In our case the teacher-student setup is used as in Universal Trading for Order Execution with Oracle Policy Distillation. It is used as an oracle that has access to all the epoch data. In this way we can generate ideal actions that can be further used as GT for a student agent. This will help & stabilize learning.
Train teacher:
python main.py train --config-file order_execution/all/single_asset_all_universal_teacher.config.txt --storage_path ./storage/universal_teacher --market-storage-dir ./storage
Export the actions:
export_actions --config-file-name order_execution/all/single_asset_all_universal_teacher_export.config.txt --storage-dir ./storage/universal_teacher --market-storage-dir ./storage
Train the student with the exported actions:
python main.py train --config-file order_execution/all/single_asset_all_universal_distillation.config.txt --storage_path ./storage/universal_distillation --market-storage-dir ./storage