Add s3_xgboost ds-model type

[billfreeman44]

What

New s3_xgboost model type in ds_models: loads an XGBoost model saved with the native save_model(*.json) format from S3 (res_type: application/json) and invokes predict_proba (model_type: XGBClassifier).

xgboost is a declared optional extra — pureskillgg-dsdk[xgboost] (which also pulls scikit-learn, required by the XGBClassifier sklearn API) — with a plain in-function import in the loader. No injection (that pattern exists for hdbscan only because hdbscan''s packaging was broken). It''s not a base dependency because csgo-ppp and other pipeline consumers of dsdk never load models and shouldn''t ship the ~200MB xgboost wheel. Consumers that already depend on xgboost directly (csgo-coach) need nothing beyond dsdk >=3.1.0.

Test trains a real tiny XGBClassifier, round-trips it through a fake S3 body, and asserts exact predict_proba parity.

Why

csgo-coach currently bakes xgb_model_v1.json (win_probability) into its Docker image and loads it from disk, bypassing the ds-models path every other model uses. This type lets that model live in the AI bucket and be registered/versioned in models.yaml like everything else. The XGBoost json format is version-stable, unlike pickles, so this is also the preferred serving route for future ML models.

Rollout (dependency chain)

1. This PR → release 3.1.0 to PyPI.
2. pureskillgg/reactor#25 (schema allows s3_xgboost in models.yaml) → release.
3. pureskillgg/csgo-assessment#90 registers xgb_win_probability; pureskillgg/csgo-coach#220 switches win_probability to get_ds_model.

A models.json entry with an unknown type is inert for consumers on dsdk <3.1.0 unless get_ds_model is called on that specific model, so merge order across consumers is safe.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds support for serving XGBoost classifier models through the existing ds_models abstraction, enabling models saved via XGBoost’s native JSON format to be fetched from S3 and invoked via predict_proba.

Changes:

• Introduces a new s3_xgboost ds-model loader (S3Xgboost) that loads application/json artifacts from S3 and runs predict_proba.
• Adds an xgboost optional extra (pureskillgg-dsdk[xgboost]) and updates the lockfile accordingly.
• Adds an integration-style test that trains an XGBClassifier, round-trips it through a fake S3 client, and checks predict_proba parity.

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
uv.lock	Locks new optional-extra dependencies (xgboost, scikit-learn and transitive deps).
pyproject.toml	Defines the xgboost optional extra and adds xgboost/sklearn to dev deps.
pureskillgg_dsdk/ds_models/s3_xgboost.py	Implements the s3_xgboost model type (S3 JSON load + predict_proba).
pureskillgg_dsdk/ds_models/s3_xgboost_test.py	Tests round-trip loading from S3 body and prediction parity vs local model.
pureskillgg_dsdk/ds_models/model.py	Registers s3_xgboost in the ds-model factory.
CHANGELOG.md	Documents the new model type and optional extra in release notes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[billfreeman44]

Addressed Copilot review:

• Bare import → generic ModuleNotFoundError: fixed — the loader now raises "xgboost is not installed: install the pureskillgg-dsdk[xgboost] extra" from the ImportError.
• xgboost/sklearn in the dev group: keeping as-is. The lint job needs them resolvable regardless — pylint statically resolves imports (E0401), so s3_xgboost_test.py's top-level import xgboost would fail lint without them. Splitting per-job --extra installs would complicate the shared setup action to save one cached wheel download.