.. _auto_ml_predict:

*ML PREDICT*
============

Applies a previously trained ML model to a new dataset to generate predictions, supporting both scalar and curve-valued targets. It accepts a saved model file (`.pkl`, `.csv` manifest, or a registered math-model ID) and returns a dataset of predicted values aligned to the input rows. Use this worker whenever you need to score new data points with a model produced by the ML Learn worker.

When to use
-----------

Tagged: ``curve prediction``, ``inference``, ``math model``, ``ml``, ``per-target model``, ``pkl``, ``predict``, ``scoring``.

Inputs
------

.. list-table::
   :header-rows: 1
   :widths: 20 20 20 20 20 20

   * - Label
     - ID
     - Type
     - Default
     - Required
     - Description
   * - Prediction Dataset
     - dataset
     - dataset
     - —
     - ✓
     - Tabular dataset whose rows are the samples to be scored; each column is a model input feature. Must be non-empty; rows are matched positionally to predictions in the output.
   * - Saved Model File
     - mfile
     - text
     - —
     - ✓
     - Path, d3VIEW math-model ID, or comma-separated list of `.pkl` file names pointing to the serialised model(s) to use for inference; may also be a `.csv` manifest produced by the ML Predict-Info worker listing per-target best models.
   * - Learn Dataset
     - reference_dataset
     - dataset
     - —
     - 
     - Optional training (learn) dataset used as a reference when reconstructing curve predictions; leave empty for scalar-target models.
   * - Raw Curve Column Name
     - raw_curve_column
     - text
     - —
     - 
     - Column name in `dataset` whose x-values are used to rebuild the predicted time/frequency curve; required only for curve-valued targets, ignored otherwise.
   * - Inputs
     - inputs
     - text
     - —
     - 
     - Comma-separated list of input feature column names; when provided together with `targets` the worker computes prediction-accuracy metrics against the supplied ground truth.
   * - Targets
     - targets
     - text
     - —
     - 
     - Comma-separated list of target column names present in `dataset`; used alongside `inputs` and `raw_vs_predictions` to evaluate and report prediction accuracy.
   * - Training Raw vs Predictions
     - raw_vs_predictions
     - dataset
     - —
     - 
     - Optional dataset containing ground-truth target values aligned to `dataset`; used to calculate and attach accuracy statistics (e.g. R², RMSE) to the output.

Outputs
-------

.. list-table::
   :header-rows: 1
   :widths: 20 20 20 20

   * - Label
     - ID
     - Type
     - Description
   * - Predictions
     - dataset
     - dataset
     - Dataset of model predictions with one row per input sample; scalar targets appear as numeric columns and curve targets are stored as serialised CurveGroup objects keyed by target name, with original row `id` values preserved.

Disciplines
-----------

- ai_ml.supervised.classification
- ai_ml.supervised.regression
- ai_ml.surrogate

.. raw:: html

   <hr style="margin-top:2em">
   <p style="font-size:11px;color:#888">
   Auto-generated from <code>platform</code> schema. Worker id: <code>ml_predict</code>. Schema hash: <code>d229c9f9b122</code>. Hand-curated docs in <code>workerexamples/</code> override this page when present.
   </p>