Compatibility for xgboost 0.90 (#90)

Author: krfricke

.github/workflows/test.yaml

@@ -189,4 +189,39 @@ jobs:
         bash ./run_ci_tests.sh
     - name: Run examples
       run: |
-        bash ./run_ci_examples.sh
+        bash ./run_ci_examples.sh
+
+  test_linux_xgboost_legacy:
+    # Tests on XGBoost 0.90 and latest Ray release
+    runs-on: ubuntu-latest
+    timeout-minutes: 14
+    strategy:
+      matrix:
+        python-version: [3.6.9]
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install codecov
+        python -m pip install -U ray
+        if [ -f requirements-test.txt ]; then python -m pip install  -r requirements-test.txt; fi
+    - name: Install package
+      run: |
+        python -m pip install -e .
+    - name: Install legacy XGBoost
+      run: |
+        python -m pip install xgboost==0.90
+    - name: Print environment info
+      run: |
+        ./xgboost_ray/tests/env_info.sh
+    - name: Run tests
+      run: |
+        bash ./run_ci_tests.sh
+    - name: Run examples
+      run: |
+        bash ./run_ci_examples.sh

examples/simple.py

@@ -34,7 +34,7 @@ def main(cpus_per_actor, num_actors):
         evals=[(test_set, "eval")],
         evals_result=evals_result,
         ray_params=RayParams(
-            max_actor_restarts=1,
+            max_actor_restarts=0,
             gpus_per_actor=0,
             cpus_per_actor=cpus_per_actor,
             num_actors=num_actors),

format.sh

@@ -104,22 +104,22 @@ format_changed() {
              yapf --in-place "${YAPF_EXCLUDES[@]}" "${YAPF_FLAGS[@]}"
         if which flake8 >/dev/null; then
             git diff --name-only --diff-filter=ACRM "$MERGEBASE" -- '*.py' | xargs -P 5 \
-                 flake8 --inline-quotes '"' --no-avoid-escape --ignore=C408,E121,E123,E126,E226,E24,E704,W503,W504,W605
+                 flake8 --inline-quotes '"' --no-avoid-escape --ignore=N,I,C408,E121,E123,E126,E226,E24,E704,W503,W504,W605
         fi
     fi
 
     if ! git diff --diff-filter=ACRM --quiet --exit-code "$MERGEBASE" -- '*.pyx' '*.pxd' '*.pxi' &>/dev/null; then
         if which flake8 >/dev/null; then
             git diff --name-only --diff-filter=ACRM "$MERGEBASE" -- '*.pyx' '*.pxd' '*.pxi' | xargs -P 5 \
-                 flake8 --inline-quotes '"' --no-avoid-escape --ignore=C408,E121,E123,E126,E211,E225,E226,E227,E24,E704,E999,W503,W504,W605
+                 flake8 --inline-quotes '"' --no-avoid-escape --ignore=N,I,C408,E121,E123,E126,E211,E225,E226,E227,E24,E704,E999,W503,W504,W605
         fi
     fi
 }
 
 # Format all files, and print the diff to stdout for travis.
 format_all() {
     yapf --diff "${YAPF_FLAGS[@]}" "${YAPF_EXCLUDES[@]}" xgboost_ray
-    flake8 --inline-quotes '"' --no-avoid-escape --ignore=C408,E121,E123,E126,E211,E225,E226,E227,E24,E704,E999,W503,W504,W605 xgboost_ray
+    flake8 --inline-quotes '"' --no-avoid-escape --ignore=N,I,C408,E121,E123,E126,E211,E225,E226,E227,E24,E704,E999,W503,W504,W605 xgboost_ray
 }
 
 # This flag formats individual files. --files *must* be the first command line

setup.py

@@ -10,5 +10,5 @@
     "distributed computing framework Ray.",
     url="https://github.com/ray-project/xgboost_ray",
     install_requires=[
-        "xgboost>=1.3", "ray", "numpy>=1.16,<1.20", "pandas", "pyarrow"
+        "xgboost>=0.90", "ray", "numpy>=1.16,<1.20", "pandas", "pyarrow"
     ])

xgboost_ray/compat/__init__.py

@@ -0,0 +1,44 @@
+import xgboost as xgb
+
+try:
+    from xgboost.callback import TrainingCallback
+    LEGACY_CALLBACK = False
+except ImportError:
+
+    class TrainingCallback:
+        def __init__(self):
+            if hasattr(self, "before_iteration"):
+                # XGBoost < 1.0 is looking up __dict__ to see if a
+                # callback should be called before or after an iteration.
+                # So here we move this to self._before_iteration and
+                # overwrite the dict.
+                self._before_iteration = getattr(self, "before_iteration")
+                self.__dict__["before_iteration"] = True
+
+        def __call__(self, callback_env: xgb.core.CallbackEnv):
+            if hasattr(self, "_before_iteration"):
+                self._before_iteration(
+                    model=callback_env.model,
+                    epoch=callback_env.iteration,
+                    evals_log=callback_env.evaluation_result_list)
+
+            if hasattr(self, "after_iteration"):
+                self.after_iteration(
+                    model=callback_env.model,
+                    epoch=callback_env.iteration,
+                    evals_log=callback_env.evaluation_result_list)
+
+        def before_training(self, model):
+            pass
+
+        def after_training(self, model):
+            pass
+
+    LEGACY_CALLBACK = True
+
+try:
+    from xgboost import RabitTracker
+except ImportError:
+    from xgboost_ray.compat.tracker import RabitTracker
+
+__all__ = ["TrainingCallback", "RabitTracker"]