Skip to main content
Inference Engine — Production Scoring API. The primary user-facing interface for applying trained AIDflow pipelines to new data. Provides access to every pipeline stage (cleaning, encoding, agent prediction) using the exact parameters locked during training.
Applicable for projects with stored (save=1) models only.

​
What It Does

  • Loads all trained models from a project’s storage.
  • Preprocesses input data through the full pipeline using trained standards.
  • Predicts with any trained agent or the champion production model.
  • Exposes intermediate stages for diagnostic or custom-chain use.

​
Pipeline Stages Available

  1. Data Quality: fe_clean — type/quality/relevance validation against trained standards.
  2. Stability: fe_stabilize, obs_stabilize — time-stability filtering and outlier labeling.
  3. Encoding: num_encode, cat_encode, ixn_encode, woe_encode, woe_stabilize, data_encode — full feature transformation pipeline.
  4. Prediction: agent1_predict … agent4_predict, champion_predict — model inference.

​
Two Usage Modes

  • preprocess=True (default): Pass original input columns. The full preprocessing pipeline is applied automatically.
  • preprocess=False: Pass pre-encoded data (e.g., WOE features). Skips preprocessing — useful for chaining stages or when encoding is applied externally.

​
Parameters

project : str | Parameters
Base training project. Accepts a project name (str) or
a Parameters object. Defines which trained models
are loaded for inference.
 path : str
Alternative project setup path for comparison. Must
reside within the base project’s 2_Output folder.
Training-Serving Symmetry:
  • Every transformation uses the parameters locked during training (binning thresholds, imputation values, encoding maps, stability filters).
  • No statistics are re-derived at inference time.
Chainable Stages:
  • Individual stages (fe_clean, woe_encode, etc.) can be called independently to inspect intermediate results or build custom pipelines.
Champion vs. Agents:
  • The champion_predict method applies the production model (selected during the Selection stage).
  • Individual agent predictions (agent1_predict … agent4_predict) are available for comparison or monitoring.

​
Examples