Feat/tgan#85

Open

marcell-mate wants to merge 64 commits intodevfrom

Commits on Dec 23, 2020

started to implement rl env and extended cryptodataset for it
danielhomola
committed
modifying the definition of p1 due to experiments in the rl notebook
danielhomola
committed
flashing out the runner and env.py a bit more
danielhomola
committed

Commits on Dec 30, 2020

now we have all the components in place and I know what to do, now let's make this work
danielhomola
committed
going through environment.py and trying to make it work
danielhomola
committed

Commits on Dec 31, 2020

working on environment
danielhomola
committed

Commits on Jan 4, 2021

ppo now trains and loss doesn't explode (nearly as often)
danielhomola
committed

Commits on Jan 7, 2021

Commits on Jan 8, 2021

init
marcell-mate
committed

Commits on Jan 9, 2021

fixing inverse linear implementation, changing past weights to past dirichlet concentrations in what we pass to the network along with the state
danielhomola
committed

Commits on Jan 11, 2021

major rewrite of the experience collection to make it parallel - doesn't work yet
danielhomola
committed

Commits on Jan 12, 2021

Commits on Jan 13, 2021

Commits on Jan 14, 2021

on track
marcell-mate
committed
disc
marcell-mate
committed

Commits on Jan 15, 2021

Commits on Jan 16, 2021

Commits on Jan 17, 2021

now it works reliably both on win/linux! hallefuckingluja
danielhomola
committed

Commits on Jan 20, 2021

adding logging of weights/portfolio and making few fixes to how actions are sampled
danielhomola
committed

Commits on Jan 21, 2021

data, runner, cleanup
marcell-mate
committed

Commits on Jan 22, 2021

preproc, volume scaling
marcell-mate
committed

Commits on Jan 23, 2021

adding immediate reward like in the pgportfolio paper@
danielhomola
committed

Commits on Jan 26, 2021

Commits on Jan 29, 2021

Commits on Jan 30, 2021

adding multi head environment to speed up experience gathering without multiprocessing
danielhomola
committed

Commits on Jan 31, 2021

num_env_heads works with 1 or many, num_env_workers works with 1 or many, created portfolio_vs_market reward, cleaned up pgportfolio qvals/adv calc
danielhomola
committed

Commits on Feb 3, 2021

Commits on Feb 9, 2021

debug
marcell-mate
committed
vaaaaa
marcell-mate
committed

Commits on Feb 18, 2021

bugz
marcell-mate
committed

Commits on Feb 22, 2021

just do it
marcell-mate
committed

Commits on Feb 23, 2021

corrections
marcell-mate
committed

Commits on Feb 24, 2021

TGAN
marcell-mate
committed

Commits on Feb 28, 2021

Commits on Mar 17, 2021

testing on local and adding more flexibility
marcell_mate
committed
black. run!
marcell_mate
committed

Commits on Mar 24, 2021

more graphs
marcell_mate
committed

Commits on Mar 25, 2021

keep models based on training loss
marcell_mate
committed
changing config
marcell_mate
committed
changing config
marcell_mate
committed
changing config
marcell_mate
committed

Commits on Jul 14, 2021

config and accessing node nb
marcell_mate
committed

Commits on Jul 15, 2021

tgan readme
marcell_mate
committed