Create Catalog of Country-Month Level Models

[Polichinel]

Objective

Compile a catalog of all country-month (cm) level of analysis (loa) models currently in the old pipeline that are being migrated to the new pipeline.

Requirements:

• Catalog Content: Each model entry in the catalog must include the following details:

  • Algorithm: Specify the model's algorithm.
  • Target: Define the model's forecasting target.
  • Input Features: List the input features (query set) used by the model.
  • Non-default Hyperparameters: Include all non-default hyperparameters.
  • Forecasting Type: Specify the type of forecasting (e.g., direct multi-step forecasting for step-shifting models).
  • Implementation Status: Indicate whether the model has been implemented in the new pipeline (note: this does not refer to whether it is activated).
  • Implementation Date & Author: These should be listed as "NA" for models not yet implemented in the new pipeline.

• File Format:

  • Create the catalog in a markdown file named cm_model_catalog.md.
  • The catalog should be structured as a table.
  • Store the file in the following directory:

    views_pipeline/documentation/catalogs/cm_model_catalog.md
    

• Branch:

  • Create a new branch named create_cm_catalog_01 to work on this task.

• Information Sources:

  • Use the current pipeline notebook to gather information about the country-month level models.
  • Consult Dr. Jim Dale if additional clarification is needed.

• Review Process:

  • Upon completion, the catalog must be reviewed by HH, JD, and MC before final approval.

Additional Actions:

• Glossary & ADRs:

  • Follow the existing templates for ADRs and glossary entries.
  • Document and create any necessary ADRs or glossary entries that emerge during the process.

• Issues:

  • If any issues arise or further definitions are needed, document them as new GitHub issues.

Exemple of table (draft)

Model Name	Algorithm	Target	Input Features	Non-default Hyperparameters	Forecasting Type	Implementation Status	Implementation Date	Author
NA	Random Forest	ln_sb_best	...	n_estimators=500, max_depth=10	Direct multi-step	No	NA	NA
magical_mystery	XGBoost	ln_sb_best	...	learning_rate=0.1, n_estimators=300	Direct multi-step	Yes	15-09-2023	Jim Dale

With model name, implementation date, and author will be NA as long as the implementation is No

---

Summary of Actionable Steps:

• Create a new branch named create_cm_catalog_01.
• Compile the model catalog with all required details: algorithm, target, input features, non-default hyperparameters, forecasting type, implementation status, and implementation date/author (where applicable).
• Format the catalog as a markdown table in cm_model_catalog.md, and store it in views_pipeline/documentation/catalogs/.
• Consult the current pipeline notebook and reach out to Dr. Jim Dale as needed.
• Submit the catalog for review by HH, JD, and MC.
• Note and create ADRs or glossary entries following existing templates.
• Log any issues encountered as GitHub issues.