a2rl.utils.backtest#

a2rl.utils.backtest(df, simulator, start_row=0, context_rows=2, predict_rows=3, return_groudtruth=False)[source]#

Utility to perform backtesting on simulator.

By using groundtruth dataset df, take context_rows number of rows as context, and groundtruth action, predict the next states (s1,s2...) and rewards (r,v). Then append the predicted next states and rewards as new context, and repeat the steps again until we get predict_rows number of new rows.

Parameters:

  • df (WiDataFrame) – WiDataFrame, this is the original dataframe before tokenized.

  • simulator (Simulator) – Pretrain whatif Simulator.

  • start_row (int) – The dataframe starting row index for backtest

  • context_rows (int) – Number of dataframe rows to be used as context, starting from start_row.

  • predict_rows (int) – Number of dataframe rows to predict.

  • return_groudtruth (bool) – Return groundtruth dataframe for comparison when true.

Return type:

DataFrame | tuple[DataFrame, DataFrame]

Returns:

Simulated Dataframe.