Request: Negative confidence

[andrew-hill]

Please allow the confidence slider to go from -100% to +100% rather than just 0-100%

This would allow things like

• simpler rules to distinguish between otherwise similar contexts
• disable a particular context under some circumstances (e.g. disable when on battery power)
• hysteresis between specific modes (e.g. if mode A, prevent mode B, and if mode B prevent mode A—if both are appropriate, a manual selection of either one will prevent the other, without enabling/disabling sticky contexts or affecting any other context matches)

[andrew-hill]

Have looked a bit into the maths in CPController.m and it'd actually require a change there, because a large negative confidence wouldn't actually do much to negate the confidence. e.g.:

• matching features +30% +70% gives +79%
• adding a -90% feature to negate would give 60% confidence

One option would be to process +ve confidences first (the current way) then multiply any negative confidences in sequence

• Currently the formula is along the lines of P = 1-(1-a)(1-b)(1-c)... where P is the context confidence, and a,b,c,... are confidences of each matching feature
• I'd change this to P = [1-(1-a)(1-b)(1-c)...] x (1-A)(1+B)...] where A,B,... are the negative confidence features (a,b,c,... are the positive confidence features).
• the 30, 70, -90 example above would instead give 8% confidence in this case
• and this wouldn't otherwise change the calculated confidence values

I've got some other ideas, but I think generally they would mean a change in the maths that might break people's current rules, which could be bad (certainly if there's no easy way to translate those rules into equivalent new ones).

[dustinrue]

I appreciate the suggestion but I'm going to pass on adding negative confidence. The confidence system is already confusing for some users as it is and adding negative confidence would further complicate it.