REST API

pip install "amesml[api] @ git+https://github.com/Ahsan-muhd-444/house-price-prediction@v0.1.1"
amesml serve --model models/regression-gbm --model models/classification-gbm

The service loads its models once at startup and refuses to start if a model is missing or invalid. Interactive documentation is served at /docs, and the OpenAPI schema at /openapi.json.

Endpoints

Method	Path	Returns
GET	/health	status, version and which tasks have a model
GET	/model/info	the metadata.json of each loaded model, by task
POST	/predict	predicted sale prices
POST	/predict/class	price class and probability for each house

Predicting

Send up to 1,000 houses as records:

curl -X POST http://127.0.0.1:8000/predict \
  -H "Content-Type: application/json" \
  -d @examples/predict_request.json

{
  "predictions": [180571.47, 145403.74],
  "model": {
    "task": "regression",
    "model": "gbm",
    "estimator": "HistGradientBoostingRegressor",
    "amesml_version": "0.1.0",
    "created_at": "2026-11-01T11:10:18+00:00"
  },
  "warnings": []
}

/predict/class takes the same body and returns:

{
  "predictions": [
    {"label": 1, "above_threshold": true, "probability": 0.9431},
    {"label": 0, "above_threshold": false, "probability": 0.0003}
  ],
  "threshold": 165000.0,
  "model": {"task": "classification", "model": "gbm"},
  "warnings": []
}

Partial records

Records use the Ames column names. You do not have to send all 79:

• Leave a field out and it takes the model's typical training value: the median for measurements, the most common value for everything else.
• Send null to say the feature is absent - for example "PoolQC": null for a house without a pool. This is how the dataset encodes absence, and the model handles it exactly as it did in training.

Errors

Status	When
422	an unknown field name (so typos fail loudly), text in a numeric field, an empty or oversized batch
503	no model is loaded for the requested task

Categories the model never saw in training (an unknown Neighborhood, say) do not fail the request. They are listed in warnings and encoded as the column's reference category.

Configuration

Variable	Default	Meaning
AMESML_MODEL_PATH	-	comma-separated model paths, one per task
AMESML_HOST	127.0.0.1	interface to bind
AMESML_PORT	8000	port

To run under your own ASGI server, use amesml.api:app (reads AMESML_MODEL_PATH) or call amesml.api.create_app([paths]).