Filtering improvements #182
Conversation
|
Recommendation: Show the "All Labels" as unchecked or in the ambiguous state if any of the labels are unchecked. I think. it is more likely you want to go back to showing all labels than showing no labels. |
|
I'm having trouble getting the filter by comparison section to do what I expect. I want to show where predictions differ from labels (or labels differ from predictions). I can't seem to pick labels on the left. When I pick predictions and differs from, picking labels on the right seems to go back to saying any. I'm not sure what is happening: filterby.mp4 |
bnmajor
force-pushed
the
filtering-improvements
branch
from
028d720 to
49fcd10
Compare
Updated!
This has been fixed now. |
|
Either there is still an issue or I am confused about what is supposed to be happening. See video. The left side of filter by only lets me pick predictions (not labels). Filter by Labels by me gets different results than filtering by labels (even though I'm the only who has labeled something); and picking labels on the right then shows "Any". filterby.mp4 |
- Fix comparison operator to ensure we're comparing user ids - Do not disable the "any" option for second comparison choice - Fix clashing component ids
- Adds the option to group by prediction or by review. - Removes the "agree/disagree" grouping option as it's too ambiguous
- Add the options to sort by predictions or reviews - Remove the option to sort by agree/disagree because of ambiguity
Add "All Labels" option to the labels filter menu. Checking the "All Labels" option selects all labels. Unchecking removes all label filters. Labels can still be checked/unchecked individually. The "no label" filter remains its own option.
This adds "All Reviews" to the reviews filter menu to mirror the setup of the labels filter menu.
For the comparison filters make sure that we confirm that the first option and operator have been set before trying to update any filtering.
- Improve naming for added clarity - Add clarifying comments - Update metadata on queued superpixels to ensure none are missed
Only disable an option if it is the exact same option as the one already selected. Disable the second selection until the first selection and operator are chosen.
bnmajor
force-pushed
the
filtering-improvements
branch
from
49fcd10 to
412427e
Compare
Add a utility function to check whether or not a value is a valid number: not NaN, not null, and not undefined.
|
I think we need the same fix for selecting option items rather than text for the differs from selector. Also, we should probably make the entire row of every filter box toggle the checkbox (e.g., the All Labels field should be clickable anywhere, not just on its text). |
|
My test data set apparently has some superpixels without any "labeller" associated with them. I don't know if this is the cause, but I still have an issue when I select the Predictions differ from Admin (it continues to show differs from Any). If this set should be blank (or, in general, if we have filtered everything away), we should put some indicator that there is nothing selected rather than just leave the field blank. |
Do not try to compute the number of superpixels to show unless there is at least one available to fetch the size from.
This adds a tiny bit more margin so that selected superpixel highlight is not covered up.
This allows finding ALL matching/differing labels and reviews in addition to user labels and user reviews.
This re-write fixes some bugs that were present as well as improves readability and adds clarifying comments.
Sets a default return value in case no fill color is found. This prevents a console error from a race condition when the grouping selection is changed but internally the actual grouping has not changed - this request a category fill color for "data" which is not actually one of the user created labels.
If the component is never mounted the watcher will never be reached.
- Do not call the annotation metadata endpoint if the annotation was saved. Saving the annotation saves the metadata so this is an extra call for no reason. Only explictly save the metadata for review changes since those do not trigger an annotation save. - Include a separate endpoint that is used to double-check that any labels that are created are not missing their associated "labeler" (user) information. - Since we are now relying on the annotation save as a way to update the metadata as well, make sure calls to updateMetadata are made *before* saveAnnotations. - Cleans up a few places where the new label value being passed in to be saved as metadata was incorrect.
- Duplicate values in filterBy list were breaking the labels and reviews menus because selections were not being removed as expected - Do not include superpixels with cleared reviews when filtering by reviewer - Fix typo that prevented filterByComparison menus from displaying correct selection - Fix filterByComaprison's conversion of value to label to make sure comparison values are correct
bnmajor
force-pushed
the
filtering-improvements
branch
from
56da328 to
149cea5
Compare
This function was originally added to make sure that any labels that were missing their labeler user id were fixed before the metadata was saved. After all of the fixes we've now made this should no longer be needed.
groupByoptionsagree/disagreegrouping option as it's too ambiguoussortByoptionsagree/disagreebecause of ambiguityreviewsfilter menu to better match thelabelsfilter menu. This adds "All Reviews" to the reviews filter menu to mirror the setup of the labels filter menu.