In T304876 we created an R function to calculate Pageview data loss.
Currently, we can filter and calculate data loss for filters like country_code, access_method, referer_class
We want to use this function for adding data loss estimates in the Readers metrics charts
De-scoped
We want to be able to get the data loss estimates by new dimensions such as-- project
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | Mayakp.wiki | T311560 Pageview Data loss | |||
| Declined | None | T305990 Additional features in Pageview Loss calculation |
This is going to pretty easy – just changing the queries and re-querying the data. The actual function is flexible enough to accommodate the additional dimension without problems.
Added scope to this task, since @mpopov is already working on it.
Update: this needs re-scoped because adding the project into the mix makes the data much, much bigger.
Should it just be the en6 countries since that's the intended application? Should it be some projects (perhaps just the wikipedias?) or just enwiki?
We want to use this function for adding data loss estimates in the Readers metrics charts
Regarding increasing the scope to package it all up for use with metrics charts… let's not? There's currently the one chart and we already have a working solution. Until there are more charts and a clear need for a different solution, let's dial back the scope here.
@mpopov :
It would be useful to keep this in the Current quarter and discuss further. @kzimmerman and I have been talking about what it would mean to 'close out pageview dataloss' and I feel this package could be a potential deliverable because we will have to keep referring to the dataloss in any future pageview analyses. I also see a need to include the dataloss on other charts like referrer class and access method.
I think this should wait until T314197: Data Loss Estimation Recommendations and T323181: Research Recommendations for Data Loss Estimation are resolved.
Declining this task, as our emphasis has shifted to unique devices, and the amount of work needed would be high.