Prediction and forecasting.
Functions
execute_predict(**kwargs) -> Dict[str, Any]
Execute prediction with support for timeseries and anomaly models.
Backs the
tuiml_predict tool. Timeseries models forecast steps ahead; anomaly detectors additionally report anomaly counts and score statistics; all other models run standard predict.Parameters
model_id
str
Identifier of a trained model from
tuiml_train (arrives via **kwargs, like all parameters below). One of model_id / model_path is required.
model_path
str
= None
Explicit path to a serialized model file.
data
str
Dataset to predict on (dataset_id, file path, or built-in name). Not needed for timeseries forecasting.
stage
str
= None
Optional stage:
'forecast' (timeseries) or 'predict_proba' (class probabilities). When None, dispatch is based on model tags.
stage_kwargs
dict
= None
Extra stage arguments (e.g.
steps for forecasting).
steps
int
= 10
Number of future steps to forecast for timeseries models.
output_path
str
= None
When given, predictions are also written to this file.
Returns
result
dict
On success:
status ('success'), num_predictions, predictions_preview (first 10), and optionally model_type, steps, output_path; anomaly models add n_anomalies, n_normal, anomaly_ratio, anomaly_scores_preview and score_stats. On failure: status ('error'), error, error_type and optionally suggestion.