Paths and file management #61
Conversation
need to adapt plot and analyze.
refine some more.
things are looking much better imho
|
This probably seems more complicated than it is. The bulk of the changes are moving blocks of arguments around into dataclasses. |
|
Actually hold on just discovered a bug |
|
Okay seems to be fine now |
| fig, axs = plt.subplots(num_transforms, 1, figsize=(20, 5 * num_transforms)) | ||
| if num_transforms == 1: | ||
| axs = [axs] | ||
|
|
||
| transform_index = 0 | ||
|
|
||
| def make_image_grid(arrays: List[FloatArray], nrow: int) -> FloatArray: |
There was a problem hiding this comment.
from looking at it, I think this function might be worth moving out of the local scope (but perhaps private to the module)?
Or do you think it's only of value within this method?
| estimates: List[Tuple[FloatArray, int]] | ||
|
|
||
|
|
||
| @dataclass |
There was a problem hiding this comment.
In general, I like the approach of typing this more strictly here
But do we need this 'ReconstructionStatistics'?
I feel like it would be enough to have a tuple returned / as a parameter where needed 🤔
| @@ -34,7 +102,7 @@ def transform_base_images( | |||
| src, _ = base_dataset[np.random.randint(base_len)] | |||
| images.append(src) | |||
|
|
|||
| results: Dict[str, Dict[str, Dict[float, List[Tensor]]]] = { | |||
| results: Dict[str, Dict[str, Dict[float, List[FloatArray]]]] = { | |||
There was a problem hiding this comment.
Why not use TransformStatistics already here?
Defining this results dict first this way makes making changes to this later harder, as also the TransformStatistics type will then have to be adjusted (same in other functions)
There was a problem hiding this comment.
really like the move towards a more traceable config system :)
So this does quite a fair bit of refactoring.
statistics.pyto return dataclass instances to make things more manageable and readable.I definitely wasn't perfectly complete or consistent with how I did this, but I think things are significantly better organized than they were before. Ideally this provides a good reference for how to factor/refactor such code in other parts of the library. Definitely open to feedback on review.x
Resolves #43