Conversation
fflorent
left a comment
There was a problem hiding this comment.
LGTM, thank you!
Waiting for @CamilleLegeron review before merging.
|
PS : regarding grist-custom-widgets-fr-admin/app/geocode/lib.ts Lines 103 to 106 in be680f4 In combination with saving the result score to a column and a manual error checking (that's what the score is there for), I found it adequate to lower the threshold to |
|
@raph-topo A doubtful result is still displayed with a choice banner, right? I don't know how much doubtful results there usually are. I can imagine allowing to customize the threshold to lower it (through an env variable) and let anyone willing to lower this value deploy their own instance. Or an option to let the user change this value ("Open configuration" in the widget menu), but that's more work. |
|
@fflorent Good you mention that… In fact, lowering the threshold applies to recherche globale & recherche spécifique. For the former, responses below the score are not returned. For the latter, it shows available alternatives only below that threshold. So that's sort of opposite behaviour. I've had good results with 40 on old surveys, with it finding reliably way renames like « X rue du Fol Penser » to « X Rue des Folles Pensees ». Agree, it could be helpful to set 2 different scores, and/or set them in the UI (Grist widget config). The |
| }, | ||
| SCORE: { | ||
| name: "score", | ||
| title: "Score (destination)", |
There was a problem hiding this comment.
If we show the user the score, I think we need to explain a little more about where it comes from, and that it is not calculated by the widget but by the Address API, and that it is a reliability score.
Maybe add a description here ?
Or some explanation in app/geocode/Instructions.tsx ?
Only ideas
|
Only on proposal, otherwise LGTM |
No description provided.