Forcing the agent to perform Feature Selection

[ajpaezm]

Hello Siyuan,

If you were to force an action like Feature Selection, in which stage of your DS-Agent would you do it? I'm looking to do this only for Time-Series problems. So I wonder what of these 3 choices is the best way to go:

1. Modify the Time-Series examples to have this. The examples are those in the benchmarks folder.
2. Modify the tasks.json to include this as an instruction.
3. Build a dedicated method or class in which the Agent can rely to call each time to force this into the process.

[guosyjlu]

Actually, I think this is a just choice of your workflow design. You can either (1) couple feature engineering with machine learning engineering in one single script, or (2) decouple feature engineering and machine learning engineering with two seperate scripts. If you choose (1), you can modify the task description to instruct LLMs to do so. If you choose (2), you can build another agent to provide preprocessed data for DS-Agent.