Validate baseline PPHC ability to approximate underlying DECA coordinates system

Question

Validate baseline PPHC ability to approximate underlying DECA coordinates system

profgrumpy opened this issue 2 years ago · comments

Alessio Gaspar commented 2 years ago

Finish code to simulate student interactions based on a DECA coordinate system.
run experiment on simulated student interaction data to ensure that PPHC is able to approximate the underlying DECA coordinate system.
Update overleaf paper accordingly
Make sure to emphasize the differences in our implementations vis-a-vis the original algorithm; e.g., mutations, other implementation specifics since the code reformulates the algorithm a bit.

Dmytro Vitel · Answer 1 · Wed Jun 29 2022 11:41:30 GMT+0800 (China Standard Time)

First version of synth_deca_axes, synth_deca_spanned was committed. TBD mapping of dimensions to StudentKnowledge

Dmytro Vitel · Answer 2 · Thu Aug 18 2022 01:44:51 GMT+0800 (China Standard Time)

I am closing this issue for now. Experiment was done and paper is on overleaf. We could create separate tickets for other experiments.

Dmytro Vitel · Answer 3 · Sat Feb 18 2023 00:59:42 GMT+0800 (China Standard Time)

Experimentation of PPHC with slightly changed behavior on Pareto non-dominance did not give better results than initial PPHC.
The idea was to move from strategy of leaving parent on non-dominance to strategy of leaving least evaluated individual (basically child). The results are similar: PPHC-1-2-1-3 is not better/slightly better that RAND-3 and worse than RAND-5. Changes are possibly statistically insignificant.
Conclusion: we need to go to different schema how algorithm could "move" towards best knowledge discovery. I reopen this ticket to continue the experimentation.