Deselecting or selecting cohorts in characterization results in page hanging or crashing

jhardin29 commented 5 years ago

Expected behavior

Deselecting or selecting cohorts to display in characterization should result in them being removed or added from the page comparisons without page displaying option to wait or kill page being displayed

Actual behavior

Deselecting or selecting cohorts on characterization page results in page unresponsive and if user waits long enough page will come back or crash

Steps to reproduce behavior

Please see characterization #58 reports for CCAE

johnSamilin commented 5 years ago

Hi @jhardin29 why there's a screenshot of the devtools of the github?

anthonysena commented 5 years ago

@vlbe - My thinking on this issue: The amount of data being returned in our environment for this characterization is causing the page to become unresponsive. Here are some record counts based on the tabular display:

All prevalence covariates: 35,452 rows
Chads2: 1 row
Drug Exposure Any Time Prior: 9,100 rows
Charlson Index: 1 row
Condition Occurrence Any Time Prior: 13,807 rows
Procedure Occurrence Any Time Prior: 13,265 rows

Furthermore, those data points are then plotted on the corresponding d3 plot to the right of the table.

Here is the cohort characterization design: cc_58.txt

I suspect you may have a tough time reproducing the problem using that design since you may not have enough data to use. Instead, you might want to try one that can produce something close to the counts above and then determine what types of fixes we could apply and which should wait until we can do more with the topic in #1320.

anthonysena commented 5 years ago

The back-end changes for this were done as part of #1357 which will address the performance problems when retrieving results. We'll move this item to the next release so that we can address the front-end changes required to toggle the filtering on the back-end.

pbr6cornell commented 5 years ago

Performance when selecting/deselecting cohorts from large characterizations is still extremely slow in our environment. This should be a top priority in ATLAS 2.8 for Characterization.

A few things to consider:

[x] 1) Default to show results from only 1 cohort (rather than default to show all cohorts)
[x] 2) Have a button that is 'select all' and 'deselect all' so that picking cohorts doesn't have to be one at a time.
[x] 3) Allow for a multi-select picker that doesn't try to refresh on each selection (since each one takes so long). Instead, user selects the set of cohorts and set of features they want, then click 'Refresh'. Fixed via #2109
[ ] 4) Don't display all feature reports on one scrolldown page, but organize them in some sort of tabular paging, so a smaller set of data is required to render (e.g. the 'all covariates' plot would be sufficient as 1 default, since you get to see everything without also seeing the specific features).
[x] 5) Decide on some minimal threshold of data to not return, probably needs to be a user input, but could be defaulted to something like 1%, which will greatly reduce the data return. Fixed via OHDSI/WebAPI#880

6) Implement paging on the resultset. _(Moving to Atlas 3.0)__

olga-ganina commented 4 years ago

Hi @anthonysena , we checked and it looks like everything was implemented as a part of #1186 . Could we close this one or anything else is required?

anthonysena commented 4 years ago

@olga-ganina - no, I think there are still some gaps to close based on Patrick's feedback above. Specifically: 1) only showing 1 cohort as a default when opening the results and 2) consider using a tabbed display to cut down on the amount of data required to render. Let's understand the level of effort for these items before implementing to see if there are alternative approaches to consider, especially for item 2.

chrisknoll commented 4 years ago

Just selecting 1 cohort from the design by default should be a quick win. Can someone implement (or just tell me to do it if no one is available).

anthonysena commented 1 year ago

The quick win referenced above https://github.com/OHDSI/Atlas/issues/1142#issuecomment-695903807 may be a good item for 2.14 so moving to that milestone.

chrisknoll commented 1 year ago

The current behavior does only show 1 cohort at first, and will not refresh the entire content as you change selections. The reports refresh on close of the dropdown.

I think this can be closed.

OHDSI / Atlas