AlMeta:YTDNM reinforcement learning

From AlMeta
Revision as of 10:55, 28 June 2022 by Cmdrtako (talk | contribs) (Created page with "Your the Dog Now Ma(n|chine) ---- #if (you got a reward) ## if reward is > 0 ###Treat =true ###add top of @History to @Trick ##else (if reward < 0) ###Treat = False ###clear...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Your the Dog Now Ma(n|chine)


  1. if (you got a reward)
    1. if reward is > 0
      1. Treat =true
      2. add top of @History to @Trick
    2. else (if reward < 0)
      1. Treat = False
      2. clear @Trick
    3. for X in @History
      1. update X in NN
      2. if (X was Conditioned)
        1. Reward = Reward minus( 1 minus likelihood of choosing X)
      3. if(X was not Conditioned)
        1. Reward = Reward times 0.5
    4. clear @History
    5. if (you are not sure what to do)
      1. if (Treat)
        1. if (second of @Trick not Conditioned)
          1. add @Trick to @Repertoire
      2. add new choice to @History as not Conditioned
    6. else (if you are sure what to do)
      1. add new choice to @History as Conditioned
  2. else (if you didn't get a reward)
    1. if (your in not vary sure of what do)
      1. if (you last action was Conditioned)
        1. reward = -1
        2. while(top of @History was Conditioned)
          1. pop top of @History as X
          2. Update X in NN
          3. reward = reward times likelihood of X
        3. Clear @History
      2. Clear @Trick
      3. add new choice to @History as not Conditioned
    2. else (if you are sure of what to do)
      1. if you last action was Conditioned)
        1. add top of @History to @Trick
      2. add new choice to @History as Conditioned.