Differences

This shows you the differences between two versions of the page.

--- flash_ai_parameter_optimization [2022/09/24 00:15] – created jhagstrand
+++ flash_ai_parameter_optimization [2023/01/12 08:58] (current) – removed jhagstrand
@@ Line 1: / Line 1: @@
-====== Flash AI Parameter Optimization ======
-The algorithm for manipulating the stacks of flash cards is implemented in coach.js.
-The algorithm is based on parameters.
-At present, we set the parameters manually.
-We should be able to use the autoplay feature, repetitively altering the parameters striving for optimal success in mastering the cards.
-We should also introduce parameters describing the human.
-  * number of cards in working set
-  * average number of attempts to learn a new cards
-Set up a meta-learning AI who repetitively sets up decks and parameter sets and runs autoplay, saving the result of each run, gradually improving the parameter set.
-class Train
-  * set parameters
-  * run with autoplay
-  * record result: number of attempts to master all cards
-  * loop
-    * adjust parameters
-    * run with autoplay, max 10000 runaway escape
-    * record result
-adjust parameters
-  * a weight multiplied times each parameter
-  * vary the weight up, see the effect on result
-  * vary the weight down, see the effect on result