XGBoost runtime fixes

.github/workflows/images.yml

@@ -130,9 +130,9 @@ jobs:
         env:
           VERSION: ${{ steps.docker-tag.outputs.value }}
         run: |
-          make docker-build docker-push PYTHON_VERSION=3.8.10
-          make docker-tag-base-python docker-push-base-python PYTHON_VERSION=3.8.10
-          docker save -o /tmp/base-wrapper.tar seldonio/seldon-core-s2i-python38:${VERSION}
+          make docker-build docker-push PYTHON_VERSION=3.10.15
+          make docker-tag-base-python docker-push-base-python PYTHON_VERSION=3.10.15
+          docker save -o /tmp/base-wrapper.tar seldonio/seldon-core-s2i-python310:${VERSION}
 
       - name: Upload artifact
         uses: actions/upload-artifact@v3

python-builder/Dockerfile

@@ -33,7 +33,7 @@ RUN apt-get upgrade -y && \
 
 
 # This is to install desired version of Python without updating conda version
-ENV PYTHON_VERSION "3.7.10"
+ENV PYTHON_VERSION "3.10.15"
 ENV CONDA_VERSION "4.7.12"
 RUN conda install --yes -c conda-forge python=$PYTHON_VERSION conda=$CONDA_VERSION
 

servers/xgboostserver/Makefile

@@ -11,7 +11,7 @@ docker-build:
 	s2i build \
 		-E environment \
 		./xgboostserver \
-		${DOCKER_REGISTRY}/seldon-core-s2i-python38:${VERSION} \
+		seldonio/seldon-core-s2i-python3:${VERSION} \
 		${IMAGE_NAME}:${VERSION}
 
 docker-push:

servers/xgboostserver/README.md

@@ -0,0 +1 @@
+# XGBoost Runtime

servers/xgboostserver/test/test_metadata_and_input.py

@@ -0,0 +1,39 @@
+import os
+import tempfile
+import numpy as np
+import pytest
+import xgboost as xgb
+from XGBoostServer import XGBoostServer
+
+@pytest.fixture
+def model_uri():
+    with tempfile.TemporaryDirectory() as temp_dir:
+        X = np.random.rand(100, 5)
+        y = np.random.randint(2, size=100)
+        dtrain = xgb.DMatrix(X, label=y)
+        params = {'objective': 'binary:logistic', 'eval_metric': 'error'}
+        booster = xgb.train(params, dtrain, num_boost_round=10)
+        model_path = os.path.join(temp_dir, "model.json")
+        booster.save_model(model_path)
+        yield temp_dir
+
+def test_init_metadata(model_uri):
+    metadata = {"key": "value"}
+    metadata_path = os.path.join(model_uri, "metadata.yaml")
+    with open(metadata_path, "w") as f:
+        yaml.dump(metadata, f)
+
+    server = XGBoostServer(model_uri)
+
+    loaded_metadata = server.init_metadata()
+
+    # Assert that the loaded metadata matches the original metadata
+    assert loaded_metadata == metadata
+
+def test_predict_invalid_input(model_uri):
+    # Create an instance of XGBoostServer with the model URI
+    server = XGBoostServer(model_uri)
+    server.load()
+    X_test = np.random.rand(10, 3)  # Incorrect number of features
+    with pytest.raises(ValueError):
+        server.predict(X_test, names=[])

servers/xgboostserver/test/test_model_and_predict.py

@@ -0,0 +1,50 @@
+import os
+import tempfile
+import numpy as np
+from unittest import mock
+import xgboost as xgb
+from XGBoostServer import XGBoostServer
+
+def test_load_json_model():
+
+    with tempfile.TemporaryDirectory() as temp_dir:
+        X = np.random.rand(100, 5)
+        y = np.random.randint(2, size=100)
+        dtrain = xgb.DMatrix(X, label=y)
+        params = {'objective': 'binary:logistic', 'eval_metric': 'error'}
+        booster = xgb.train(params, dtrain, num_boost_round=10)
+        model_path = os.path.join(temp_dir, "model.json")
+        booster.save_model(model_path)
+
+        server = XGBoostServer(temp_dir)
+        server.load()
+
+        assert server.ready
+        assert isinstance(server._booster, xgb.Booster)
+
+def test_predict():
+    # Create a temporary directory for the model file
+    with tempfile.TemporaryDirectory() as temp_dir:
+        # Train a dummy XGBoost model and save it in .json format
+        X = np.random.rand(100, 5)
+        y = np.random.randint(2, size=100)
+        dtrain = xgb.DMatrix(X, label=y)
+        params = {'objective': 'binary:logistic', 'eval_metric': 'error'}
+        booster = xgb.train(params, dtrain, num_boost_round=10)
+        model_path = os.path.join(temp_dir, "model.json")
+        booster.save_model(model_path)
+
+        # Create an instance of XGBoostServer with the model URI
+        server = XGBoostServer(temp_dir)
+
+        server.load()
+
+        # Prepare test data
+        X_test = np.random.rand(10, 5)
+
+        with mock.patch("seldon_core.Storage.download", return_value=temp_dir):
+            predictions = server.predict(X_test, names=[])
+
+        # Assert the expected shape and type of predictions
+        assert isinstance(predictions, np.ndarray)
+        assert predictions.shape == (10,)

servers/xgboostserver/test/xgboost_iris.ipynb

@@ -9,21 +9,55 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 4,
    "metadata": {},
    "outputs": [],
    "source": [
     "import os\n",
-    "\n",
     "import xgboost as xgb\n",
-    "from sklearn.datasets import load_iris\n",
-    "\n",
+    "from sklearn.datasets import load_iris"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [],
+   "source": [
     "model_dir = \".\"\n",
-    "BST_FILE = \"model.bst\"\n",
-    "\n",
+    "BOOSTER_FILE = \"model.json\"\n",
+    "BOOSTER_FILE_DEPRECATED = \"model.bst\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
     "iris = load_iris()\n",
     "y = iris[\"target\"]\n",
-    "X = iris[\"data\"]\n",
+    "X = iris[\"data\"]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/home/ramonpzg/micromamba/envs/mlserver-quick/lib/python3.11/site-packages/xgboost/core.py:158: UserWarning: [18:32:52] WARNING: /workspace/src/learner.cc:740: \n",
+      "Parameters: { \"silent\" } are not used.\n",
+      "\n",
+      "  warnings.warn(smsg, UserWarning)\n"
+     ]
+    }
+   ],
+   "source": [
+    "\n",
     "dtrain = xgb.DMatrix(X, label=y)\n",
     "param = {\n",
     "    \"max_depth\": 6,\n",
@@ -34,8 +68,10 @@
     "    \"objective\": \"multi:softmax\",\n",
     "}\n",
     "xgb_model = xgb.train(params=param, dtrain=dtrain)\n",
-    "model_file = os.path.join((model_dir), BST_FILE)\n",
-    "xgb_model.save_model(model_file)"
+    "model_file_json = os.path.join((model_dir), BOOSTER_FILE)\n",
+    "model_file_bst= os.path.join((model_dir), BOOSTER_FILE_DEPRECATED)\n",
+    "xgb_model.save_model(model_file_json)\n",
+    "xgb_model.save_model(model_file_bst)"
    ]
   },
   {
@@ -319,7 +355,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "Python 3 (ipykernel)",
    "language": "python",
    "name": "python3"
   },
@@ -333,7 +369,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.7.6"
+   "version": "3.11.9"
   },
   "varInspector": {
    "cols": {

servers/xgboostserver/xgboostserver/XGBoostServer.py

@@ -6,10 +6,13 @@
 import yaml
 import logging
 import xgboost as xgb
+import json
+from packaging import version
 
 logger = logging.getLogger(__name__)
 
-BOOSTER_FILE = "model.bst"
+BOOSTER_FILE = "model.json"
+BOOSTER_FILE_DEPRECATED = "model.bst"
 
 
 class XGBoostServer(SeldonComponent):
@@ -22,7 +25,27 @@ def load(self):
         model_file = os.path.join(
             seldon_core.Storage.download(self.model_uri), BOOSTER_FILE
         )
-        self._booster = xgb.Booster(model_file=model_file)
+        if not os.path.exists(model_file):
+            # Fallback to deprecated .bst format
+            model_file = os.path.join(
+                seldon_core.Storage.download(self.model_uri), BOOSTER_FILE_DEPRECATED
+            )
+            if os.path.exists(model_file):
+                logger.warning(
+                    "Using deprecated .bst format for XGBoost model. "
+                    "Please update to the .json format in the future."
+                )
+            else:
+                raise FileNotFoundError(f"Model file not found: {BOOSTER_FILE} or {BOOSTER_FILE_DEPRECATED}")
+
+        if version.parse(xgb.__version__) < version.parse("1.7.0"):
+            # Load model using deprecated method for older XGBoost versions
+            self._booster = xgb.Booster(model_file=model_file)
+        else:
+            # Load model using the new .json format for XGBoost >= 1.7.0
+            self._booster = xgb.Booster()
+            self._booster.load_model(model_file)
+
         self.ready = True
 
     def predict(
@@ -39,12 +62,14 @@ def init_metadata(self):
 
         try:
             with open(file_path, "r") as f:
-                return yaml.safe_load(f.read())
+                metadata = yaml.safe_load(f.read())
+                # Validate and sanitize the loaded metadata if needed
+                return metadata
         except FileNotFoundError:
-            logger.debug(f"metadata file {file_path} does not exist")
+            logger.debug(f"Metadata file {file_path} does not exist")
             return {}
         except yaml.YAMLError:
             logger.error(
-                f"metadata file {file_path} present but does not contain valid yaml"
+                f"Metadata file {file_path} present but does not contain valid YAML"
             )
-            return {}
+            return {}

servers/xgboostserver/xgboostserver/requirements.txt

@@ -1,3 +1,3 @@
-scikit-learn == 1.0.2
+scikit-learn >= 1.0.2, <=1.5.0
 numpy >= 1.8.2
-xgboost == 1.4.2
+xgboost >= 1.7.0, <=2.2.0

wrappers/s2i/python/Makefile

@@ -3,9 +3,9 @@ VERSION ?= $(shell cat ../../../version.txt)
 
 DOCKER_REGISTRY ?= seldonio
 
-CONDA_DOWNLOAD_VERSION ?= py38_23.3.1-0
+CONDA_DOWNLOAD_VERSION ?= py310
 CONDA_VERSION ?= 23.5.0
-PYTHON_VERSION ?= 3.8.10
+PYTHON_VERSION ?= 3.10.15
 
 IMAGE_PYTHON_VERSION = $(shell echo -n $(PYTHON_VERSION) | cut -d. -f1-2 | sed 's/\.//g')
 DEFAULT_IMAGE_PYTHON_VERSION = $(shell echo -n $(PYTHON_VERSION) | cut -d. -f1)

wrappers/s2i/python/build_scripts/build_all.sh

@@ -2,10 +2,10 @@
 make -C ../ docker-build-conda-base
 
 # Build standard wrapper images
-make -C ../ docker-build PYTHON_VERSION=3.8.10
+make -C ../ docker-build PYTHON_VERSION=3.10.15
 
 # Push default tag image
-make -C ../ docker-tag-base-python PYTHON_VERSION=3.8.10
+make -C ../ docker-tag-base-python PYTHON_VERSION=3.10.15
 
 # Build GPU images
-make -C ../ docker-build-gpu PYTHON_VERSION=3.8.10
+make -C ../ docker-build-gpu PYTHON_VERSION=3.10.15

wrappers/s2i/python/build_scripts/push_all.sh

@@ -2,10 +2,10 @@
 make -C ../ docker-push-conda-base
 
 # Push standard wrapper images
-make -C ../ push_to_dockerhub PYTHON_VERSION=3.8.10
+make -C ../ push_to_dockerhub PYTHON_VERSION=3.10.15
 
 # Push default tag image
-make -C ../ docker-push-base-python PYTHON_VERSION=3.8.10
+make -C ../ docker-push-base-python PYTHON_VERSION=3.10.15
 
 # Push GPU images
-make -C ../ docker-push-gpu PYTHON_VERSION=3.8.10
+make -C ../ docker-push-gpu PYTHON_VERSION=3.10.15