Ideology
Old day people study probability forcus on sample distribute. [fix parameters]
Bayes probability study forcus on parameters.[fix samples]
1 | prior distribution + sample info -> Posterior distribution |
the new experiment result will refresh people cognition on old things.
It’s fit people cognition law on nation explor method.
formula
First:
the probability event A, event B happen at the same time
, equals the probability A happen multiply with the probability
B happen when A happen.
So we have:
=>
=>
a new knowledge = a priori knowledge * a new factor affect knowledge / a new factor
We use sample experiment’s phenomenon verify rules.
Use adjust factor
to adust parameters, make rules more accurate.
Example
We need to check a rule:
watermelon sugar content > 10, good probability is 70%
- prior condition is good watermelon probability is 20%
- exp:bad watermelon has probability 25%, sugar content > 10
- exp:good watermelon has probability 99%, sugar content > 10
so we have :
Now we found the threshold is not right, base on adjust factor, we change rule:
watermelon sugar content > 15, good probability is 70%
- prior condition is good watermelon probability is 20%
- exp:bad watermelon has probability 15%, sugar content > 15
- exp:good watermelon has probability 90%, sugar content > 15
so we have :
Keep on adjust:
watermelon sugar content > 20, good probability is 70%
- prior condition is good watermelon probability is 20%
- exp:bad watermelon has probability 9%, sugar content > 20
- exp:good watermelon has probability 85%, sugar content > 20