ppo act ret pen turn 3e-4 pi lr nstepp 200 reward accumulation

ppo active return penalized turnovers 1e-3 pi_lr + ppo nstep 20 reward acc with grad clip

ppo with act ret w/ turnovers 3e-4 pi_lr + 100x scaled reward + ppo nstep 20 reward accumulation

ppo active return penalized turnovers

training logs ppo act ret clipped actions [0, 1] + ppo stability update (clipped actions, clip ration 0.1, bptt 256, ent bonus)

ppo act ret true rnn 200 epochs 1e-5 pi lr

add model

add models

ppo true rnn with correct update and reset

ppo true rnn with correct update and reset

ppo active return lstm cidl

fix models names

ppo max exposure lstm cidl remove actions mean + log std variations

ppo max exposure lstm cidl remove actions mean + log std variations

ppo max exposure lstm cidl features diagnostics

fix model names

minor name change

trained ppo for diagnostics

trained ppo

rename lstm models

ppo llm max exposure cidl features

ppo rnn with cidl features

add trained ppo longer lookback and sequence

rename model

ppo lstm no leverage env variations + sentiment no action scaling

cleaup models

ppo sentiment, ppo env variations

ppo base features sentiment

trained ppo agents 1e21 cash

ppo trained agents max exposure + asset cap pct with 1e15 cash

trained ppo lstm agents

trained ppo asset cap rnn

trained ppo agents

ppo max exposure correct actions from env

trained ppo lstm max exposure

trained ppo with max exposure

trained ppo agents correct model actions in [-1, 1]

agents ppo clipped actions [-1, 1] + fixed implicit leverage

ppo agents smaller mlp

add trained agents

ppo agents 168/4320 window lens

rename model

trained ppo agents

ppo agents 24/8640 min/max window lens

rename agent folder

ppo agents normalized stretched windows new scales

ppo agent with random assets initialization

add ppo agent no gamma return

trained ppo agents

only feats normalized data trained ppo agents