Hotfix for no dependency config loading

luxonis_train/config/predefined_models/base_predefined_model.py

@@ -1,4 +1,5 @@
 from abc import abstractmethod
+from contextlib import suppress
 from typing import Literal
 
 from loguru import logger
@@ -208,42 +209,47 @@ def nodes(self) -> list[NodeConfig]:
         return nodes
 
     def _generate_metrics(self) -> list[MetricModuleConfig]:
-        if self._per_class_metrics is None:
-            return [
-                MetricModuleConfig(
-                    name=metric,
-                    params=self._metrics_params,
-                    is_main_metric=metric == self._main_metric,
-                )
-                for metric in self._metrics
-            ]
-
-        task = NODES.get(self._head).task
-        metrics = []
-        applied_per_class_override = False
-
-        for metric in self._metrics:
-            metric_params = dict(self._metrics_params)
-            metric_cls = METRICS.get(metric)
-            aliases = metric_cls.get_predefined_model_params_aliases(task)
-            param_name = aliases.get("per_class_metrics")
-            if param_name is not None:
-                metric_params[param_name] = self._per_class_metrics
-                applied_per_class_override = True
-
-            metrics.append(
-                MetricModuleConfig(
-                    name=metric,
-                    params=metric_params,
-                    is_main_metric=metric == self._main_metric,
-                )
-            )
+        if self._per_class_metrics is not None:
+            try:
+                task = NODES.get(self._head).task
+                metrics = []
+                applied_per_class_override = False
+
+                for metric in self._metrics:
+                    metric_params = dict(self._metrics_params)
+                    metric_cls = METRICS.get(metric)
+                    aliases = metric_cls.get_predefined_model_params_aliases(
+                        task
+                    )
+                    param_name = aliases.get("per_class_metrics")
+                    if param_name is not None:
+                        metric_params[param_name] = self._per_class_metrics
+                        applied_per_class_override = True
 
-        if self._metrics and not applied_per_class_override:
-            logger.warning(
-                "Ignoring `per_class_metrics` for predefined model metrics "
-                f"{self._metrics} because none of them support a per-class "
-                "override."
-            )
+                    metrics.append(
+                        MetricModuleConfig(
+                            name=metric,
+                            params=metric_params,
+                            is_main_metric=metric == self._main_metric,
+                        )
+                    )
+
+                if self._metrics and not applied_per_class_override:
+                    logger.warning(
+                        "Ignoring `per_class_metrics` for predefined model metrics "
+                        f"{self._metrics} because none of them support a per-class "
+                        "override."
+                    )
+                return metrics  # noqa: TRY300
 
-        return metrics
+            except Exception:
+                logger.warning("Unable to ")
+
+        return [
+            MetricModuleConfig(
+                name=metric,
+                params=self._metrics_params,
+                is_main_metric=metric == self._main_metric,
+            )
+            for metric in self._metrics
+        ]