Release 1.1.0: port integration logging fixes from whitebox-xai-azure/sdk

CHANGELOG.md

@@ -7,6 +7,61 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.1.0] - 2026-07-30
+
+Ports integration fixes and compatibility improvements developed in the
+`whitebox-xai-azure` monorepo's `sdk/` directory since the 1.0.0 cut.
+
+### Fixed
+- **XGBoost/LightGBM**: `XGBoostMonitor.predict()`/`LightGBMMonitor.predict()`
+  and `wrap_xgboost_model()`/`wrap_lightgbm_model()` called a nonexistent
+  `log_predictions()` method (raising `AttributeError`) instead of
+  `ModelMonitor.log_batch()` — the primary documented usage of both
+  integrations was broken.
+- **TensorFlow/Keras**: `KerasMonitor.predict()` called `log_prediction()`/
+  `log_batch()` with invalid keyword arguments (`prediction=`, `actual=`,
+  `predictions=`, `actuals=`), raising `TypeError` on every call.
+  `KerasMonitor.log_epoch()`/`log_checkpoint()` (and the
+  `WhiteBoxXAICallback` training callback) called a `log_custom_metric()`
+  method that didn't exist, raising `AttributeError`.
+  `KerasMonitor.set_baseline()` raised `TypeError` when computing baseline
+  predictions, since it called the parent `ModelMonitor.set_baseline()`
+  with two positional arguments against a signature that only accepted one
+  (see the `ModelMonitor.set_baseline()` fix below).
+  `register_saved_model()` raised `ValueError` unless `self.model` was set,
+  even though it exists specifically to register a model saved to disk.
+  `wrap_keras_model()` also called `log_batch()` with invalid keyword
+  arguments.
+- **Hugging Face Transformers**: `TransformersMonitor.log_prediction_transformers()`/
+  `log_batch_transformers()` called `log_prediction()`/`log_batch()` with
+  invalid keyword arguments, raising `TypeError` on every prediction.
+  `TransformersMonitor.set_baseline()` had the same `ModelMonitor.set_baseline()`
+  argument-count crash as `KerasMonitor` above.
+  `wrap_transformers_pipeline()` reassigned `pipeline.__call__` on a pipeline
+  *instance*, which Python never actually invokes (dunder-method lookup
+  happens on the type, not the instance) — wrapped pipelines silently never
+  logged any predictions.
+- **LangChain**: `LangChainMonitor.log_chain_execution()`,
+  `log_agent_execution()`, `log_llm_call()`, `log_tool_call()`, and
+  `log_rag_retrieval()` all passed `prediction=` to `log_prediction()`,
+  which only accepts `output=` — all five convenience-logging methods
+  raised `TypeError`.
+- `OfflineQueue.clear_completed()` used an exclusive day-boundary comparison
+  (`created_at < ...`), so a completed record exactly `older_than_days` old
+  was never cleared. Now inclusive (`<=`).
+
+### Added
+- `whiteboxxai.integrations.langchain`/`langchain_agents` now try
+  `langchain_core.*` first (the modern, lighter-weight package) and fall
+  back to the legacy `langchain.callbacks.base`/`langchain.schema.*`
+  locations, so the SDK works with `langchain-core`-only installs.
+  `AgentExecutor`/`Chain` (higher-level `langchain`-only abstractions) are
+  probed independently so their absence no longer disables the rest of the
+  integration.
+- `ModelMonitor.set_baseline()` accepts an optional `labels` argument for
+  baseline predictions/labels.
+- `KerasMonitor.log_custom_metric()`.
+
 ## [1.0.0] - 2026-07-14
 
 First stable, production release. This release realigns the SDK with the
@@ -153,6 +208,9 @@ from the git history between the `0.2.0` and `0.2.1` tags:
 - PII detection and masking
 - Secure API key handling
 
-[Unreleased]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/v0.2.0...HEAD
-[0.2.0]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/v0.1.0...v0.2.0
-[0.1.0]: https://github.com/AgentaFlow/whitebox-python-sdk/releases/tag/v0.1.0
+[Unreleased]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/1.1.0...HEAD
+[1.1.0]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/1.0.0...1.1.0
+[1.0.0]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/0.2.1...1.0.0
+[0.2.1]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/0.2.0...0.2.1
+[0.2.0]: https://github.com/AgentaFlow/whitebox-python-sdk/compare/0.1.0...0.2.0
+[0.1.0]: https://github.com/AgentaFlow/whitebox-python-sdk/releases/tag/0.1.0

docs/api-reference.md

@@ -141,7 +141,7 @@ log_batch(predictions: List[dict])
 Log multiple predictions.
 
 ```python
-set_baseline(data: np.ndarray, **kwargs)
+set_baseline(data: np.ndarray, labels: Optional[np.ndarray] = None)
 ```
 Set baseline data for drift detection.
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "whitebox-xai-sdk"
-version = "1.0.0"
+version = "1.1.0"
 description = "Python SDK for WhiteBox XAI - AI Observability & Explainability Platform"
 readme = "README.md"
 authors = [

tests/unit/test_monitor.py

@@ -6,6 +6,7 @@
 
 from unittest.mock import AsyncMock, Mock, patch
 
+import numpy as np
 import pytest
 
 from whiteboxxai.client import WhiteBoxXAI
@@ -178,6 +179,33 @@ def test_get_drift_reports(self):
         assert reports[0]["drift_score"] == 0.8
 
 
+class TestBaselineData:
+    """Tests for ModelMonitor.set_baseline()."""
+
+    def test_set_baseline_data_only(self):
+        """Test setting baseline data without labels."""
+        mock_client = Mock(spec=WhiteBoxXAI)
+        monitor = ModelMonitor(client=mock_client, model_id="model-123")
+        data = np.array([[1.0, 2.0], [3.0, 4.0]])
+
+        monitor.set_baseline(data)
+
+        assert monitor._baseline_data is data
+        assert monitor._baseline_labels is None
+
+    def test_set_baseline_with_labels(self):
+        """Test setting baseline data with labels."""
+        mock_client = Mock(spec=WhiteBoxXAI)
+        monitor = ModelMonitor(client=mock_client, model_id="model-123")
+        data = np.array([[1.0, 2.0], [3.0, 4.0]])
+        labels = np.array([0, 1])
+
+        monitor.set_baseline(data, labels=labels)
+
+        assert monitor._baseline_data is data
+        assert monitor._baseline_labels is labels
+
+
 class TestAlertManagement:
     """Tests for alert management."""
 

tests/unit/test_offline.py

@@ -0,0 +1,45 @@
+"""
+Tests for SDK Offline Queue Module
+
+Tests for offline mode queueing functionality.
+"""
+
+import sqlite3
+
+from whiteboxxai.offline import OfflineQueue, OperationPriority, OperationType
+
+
+class TestOfflineQueueClearCompleted:
+    """Tests for OfflineQueue.clear_completed()."""
+
+    def test_clears_records_exactly_older_than_days_old(self, tmp_path):
+        """A completed record whose age is exactly `older_than_days` must be
+        cleared (boundary is inclusive: created_at <= cutoff, not <)."""
+        db_path = str(tmp_path / "queue.db")
+        queue = OfflineQueue(db_path=db_path)
+
+        op_id = queue.enqueue(OperationType.PREDICT, {"id": 1}, OperationPriority.NORMAL)
+        queue.mark_success(op_id)
+
+        with sqlite3.connect(db_path) as conn:
+            conn.execute(
+                "UPDATE queue SET created_at = datetime('now', '-7 days') WHERE id = ?",
+                (op_id,),
+            )
+            conn.commit()
+
+        queue.clear_completed(older_than_days=7)
+
+        assert queue.get_statistics()["total"] == 0
+
+    def test_keeps_records_younger_than_older_than_days(self, tmp_path):
+        """A completed record younger than `older_than_days` must be kept."""
+        db_path = str(tmp_path / "queue.db")
+        queue = OfflineQueue(db_path=db_path)
+
+        op_id = queue.enqueue(OperationType.PREDICT, {"id": 1}, OperationPriority.NORMAL)
+        queue.mark_success(op_id)
+
+        queue.clear_completed(older_than_days=7)
+
+        assert queue.get_statistics()["total"] == 1

whiteboxxai/integrations/boosting.py

@@ -64,6 +64,31 @@
     lgb = None
 
 
+def _build_prediction_batch(
+    inputs: np.ndarray,
+    predictions: np.ndarray,
+    actuals: Optional[np.ndarray] = None,
+    probabilities: Optional[np.ndarray] = None,
+) -> List[Dict[str, Any]]:
+    """Build per-sample dicts for ModelMonitor.log_batch() from parallel arrays."""
+
+    def _item(value: Any) -> Any:
+        return value.tolist() if isinstance(value, np.ndarray) else value
+
+    batch = []
+    for i in range(len(inputs)):
+        item: Dict[str, Any] = {
+            "inputs": _item(inputs[i]),
+            "output": _item(predictions[i]),
+        }
+        if actuals is not None:
+            item["actual"] = _item(actuals[i])
+        if probabilities is not None:
+            item["probability"] = _item(probabilities[i])
+        batch.append(item)
+    return batch
+
+
 class XGBoostMonitor(ModelMonitor):
     """
     Monitor for XGBoost models.
@@ -241,23 +266,13 @@ def predict(
 
         # Log predictions
         if log_predictions:
-            pred_metadata = metadata or {}
-
-            # Add feature importance for this batch if enabled
-            if self.track_feature_importance:
-                try:
-                    importance = self._get_feature_importance(model)
-                    if importance:
-                        pred_metadata["feature_importance"] = importance
-                except Exception as e:
-                    warnings.warn(f"Failed to extract feature importance: {e}")
-
-            self.log_predictions(
-                inputs=X,
-                predictions=predictions,
-                actuals=y_true,
-                probabilities=probabilities,
-                metadata=pred_metadata,
+            self.log_batch(
+                _build_prediction_batch(
+                    inputs=X,
+                    predictions=predictions,
+                    actuals=y_true,
+                    probabilities=probabilities,
+                )
             )
 
         return predictions
@@ -476,23 +491,13 @@ def predict(
 
         # Log predictions
         if log_predictions:
-            pred_metadata = metadata or {}
-
-            # Add feature importance for this batch if enabled
-            if self.track_feature_importance:
-                try:
-                    importance = self._get_feature_importance(model)
-                    if importance:
-                        pred_metadata["feature_importance"] = importance
-                except Exception as e:
-                    warnings.warn(f"Failed to extract feature importance: {e}")
-
-            self.log_predictions(
-                inputs=X,
-                predictions=predictions,
-                actuals=y_true,
-                probabilities=probabilities,
-                metadata=pred_metadata,
+            self.log_batch(
+                _build_prediction_batch(
+                    inputs=X,
+                    predictions=predictions,
+                    actuals=y_true,
+                    probabilities=probabilities,
+                )
             )
 
         return predictions
@@ -558,7 +563,7 @@ def wrap_xgboost_model(model: Any, monitor: XGBoostMonitor, auto_register: bool
         >>> wrapped_model = wrap_xgboost_model(model, monitor)
         >>> predictions = wrapped_model.predict(X_test)  # Auto-logged
     """
-    if auto_register and not monitor._model_id:
+    if auto_register and not monitor.model_id:
         monitor.register_from_model(model)
 
     # Store original methods
@@ -574,7 +579,7 @@ def wrapped_predict(X, *args, **kwargs):
 
         # Log predictions
         try:
-            monitor.log_predictions(inputs=X, predictions=predictions)
+            monitor.log_batch(_build_prediction_batch(inputs=X, predictions=predictions))
         except Exception as e:
             warnings.warn(f"Failed to log predictions: {e}")
 
@@ -589,8 +594,10 @@ def wrapped_predict_proba(X, *args, **kwargs):
             # Log with probabilities
             try:
                 predictions = np.argmax(probabilities, axis=1)
-                monitor.log_predictions(
-                    inputs=X, predictions=predictions, probabilities=probabilities
+                monitor.log_batch(
+                    _build_prediction_batch(
+                        inputs=X, predictions=predictions, probabilities=probabilities
+                    )
                 )
             except Exception as e:
                 warnings.warn(f"Failed to log predictions: {e}")
@@ -624,7 +631,7 @@ def wrap_lightgbm_model(model: Any, monitor: LightGBMMonitor, auto_register: boo
         >>> wrapped_model = wrap_lightgbm_model(model, monitor)
         >>> predictions = wrapped_model.predict(X_test)  # Auto-logged
     """
-    if auto_register and not monitor._model_id:
+    if auto_register and not monitor.model_id:
         monitor.register_from_model(model)
 
     # Store original methods
@@ -640,7 +647,7 @@ def wrapped_predict(X, *args, **kwargs):
 
         # Log predictions
         try:
-            monitor.log_predictions(inputs=X, predictions=predictions)
+            monitor.log_batch(_build_prediction_batch(inputs=X, predictions=predictions))
         except Exception as e:
             warnings.warn(f"Failed to log predictions: {e}")
 
@@ -655,8 +662,10 @@ def wrapped_predict_proba(X, *args, **kwargs):
             # Log with probabilities
             try:
                 predictions = np.argmax(probabilities, axis=1)
-                monitor.log_predictions(
-                    inputs=X, predictions=predictions, probabilities=probabilities
+                monitor.log_batch(
+                    _build_prediction_batch(
+                        inputs=X, predictions=predictions, probabilities=probabilities
+                    )
                 )
             except Exception as e:
                 warnings.warn(f"Failed to log predictions: {e}")