Fix local auto labeling #600
Conversation
| log_to_field: Optional[str] = None, | ||
| is_prediction: Optional[bool] = True | ||
| ) -> Dict[str, Any]: |
There was a problem hiding this comment.
Having a is_prediction kwarg on a predict_with_... function is IMO really confusing API design.
I won't put my foot down and I will push this through for whatever you need right now, but we will probably need to change this in the near future if you want people to actually be using this.
Either way, thanks for adding the docstring
| for prediction, remote_path in zip( | ||
| predict_fn(local_paths), | ||
| [predict_fn(local_paths)], | ||
| [result.path for result in self[idx * batch_size : (idx + 1) * batch_size]], |
There was a problem hiding this comment.
[blocker] This doesn't look right.
The signature of CustomPredictor (predict_fn) is:
CustomPredictor = Callable[
[List[str]],
List[Tuple[Any, Optional[float]]],
]So the function should be accepting a list of datapoints, and return a list of prediction tuples.
Why are you wrapping the result in another list? This zip will end up processing all of the returned predictions for one single path.
There was a problem hiding this comment.
How does the return from the prediction function you're using right now look like?
Fixed an issue where the auto-labeling expected format for local and dockerized instances was different.
Need to update the docs active learning page