Porting MetaLearning to PyTorch #3653

NimishaDey · 2023-11-03T10:27:59Z

Description

Fix #(issue)

Type of change

Please check the option that is related to your PR.

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to not work as expected)
- In this case, we recommend to discuss your modification on GitHub issues before creating the PR
Documentations (modification for documents)

Checklist

deepchem/metalearning/torch_maml.py

rbharath · 2023-11-03T16:30:38Z

deepchem/metalearning/tests/test_maml_pytorch.py

+
+        # Optimize it.
+
+    learner = SineLearner()


This should be in an explicit test_ method

rbharath · 2023-11-15T17:30:48Z

deepchem/metalearning/torch_maml.py

+    task.
+    Example
+    --------
+    import deepchem as dc


This isn't formatted correctly as an example

rbharath · 2023-11-15T17:31:00Z

deepchem/metalearning/torch_maml.py

+
+    def __init__(
+        self,
+        learner,


Add type annotations

rbharath · 2023-11-15T17:32:18Z

deepchem/metalearning/torch_maml.py

+
+    Example
+    --------
+    class SineLearner(dc.metalearning.TorchMetaLearner):


rbharath · 2023-11-20T17:30:05Z

docs/source/api_reference/metalearning.rst

+.. autoclass:: deepchem.metalearning.TorchMetaLearner
+  :members:
+
+ Torch MAML


Nit, space here?

shreyasvinaya · 2023-11-22T09:40:05Z

deepchem/metalearning/tests/test_maml_pytorch.py

+    has_pytorch = False
+
+
+class SineLearner(TorchMetaLearner):


add this to a try block, TorchMetaLearner may not be imported always (might be a cause for the failures of the Jax Tests

Added but now "SineLearner" is not defined.

you can try running without the try block locally to see what might be causing the error, that way you would be able to check if there is some dependency that is causing the error on CI or if there is some bug in the code which is causing the code to skip the definition

shreyasvinaya · 2023-11-22T09:40:41Z

deepchem/metalearning/tests/test_maml_reload_torch.py

+    has_pytorch = False
+
+
+class SineLearner(TorchMetaLearner):


add this to a try block, TorchMetaLearner may not be imported always (might be a cause for the failures of the Jax Tests

shreyasvinaya · 2023-11-27T10:56:53Z

deepchem/metalearning/tests/test_maml_pytorch.py

+    @pytest.mark.torch
+    def test_maml_pytorch(self):
+
+        class SineLearner(dc.metalearning.TorchMetaLearner):


you can try dc.metalearning.torch_maml.TorchMetaLearner if the tests fail again

Yes I tried this. torch_maml is not recognized now in CI.

rbharath · 2023-11-27T17:21:06Z

deepchem/metalearning/__init__.py

 try:
    from deepchem.metalearning.maml import MAML, MetaLearner
-except ModuleNotFoundError:
-    pass
+    from deepchem.metalearning.torch_maml import TorchMAML, TorchMetaLearner


It should just be MAML/MetaLearner.

dc.metalearning.MAML for TF and then dc.metalearning.torch_mamll.MAML for Torch. Then in a later version we swap the default

rbharath · 2023-11-27T17:22:17Z

deepchem/metalearning/tests/test_maml_pytorch.py

+
+try:
+    import torch
+    import torch.nn.functional as F


From deepechem.metalearning.torch_maml import MetaLearner

rbharath · 2023-11-27T17:23:01Z

deepchem/metalearning/tests/test_maml_reload_torch.py

+    import torch
+    import torch.nn.functional as F
+
+    class SineLearner(dc.metalearning.TorchMetaLearner):


Try the same thing as above

rbharath

Let's break this into 2 PRs, one for MetaLearning and one for MAML

rbharath · 2023-11-29T17:20:52Z

deepchem/metalearning/tests/test_maml_pytorch.py

+    has_pytorch = False
+
+
+class TestMAML(unittest.TestCase):


You need the torch marker to prevent this beining.

rbharath · 2023-11-29T17:21:52Z

deepchem/metalearning/tests/test_maml_pytorch.py

+    @pytest.mark.torch
+    def test_maml_pytorch(self):
+
+        class SineLearner(MetaLearner):


Maybe move this out of the test and into the try/catch block above

Jax tests are passing now sir.

ARY2260 · 2023-12-06T15:15:31Z

deepchem/metalearning/tests/test_maml_pytorch.py

+import pytest
+from flaky import flaky
+import unittest
+from deepchem.metalearning.torch_maml import MetaLearner, MAML


Add this line within the try-except block as it is also indirectly importing torch.

Done
Jax tests are now passing.

ARY2260 · 2023-12-06T15:15:49Z

deepchem/metalearning/tests/test_maml_reload_torch.py

+import deepchem as dc
+import numpy as np
+import pytest
+from deepchem.metalearning.torch_maml import MetaLearner, MAML


Add this line within the try-except block as it is also indirectly importing torch.

Yes
Jax tests are passing now.

rbharath · 2023-12-08T22:32:33Z

deepchem/metalearning/torch_maml.py

+    has_pytorch = False
+
+
+class MetaLearner(object):


@NimishaDey Is this a duplicate of the existing MetaLearner class? We shouldn't reimplement the abstract superclass

Sir I have added the 'parameter' method to the existing methods of MetaLearner class. It returns an iterator over the learner parameters. This is needed to construct PyTorch optimizer.

rbharath · 2023-12-08T22:32:54Z

docs/source/api_reference/metalearning.rst

+
+MetaLearner
+----------------
+This is the abstract superclass for metalearning algorithms.


@NimishaDey This is a duplicate of the documentation above. We shouldn't reimplement the abstract superclass. Instead let's fix it as needed

rbharath · 2023-12-08T22:33:09Z

docs/source/api_reference/metalearning.rst

+.. autoclass:: deepchem.metalearning.torch_maml.MetaLearner
+  :members:
+
+MAML


Mention this is the torch implementation and the earlier implementation is the TensorFlow implementation

Let's add the new MAML abstract class to the shared documentation

rbharath

@NimishaDey Getting closer but we still need one more round of fixes. We shouldn't reimplement the abstract superclass in particular

rbharath · 2023-12-13T17:08:13Z

deepchem/metalearning/__init__.py

+        raise NotImplementedError("Subclasses must implement this")
+
+    def parameters(self):
+        """This method is specific to PyTorch implementation.


Abstract classes should not refer to implementation (so no pytorch mention). Talk about the general idea). Also formatting for this string is off.

rbharath · 2023-12-13T17:08:27Z

deepchem/metalearning/__init__.py

+    def parameters(self):
+        """This method is specific to PyTorch implementation.
+        Returns an iterator over the MetaLearner parameters."""
+        print("Subclasses must implement this for PyTorch implementation.")


Raise not implemented error not a print

rbharath · 2023-12-13T17:10:39Z

deepchem/metalearning/torch_maml.py

+    has_pytorch = False
+
+
+class MAML(object):


This is also duplicated with Tensorflow. We should have a shared abstract class instead

Documenting offline discussion, there is currently no abstract MAML superclass so we made a parallel torch implementation to existing TF. We may want to factor out an abstract class later

rbharath · 2023-12-13T17:12:06Z

docs/source/api_reference/metalearning.rst

@@ -1,12 +1,16 @@
 Metalearning
 ============
-One of the hardest challenges in scientific machine learning is lack of access of sufficient data. Sometimes experiments are slow and expensive and there's no easy way to gain access to more data. What do you do then? 
+One of the hardest challenges in scientific machine learning is lack of access of sufficient data. Sometimes experiments are slow and expensive and there's no easy way to gain access to more data. What do you do then?

 This module contains a collection of techniques for doing low data
 learning. "Metalearning" traditionally refers to techniques for
 "learning to learn" but here we take it to mean any technique which
 proves effective for learning with low amounts of data.


Have the abstract classes listed here.

rbharath · 2023-12-20T17:22:57Z

deepchem/metalearning/__init__.py

+        f'Skipped loading some PyTorch models, missing a dependency. {e}')
+
+
+class MetaLearner(object):


This will cause a circular import issue. Move this into its own file metalearner.py and importt it before the other files up top

It's not a circular import because we are not inheriting from it.

rbharath · 2023-12-20T17:23:20Z

deepchem/metalearning/maml.py

@@ -10,60 +10,6 @@
 from deepchem.models.optimizers import Adam, GradientDescent


-class MetaLearner(object):
-    """Model and data to which the MAML algorithm can be applied.


This will break the tensorflow code to break because the import will be gone

Since there is no circular import Tensorflow tests are running and I have the screenshot of it below.

rbharath

Ready to merge once @NimishaDey does final check of tests

NimishaDey · 2023-12-22T10:58:33Z

Ready to merge once @NimishaDey does final check of tests

Both PyTorch and Tensorflow tests are running.

rbharath

LGTM

We may still need to do some refinement but we can do in future PR

rbharath reviewed Nov 3, 2023

View reviewed changes

deepchem/metalearning/torch_maml.py Outdated Show resolved Hide resolved

rbharath reviewed Nov 3, 2023

View reviewed changes

deepchem/metalearning/torch_maml.py Show resolved Hide resolved

rbharath reviewed Nov 3, 2023

View reviewed changes

rbharath reviewed Nov 15, 2023

View reviewed changes

NimishaDey force-pushed the meta_torch branch from d38e587 to dc9ae93 Compare November 20, 2023 17:20

rbharath reviewed Nov 20, 2023

View reviewed changes

shreyasvinaya reviewed Nov 22, 2023

View reviewed changes

NimishaDey force-pushed the meta_torch branch from 73fcf27 to e5ffa0b Compare November 26, 2023 13:50

shreyasvinaya reviewed Nov 27, 2023

View reviewed changes

rbharath reviewed Nov 27, 2023

View reviewed changes

NimishaDey added 14 commits November 29, 2023 16:24

Porting MetaLearning to PyTorch

fe73c33

Adding to docs

ee9279b

Adding usage example and explict test

60b3cac

Adding type annotations and usage example

87a0ecc

Making corrections

9963fba

Resolving errors

d8cd705

Corrections in docs

4f5833a

Including in try block

fdc9026

Unittest

4f3ee1d

Resolving error in CI

5f403db

Correcting Module dependencies

be4fd53

Corrections

c4635a1

Fixes in init

e58f829

Corrections in init

d4ed58b

NimishaDey added 2 commits November 29, 2023 17:34

Corrections for Jax tests

9586963

Corrections

fb6d991

rbharath reviewed Nov 29, 2023

View reviewed changes

NimishaDey added 2 commits December 2, 2023 11:06

Corrections in type annotations

e3ea2f6

Correcting type annotations

6be2150

ARY2260 reviewed Dec 6, 2023

View reviewed changes

Correcting import statements

aafd954

NimishaDey marked this pull request as ready for review December 7, 2023 12:21

rbharath reviewed Dec 8, 2023

View reviewed changes

NimishaDey added 5 commits December 10, 2023 21:06

Correcting docs

b9bc141

Adding method to MetaLearner class

4145268

Corrections to import statements

8686f0a

Changing import statements

32a8a36

Corrections

14bede8

rbharath reviewed Dec 13, 2023

View reviewed changes

Corrections in abstract class and docs

3a52c3c

rbharath reviewed Dec 20, 2023

View reviewed changes

rbharath approved these changes Dec 26, 2023

View reviewed changes

rbharath merged commit 545540a into deepchem:master Dec 26, 2023
23 of 33 checks passed

		f'Skipped loading some PyTorch models, missing a dependency. {e}')


		class MetaLearner(object):

Porting MetaLearning to PyTorch #3653

Porting MetaLearning to PyTorch #3653

Conversation

NimishaDey commented Nov 3, 2023

Description

Type of change

Checklist

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rbharath left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rbharath left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rbharath left a comment

Choose a reason for hiding this comment

NimishaDey commented Dec 22, 2023

rbharath left a comment

Choose a reason for hiding this comment