files for DeepONet implementation from OperatorLearning.jl and added tests for the architecture.