conda-env-train

Prompt file imported from walterpinson/homeprice-predictor-vibe (.github/prompts/00-conda-env-train.prompt.md). Copyright stays with the author.

You are helping define the Conda environment for this project's HOUSE PRICE training workflow.

Task:

• Create or fully overwrite src/deploy/env-train.yml.

Goals:

• The same Conda environment should work for:
  • Local development on my machine (running register_data.py, train.py, register_model.py directly).
  • Azure ML training jobs that use train.py and the Azure ML Python SDK v2.

Requirements for env-train.yml:

• Use a standard Conda YAML structure with:
  • A name (e.g., vibe-ml-train).
  • Channels including at least conda-forge (and defaults if needed).
  • dependencies that include:
    • Python 3.10 or 3.11 (version that works well with azure-ai-ml).
    • numpy
    • pandas
    • scikit-learn
    • joblib (or ensure scikit-learn's deps cover it)
    • mltable (for loading Azure ML MLTable data assets)
    • pip
    • Under pip::
      • azure-ai-ml
      • azure-identity
      • (Optionally) mlflow if helpful for basic logging, but it's not required.

Constraints:

• Keep versions reasonably current and mutually compatible.
• Do not include heavy, unnecessary packages that would slow down environment creation (keep it lean for a demo).

Usage expectations (for humans):

• Locally, I will run:
  • conda env create -f src/deploy/env-train.yml
  • conda activate vibe-ml-train
• In Azure ML, I may reference this file when defining a custom environment for jobs.

Important:

• Do NOT print the file contents in chat.
• Write the complete YAML directly into src/deploy/env-train.yml, overwriting any existing content.