Confidence level of candidates

[subhash-velou]

I would like to know if there is a way to access confidence level/score for each candidate. For example, in the output below, can we know that the model is 90% confident it's "table", 4% confident its "tale" etc.?
{ "results": [ { "candidates": [ "table", "tale", "tacle", "tadle", "ladle", "tadre" ], "len": 5, "pos_from": 0 } ] }

I'm having a requirement where I need to apply the fix with top candidate only if the confidence level is above 90%. Would like to understand if you already have this facility or any plans to implement in future.

[hejazizo]

hey. Any luck with this? I exactly need this as sometimes the model makes mistake on task specific texts. Is this going to be implemented or is already implemented?

[hejazizo]

Anybody here?!!!

[jackneil]

i've created a mod that can output JSON with scores assigned to each candidate. Can't open a pull request here to push it though.

[bakwc]

why not? will be glad to accept it

[jackneil]

sorry i was trying to create a pull request on master and it was greyed out. need to branch first?

[bakwc]

yep

[jackneil]

ok i just branched locally to confidenceIntervals but when trying to publish the branch is says permission denied. using vscode.

[bakwc]

you need to publish it to local repository and then create pull request

[rprilepskiy]

@jackneil, how is it going with confidenceIntervals feature? Could you create pull request?
Do you need any help?

[jackneil]

Yes we got it implemented and functional. I'll see if the dev on that project can get it pulled and merged. Best

…

-jack Jack Neil CEO / Hank AI, Inc. jack.neil@hank.ai (w) +1.828.490.1203 (m) +1.803.466.7551 Sent via Superhuman ( https://sprh.mn/?vip=jack.neil@hank.ai )

On Mon, Jan 27, 2020 at 8:40 AM, Roman Prilepskiy < ***@***.*** > wrote: @ jackneil ( https://github.com/jackneil ) , how is it going with confidenceIntervals feature? Could you create pull request? Do you need any help? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub ( #53?email_source=notifications&email_token=AGFRM5UBXSCKM5LZJWYYGILQ73P3NA5CNFSM4FUTGMO2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJ7QTTA#issuecomment-578750924 ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AGFRM5XWDQJSDGNGYC5IOFLQ73P3NANCNFSM4FUTGMOQ ).