Ensemble Forecasting

Ensemble Forecasting

We are excited to introduce the new POST /v1/forecast/ensemble endpoint, which automatically selects and blends the best-performing models for your specific data.

How it works

When you submit a forecast request to the ensemble endpoint, TSFM.ai:

1. Backtests your data against all available models using a configurable holdout window
2. Scores each model using your preferred metric (SMAPE, MASE, or CRPS)
3. Ranks models by performance and selects the top performers
4. Blends their predictions using optimized weights to produce a single ensemble forecast

Request example

curl -X POST https://api.tsfm.ai/v1/forecast/ensemble \
  -H "Authorization: Bearer $TSFM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "data": { "values": [1.2, 3.4, 5.6, ...], "freq": "h" },
    "prediction_length": 24,
    "scoring_metric": "smape",
    "top_n": 5,
    "max_models": 3
  }'

Response

The response includes the ranked results with individual model scores, the computed weights for each model in the ensemble, and the final blended prediction.

Configuration

• top_n — Number of models to evaluate during backtesting (default: 10)
• max_models — Maximum number of models to include in the final ensemble (default: 3)
• scoring_metric — One of smape, mase, or crps (default: smape)
• budget — Maximum compute budget in seconds (default: 30)

Ensemble forecasting is available on all paid plans.