8000 Commits · almanar/AutoTrading · GitHub

More Web Proxy on the site http://driver.im/

Commits

Commits on May 12, 2019

backtesting strategy using 1month trading and 1 week walk forward as test (pending trade profiler for evaluation)
poryee
committed

Commits on May 5, 2019

Added Tensorboard capabilities to track model over time (currently only the loss) which is the value we want to minimise
poryee
committed

Commits on Apr 29, 2019

fixed incorrect popping of position from stack resulting in wrong reward calculation
poryee
committed

Commits on Apr 24, 2019

completed resampling of data so we can trade at higher timeframe
poryee
committed

Commits on Apr 17, 2019

fixed close position (Did not factor fx and lot size) over profit retrain
poryee
committed

Commits on Apr 16, 2019

multiday seems legit now will test awhile more to see if stable then explore different training duration and steps
poryee
committed

Commits on Apr 13, 2019

included multiday training did not fare as well as i wanted, will proceed to chart positive vs negative reward ratio and then try to train resampled timeframe
poryee
committed

Commits on Apr 6, 2019

included average plot line and text for evaluation result chart and removed training of model during model evaluation
poryee
committed

Commits on Apr 1, 2019

added subplot to show acc balance over time note this is still single day training and evaluation
poryee
committed

Commits on Mar 30, 2019

Removed old code, and introduce visualisation of trading result for evaluation of model
poryee
committed

Commits on Mar 25, 2019

pending multi day training evaluating model past on single day train data still has plenty of downside volatility
poryee
committed

Commits on Mar 18, 2019

Update object and key outcome (OKR)
poryee
committed

Commits on Mar 17, 2019

Commits on Mar 13, 2019

Experience replay
poryee
committed

Commits on Mar 12, 2019

Upgrade model to Dueling DQN, start reward from 0 instead of account balance
poryee
committed

Commits on Mar 8, 2019

Commits on Mar 6, 2019

Commits on Jan 28, 2019

refactor code and stub for rl model
poryee
committed

Commits on Jan 20, 2019

download from IG as csv 30day is roughly the cap for 5min resolution
poryee
committed

Commits on Jan 9, 2019

ignore
poryee
committed

Commits on Jan 2, 2019

Delete ig_service_config.py
poryee
authored

Commits on Nov 16, 2018

update with negative reward accuracy is pretty shit
poryee
committed

Commits on Nov 13, 2018

Completed RL classification of lowest low mocking daily close. will proceed with caution using 30min time frame from IG
poryee
committed

Commits on Oct 30, 2018

Noticed that model overfits to training data.
poryee
committed

Commits on Oct 24, 2018

basic rl dqn functionality in replacement for prediction (will be enhanced or futher optimised to d4pg)
poryee
committed

Commits on Oct 23, 2018

updates
poryee
committed

Commits on Sep 30, 2018

Refactor code, and update ml_model with cross validation data split so that our prediction model don't overfit
poryee
committed

Commits on Sep 22, 2018

integrated performance test from risk adjusted metrics model, refactor ml model
poryee
committed

Commits on Sep 19, 2018

prediction model test using mnist
poryee
committed

Commits on Sep 16, 2018

basic calmar implementation to measure strategy risk adjusted performance
poryee
committed

Commits on Sep 12, 2018

complete construction of an oscillator indicator (14 day Momentum)
poryee
committed

0