Made Adapters Sliceable #285
Conversation
…ed to define set item, might be nice to list indecies of transforms in the default print
… transform was not added to the list for things to imported in the adapter file so added that
|
Thank you! I will take a look soon! |
|
A few short questions:
|
|
Essentially what i meant with my last comment is that you can add a transform to the end of the adapter by slicing. |
|
Ah, I see. Could you show a code example? Let me know once you have added everything you wanted to add to this PR. |
|
Appending via slicing is fine, imo. We should have checks in place and error if slices are assigned non-Adapter objects (thus making the adapter self-recursive, each slice is an Adapter itself). That said, best practice should be to use add_transform (could just call this append as well). We could add an insert method to mimic list behavior, too. |
bayesflow/adapters/adapter.py
Outdated
| @@ -76,7 +76,68 @@ def __call__(self, data: dict[str, any], *, inverse: bool = False, **kwargs) -> | |||
| return self.forward(data, **kwargs) | |||
|
|
|||
| def __repr__(self): | |||
| return f"Adapter([{' -> '.join(map(repr, self.transforms))}])" | |||
| str_transf = '' | |||
There was a problem hiding this comment.
Use double quotes for strings to conform to our style guide.
bayesflow/adapters/adapter.py
Outdated
| def __getitem__(self, index): | ||
| if isinstance(index, slice): | ||
| if index.start > index.stop: | ||
| raise IndexError("Index slice must be positive integers such that a < b for adapter[a:b]") |
There was a problem hiding this comment.
Should just return an empty adapter.
bayesflow/adapters/adapter.py
Outdated
| if isinstance(index, slice): | ||
| if index.start > index.stop: | ||
| raise IndexError("Index slice must be positive integers such that a < b for adapter[a:b]") | ||
| if index.stop < len(self.transforms): |
There was a problem hiding this comment.
I don't think this works for index.stop = None, which happens when you call adapter[start:]
Best would be to just mimic list behavior via returning Adapter(self.transforms[index])
bayesflow/adapters/adapter.py
Outdated
| new_adapter = Adapter(transforms=sliced_transforms) | ||
| return new_adapter | ||
| else: | ||
| raise IndexError("Index slice out of range") |
There was a problem hiding this comment.
We can let the internal list handle this
bayesflow/adapters/adapter.py
Outdated
| raise IndexError("Adapter index out of range.") | ||
| sliced_transforms = self.transforms[index] | ||
| new_adapter = Adapter(transforms=sliced_transforms) | ||
| return new_adapter |
There was a problem hiding this comment.
Same idea here, let the internal list handle edge cases
| else: | ||
| raise TypeError("Invalid index type. Must be int or slice.") | ||
|
|
||
| def __setitem__(self, index, new_value): |
There was a problem hiding this comment.
Seems a bit verbose, does the following snippet not work?
self.transforms[index] = new_value.transforms[:]Or for an integer index:
self.transforms[index:index + 1] = new_value.transforms[:]…ts that I had previously written
I made the adapters sliceable but one concern I have is that when I ask for
generic_adapter[0:2]this will return an entirely new adapter instance. Depending on the use case this one matter but could cause problems. I also made a test file but it's not tracked by github, please let me know if I should also upload that.I also noticed that ElementwiseTransform wasn't being imported so added that as well.
I thought it might also be nice to add the indices of the transforms to the default adapter print, but didn't add that particular feature. I also thought it might be nice to add the feature of appending a new transform via slice assignment so if
index == len(adapter.transforms)then simply use theadd_transformmethod.Resolves #274