src.eval.torf_rfvsnn_evaluator

Classes

TORFRFvsNNEvaluator(**kwargs)

class src.eval.torf_rfvsnn_evaluator.TORFRFvsNNEvaluator(**kwargs)
Author:

Alberto M. Esmoris Pena

Evaluator that compares Random Forest (RF) and Neural Network (NN) predictions for a trained TransfOctoRF model.

The evaluator runs the RF and NN stages independently on the provided centroid features and collects the probabilities needed for the comparison evaluation.

Can be used:

  1. Internally by the TORF model during training via rfvsnn_plot_path / rfvsnn_report_path.

  2. As a pipeline component referenced in JSON as "eval": "TORFRFvsNNEvaluator".

See TORFRFvsNNEvaluation.

Variables:

  • plot_path (str or None) – Path for the comparison plot.

  • report_path (str or None) – Directory path for the CSV reports.

static extract_eval_args(spec)

Extract arguments from a pipeline JSON specification.

Parameters:

spec – The key-word specification.

Returns:

The arguments to initialize/instantiate a TORFRFvsNNEvaluator.

Return type:

dict

__init__(**kwargs)

Initialize a TORFRFvsNNEvaluator.

Parameters:

kwargs – Evaluator attributes.

eval(model, X=None, y=None, **kwargs)

Evaluate a trained TransfOctoRF model by comparing RF and NN predictions.

Parameters:

  • model (TransfOctoRFClassificationModel) – The trained TransfOctoRF model.

  • X (np.ndarray) – Centroid feature matrix (S, n_f).

  • y (np.ndarray) – Ground-truth centroid labels (S,).

Returns:

The RF vs NN evaluation.

Return type:

TORFRFvsNNEvaluation

__call__(pcloud, **kwargs)

Pipeline-based execution entry point.

Parameters:

  • pcloud (PointCloud) – The point cloud with predictions.

  • model – The TORF model.

  • y – Ground-truth labels.

  • out_prefix – Output prefix for path expansion.

eval_args_from_state(state)

Obtain arguments from the pipeline state.

Parameters:

state (SimplePipelineState) – The pipeline’s state.

Returns:

Dictionary of arguments for __call__.

Return type:

dict