- This is a very simple trading agent that uses statistics as features from training data to decide if the agent should buy, sell, or hold stock.
- Agent take the training data, calculate the statistics from this dataset.
- The statistic is trained online, not offline (no pre-training), before the testing data is ingested
- Once the statistics is calculated, the results is used to let agent decide and output the action file.
- During testing time, model distribution shift is fixed and closing price is submitted for agent evaluation.