Evaluate regression with metric operators

fiftyone/utils/eval/regression.py

@@ -34,6 +34,7 @@ def evaluate_regressions(
     pred_field,
     gt_field="ground_truth",
     eval_key=None,
+    eval_metrics=None,
     missing=None,
     method=None,
     progress=None,
@@ -69,6 +70,8 @@ def evaluate_regressions(
         gt_field ("ground_truth"): the name of the field containing the
             ground truth :class:`fiftyone.core.labels.Regression` instances
         eval_key (None): a string key to use to refer to this evaluation
+        eval_metrics (None): a list of tuples of ``fiftyone.operators.Operator``
+            for metric computation and the corresponding kwargs for the operator.
         missing (None): a missing value. Any None-valued regressions are
             given this value for results purposes
         method (None): a string specifying the evaluation method to use. The
@@ -97,7 +100,11 @@ def evaluate_regressions(
     eval_method.register_samples(samples, eval_key)
 
     results = eval_method.evaluate_samples(
-        samples, eval_key=eval_key, missing=missing, progress=progress
+        samples,
+        eval_metrics=eval_metrics,
+        eval_key=eval_key,
+        missing=missing,
+        progress=progress,
     )
     eval_method.save_run_results(samples, eval_key, results)
 
@@ -155,7 +162,12 @@ def register_samples(self, samples, eval_key):
             dataset.add_sample_field(eval_key, fof.FloatField)
 
     def evaluate_samples(
-        self, samples, eval_key=None, missing=None, progress=None
+        self,
+        samples,
+        eval_metrics=None,
+        eval_key=None,
+        missing=None,
+        progress=None,
     ):
         """Evaluates the regression predictions in the given samples with
         respect to the specified ground truth values.
@@ -254,10 +266,14 @@ class SimpleEvaluation(RegressionEvaluation):
     """
 
     def evaluate_samples(
-        self, samples, eval_key=None, missing=None, progress=None
+        self,
+        samples,
+        eval_metrics=None,
+        eval_key=None,
+        missing=None,
+        progress=None,
     ):
         metric = self.config._metric
-
         if metric == "squared_error":
             error_fcn = lambda yp, yt: (yp - yt) ** 2
         elif metric == "absolute_error":
@@ -293,12 +309,27 @@ def evaluate_samples(
             _confs = confs
             _ids = ids
 
+        # Metric operators.
+        agg_metric_ops = []
+        if eval_metrics:
+            for metric_op, metric_kwargs in eval_metrics:
+                if not metric_op.is_aggregate:  # check if per-sample/frame
+                    try:
+                        metric_op(
+                            samples, ytrue, ypred, eval_key, **metric_kwargs
+                        )
+                    except Exception as e:
+                        print(e)
+                else:
+                    agg_metric_ops.append((metric_op, metric_kwargs))
+
         results = RegressionResults(
             samples,
             self.config,
             eval_key,
             _ytrue,
             _ypred,
+            agg_metric_ops=agg_metric_ops,
             confs=_confs,
             ids=_ids,
             missing=missing,
@@ -312,10 +343,8 @@ def compute_error(yp, yt):
             if missing is not None:
                 if yp is None:
                     yp = missing
-
                 if yt is None:
                     yt = missing
-
             try:
                 return error_fcn(yp, yt)
             except:
@@ -371,6 +400,7 @@ def __init__(
         eval_key,
         ytrue,
         ypred,
+        agg_metric_ops=None,
         confs=None,
         ids=None,
         missing=None,
@@ -387,6 +417,7 @@ def __init__(
         self.confs = confs
         self.ids = ids
         self.missing = missing
+        self.agg_metrics_ops = agg_metric_ops
 
     def metrics(self, weights=None):
         """Computes various popular regression metrics for the results.
@@ -431,7 +462,7 @@ def metrics(self, weights=None):
             max_error = 0.0
             support = 0
 
-        return {
+        results = {
             "mean_squared_error": mse,
             "root_mean_squared_error": rmse,
             "mean_absolute_error": mae,
@@ -442,6 +473,19 @@ def metrics(self, weights=None):
             "support": support,
         }
 
+        if self.agg_metrics_ops:
+            for metric_op, metric_kwargs in self.agg_metrics_ops:
+                metric_val = metric_op(self.samples, **metric_kwargs)
+                sample_eval_key = metric_kwargs.get("sample_eval_key", None)
+                metric_name = (
+                    f"{metric_op.name}_{sample_eval_key}"
+                    if sample_eval_key
+                    else metric_op.name
+                )
+                results[metric_name] = metric_val.result
+
+        return results
+
     def print_metrics(self, weights=None, digits=2):
         """Prints the regression metrics computed via :meth:`metrics`.