240 architecture implement quality control pipeline orchestration

config/quality_control.yaml

@@ -18,7 +18,6 @@ metrics:
     enabled: true
     required_resources:
       - psf_models.standard
-
     params:
       statistic: reduced_chi_square
       normalize_residuals: true
@@ -28,7 +27,9 @@ rejection:
 
   mask_obscuration:
     enabled: true
-    threshold: 0.25
+    policy:
+      threshold:
+        value: 0.25
 
   goodness_of_fit:
     enabled: false

src/wf_psf/quality_control/context.py

@@ -0,0 +1,31 @@
+"""Quality control context.
+
+Encapsulates contextual information, such as datasets and resolved resources,
+required by the quality control pipeline and its metrics.
+
+:Authors:
+    Jennifer Pollack <jennifer.pollack@cea.fr>
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+
+@dataclass
+class QualityControlContext:
+    """Context shared across the quality control pipeline.
+
+    Attributes
+    ----------
+    dataset : Any
+        Dataset or data container supplied to the quality control pipeline.
+
+    resources : dict[str, Any]
+        Ready-to-use resources required by enabled quality metrics, keyed by
+        resource identifier.
+    """
+
+    dataset: Any
+    resources: dict[str, Any] = field(default_factory=dict)

src/wf_psf/quality_control/pipeline.py

@@ -2,11 +2,10 @@
 
 Defines the orchestration layer for dataset quality control.
 
-The QualityControlPipeline coordinates quality metric evaluation,
-application of rejection policies, reporting, and dataset filtering.
-Individual quality metrics and rejection policies are provided through
-their respective interfaces, allowing new methods to be added without
-modifying the pipeline implementation.
+The QualityControlPipeline coordinates quality metric evaluation and
+sample rejection. Individual quality metrics and rejection policies are
+provided through their respective interfaces, allowing new methods to
+be added without modifying the pipeline implementation.
 
 :Authors: Jennifer Pollack <jennifer.pollack@cea.fr>
 
@@ -15,6 +14,19 @@
 from dataclasses import dataclass
 import numpy as np
 
+from wf_psf.quality_control.config import QualityControlConfigHandler
+from wf_psf.quality_control.context import QualityControlContext
+from wf_psf.quality_control.metrics.base import QualityMetric
+from wf_psf.quality_control.metrics.registry import build_metrics_registry
+from wf_psf.quality_control.rejection.base import RejectionPolicy
+from wf_psf.quality_control.rejection.registry import build_rejection_policy_registry
+from wf_psf.quality_control.resources import Resources
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
 @dataclass
 class QualityControlResult:
     """Results produced by the quality control pipeline.
@@ -24,7 +36,7 @@ class QualityControlResult:
     metrics
         Computed quality metrics indexed by metric name.
 
-    rejection_masks
+    validity_masks
         Boolean validity masks produced by each rejection policy.
 
     valid_mask
@@ -34,7 +46,7 @@ class QualityControlResult:
 
     metrics: dict[str, np.ndarray]
 
-    rejection_masks: dict[str, np.ndarray]
+    validity_masks: dict[str, np.ndarray]
 
     valid_mask: np.ndarray
 
@@ -50,15 +62,129 @@ class QualityControlPipeline:
     of the pipeline execution.
     """
 
-    def __init__(
-            self,
-            metrics,
-            rejection_policies,
-    ):
-        ...
+    def __init__(self, qc_config_path):
+        self.config = QualityControlConfigHandler(qc_config_path).load()
+        self.metrics_registry = build_metrics_registry()
+        self.rejection_registry = build_rejection_policy_registry()
+
+    def _instantiate_metrics(self) -> dict[str, QualityMetric]:
+        """Instantiate enabled quality metric implementations from configuration.
+
+        Returns
+        -------
+        dict[str, QualityMetric]
+            Enabled quality metric implementations keyed by metric name.
+
+        Notes
+        -----
+        The quality control configuration is assumed to have been validated
+        before policy instantiation.
+        """
+        metrics = {}
+
+        for name, metric_config in self.config.metrics.items():
+            if not metric_config.enabled:
+                logger.debug("Skipping metric %s: not enabled.", name)
+                continue
+
+            metric_cls = self.metrics_registry.get(name)
+
+            metrics[name] = metric_cls()
+
+        logger.debug("Instantiated metrics: %s", list(metrics))
+
+        return metrics
+
+    def _instantiate_rejection_policies(self) -> dict[str, RejectionPolicy]:
+        """Instantiate enabled rejection policy implementations.
+
+        Returns
+        -------
+        dict[str, RejectionPolicy]
+            Enabled rejection policy implementations keyed by metric name.
+
+        Notes
+        -----
+        The quality control configuration is assumed to have been validated
+        before policy instantiation.
+        """
+        rejection_policies = {}
+
+        for metric_name, rejection_config in self.config.rejection.items():
+            if not rejection_config.enabled:
+                logger.debug("Skipping rejection policy %s: not enabled.", metric_name)
+                continue
+
+            policy_name, policy_params = next(iter(rejection_config.policy.items()))
+            policy_cls = self.rejection_registry.get(policy_name)
+
+            rejection_policies[metric_name] = policy_cls(**policy_params)
+
+        logger.debug("Instantiated rejection policies: %s", list(rejection_policies))
+
+        return rejection_policies
+
+    def _resolve_resources(self, provided_resources):
+        """Resolve resources required by enabled quality metrics.
+
+        Parameters
+        ----------
+        provided_resources : Mapping[str, Any] or None
+            Ready-to-use resources supplied by the pipeline caller, keyed by
+            resource identifier.
+
+        Returns
+        -------
+        dict[str, Any]
+            Resolved resources required by enabled quality metrics.
+        """
+        resource_manager = Resources(self.config)
+        return resource_manager.resolve(provided_resources)
+
+    def run(self, dataset, provided_resources=None):
+        """Run quality control pipeline.
+
+        Parameters
+        ----------
+        dataset : Any
+            Dataset or data container supplied to the quality control pipeline.
+
+        provided_resources : Mapping[str, Any] or None
+            Ready-to-use resources supplied by the pipeline caller, keyed by resource identifier.
+
+        Notes
+        -----
+        The pipeline is expected to be invoked only when at least one quality metric is enabled in the quality control configuration.
+        """
+        resolved_resources = self._resolve_resources(
+            provided_resources=provided_resources
+        )
+
+        context = QualityControlContext(dataset, resolved_resources)
+
+        metrics = self._instantiate_metrics()
+
+        metric_results = {
+            name: metric.compute(context) for name, metric in metrics.items()
+        }
+
+        rejection_policies = self._instantiate_rejection_policies()
 
-    def run(self, dataset):
+        validity_masks = {
+            name: policy.apply(metric_results[name])
+            for name, policy in rejection_policies.items()
+        }
 
-        ...
+        if validity_masks:
+            # True indicates a valid sample. A sample is valid only if it passes
+            # every enabled rejection policy.
+            valid_mask = np.logical_and.reduce(list(validity_masks.values()))
+        else:
+            metric_result = next(iter(metric_results.values()))
+            valid_mask = np.ones(metric_result.shape, dtype=bool)
 
-        return QualityControlResult(...)
+        return QualityControlResult(
+            metrics=metric_results,
+            validity_masks=validity_masks,
+            valid_mask=valid_mask,
+        )

src/wf_psf/quality_control/rejection/base.py

@@ -37,7 +37,12 @@ class RejectionPolicy(ABC):
 
     @abstractmethod
     def apply(self, metric: np.ndarray) -> np.ndarray:
-        """Return a boolean mask identifying valid dataset samples.
-
-        The returned mask has one entry per dataset sample.
+        """Apply the rejection policy to metric values.
+
+        Returns
+        -------
+        np.ndarray
+            Boolean validity mask with one entry per dataset sample. ``True``
+            indicates that the sample passes the rejection policy and should
+            be retained; ``False`` indicates that it should be rejected.
         """

src/wf_psf/quality_control/rejection/threshold.py

@@ -16,10 +16,23 @@
 
 
 class ThresholdRejectionPolicy(RejectionPolicy):
-    """Reject dataset samples based on configurable metric thresholds."""
+    """Reject dataset samples based on configurable metric threshold.
+
+    Attributes
+    ----------
+    name : str
+        Policy identifier used by the rejection policy registry.
+
+    value : float
+        Threshold applied by the rejection policy.
+
+    """
 
     name = "threshold"
 
+    def __init__(self, value: float):
+        self.value = value
+
     def apply(self, metric: np.ndarray) -> np.ndarray:
         """Apply threshold-based rejection to metric values."""
         raise NotImplementedError

src/wf_psf/tests/test_quality_control/data/invalid/rejection_metric_not_enabled.yaml

@@ -19,4 +19,6 @@ rejection:
 
   mask_obscuration:
     enabled: true
-    threshold: 0.25
+    policy:
+      threshold:
+        value: 0.25

src/wf_psf/tests/test_quality_control/data/valid/imaginary_metric.yaml

@@ -0,0 +1,4 @@
+metrics:
+
+  imaginary_metric:
+    enabled: true

src/wf_psf/tests/test_quality_control/data/valid/quality_control.yaml

@@ -18,7 +18,6 @@ metrics:
     enabled: true
     required_resources:
       - psf_models.standard
-
     params:
       statistic: reduced_chi_square
       normalize_residuals: true

src/wf_psf/tests/test_quality_control/data/valid/quality_control_multiple_rejection_policies.yaml

@@ -0,0 +1,42 @@
+resources:
+
+  psf_models:
+    standard:
+      inference_config: inference_standard.yaml
+    oversampled:
+      inference_config: inference_oversampled.yaml
+
+metrics:
+
+  mask_obscuration:
+    enabled: true
+    params:
+      aperture: gaussian
+      sigma: 2.5
+
+  goodness_of_fit:
+    enabled: true
+    required_resources:
+      - psf_models.standard
+    params:
+      statistic: reduced_chi_square
+      normalize_residuals: true
+
+rejection:
+
+  mask_obscuration:
+    enabled: true
+    policy:
+      threshold:
+        value: 3.0
+
+  goodness_of_fit:
+    enabled: true
+    policy:
+      threshold:
+        value: 3.0
+
+reporting:
+
+  save_metrics: true
+  log_statistics: true