README.md

Pythia

Times series deep learning model for trading.

Main Classes

• experiment is the configurations of everything. In theory, this should be reproducible and deterministic (maybe I need to add a seed in the json). The experiment determines the temporal framework used. Values can be for instance walk-forward backtest, cross-validation, standard (=train-val-test), robustness analysis, transfer learning and so on
• market identifies the timeseries used and the logic used in calculating the trading costs
• agent is the trading agent. In the first case, agent is supervised and has two parameters: predictor and trader. Otherwise, the agent is RL based.

LaTeX Docs

Please find our LaTeX documentation here

Execution

In order to run an experiment, run the following command in the terminal:

python pythia --run [SETTINGS_FILE]

For example:

python pythia --run data/experiments/chalvatzis/settings.json