2025.04.02 meeting notes

Recap

• quick recap of bug fixes:
  • train eval plot with and without market bound now works
  • optimal strat value now is correct
  • action space now relaxed to [-1_000, 1_000]
• show market bound plot
• show sharpe ratio on train eval plot
• show grid plot of fbm opt strats
• show recreated sharpe ratio plot from Rasonyi Nika paper
• show experiments with T being bigger than num_steps

TODO

• show sharpe ratio of the optimal strategy on the train eval plot
• do the same as above but split the plot into two instead of a new y scale
• show the asimptotic optimal strategy on the market bound plot
• do an experiment where we train on H = 0.1 with increasing T and we keep ratio of T / num_steps constant
• show side by side of H0.1 T1024 with and without clipped action space, that is [-10, 10] and [-1_000, 1_000]
• szeretnenk latni a hisztogramjat az actionoknek
• if the training works for small T then we can try to incrementally extend the time horizon

Remaining from last week

• when running parallell envs try to shuffle the starting point to cover more of the episode length
• pump up the num_steps numbers
• train model on $T=128$ with $H=0.45,0.55$ with big num_steps_eval number (5_000)
• fix training for $H>0.5$
  • Liquidation test env continue
  • make a grid search where there are three axis: clip_coef, learning_rate, H = 0.1, 0.7
  • make an env that is a child of FBMEnv and has a price process of a linear function with random slope