Patch release 0.8.3 (#4155)

.github/workflow_scripts/env_setup.sh

@@ -57,28 +57,14 @@ function install_tabular_platforms {
     install_local_packages "tabular/$1"
 }
 
-function install_multimodal_no_groundingdino {
-    # groundingdino has issue when installing on Windows
-    # https://github.com/IDEA-Research/GroundingDINO/issues/57
-    source $(dirname "$0")/setup_mmcv.sh
-
-    # launch different process for each test to make sure memory is released
-    python3 -m pip install --upgrade pytest-xdist
-    install_local_packages "multimodal/$1"
-    setup_mmcv
-    # python3 -m pip install --upgrade "mmocr<1.0"  # not compatible with mmcv 2.0
-}
-
 function install_multimodal {
     source $(dirname "$0")/setup_mmcv.sh
-    source $(dirname "$0")/setup_groundingdino.sh
 
     # launch different process for each test to make sure memory is released
     python3 -m pip install --upgrade pytest-xdist
     install_local_packages "multimodal/$1"
     setup_mmcv
     # python3 -m pip install --upgrade "mmocr<1.0"  # not compatible with mmcv 2.0
-    setup_groundingdino
 }
 
 function install_all {
@@ -89,7 +75,7 @@ function install_all {
 
 function install_all_windows {
     install_local_packages "common/[tests]" "core/[all]" "features/" "tabular/[all,tests]" "timeseries/[all,tests]" "eda/[tests]"
-    install_multimodal_no_groundingdino "[tests]"
+    install_multimodal "[tests]"
     install_local_packages "autogluon/"
 }
 

.github/workflow_scripts/setup_mmcv.sh

@@ -1,4 +1,4 @@
 function setup_mmcv {
-  mim install mmcv --timeout 60
+  mim install "mmcv==2.1.0" --timeout 600
   python3 -m pip install "mmdet>=3.0.0"
 }

.github/workflow_scripts/test_eda.sh

@@ -11,7 +11,8 @@ export CUDA_VISIBLE_DEVICES=0
 install_local_packages "common/[tests]" "core/[all,tests]" "features/" "tabular/[all,tests]" "eda/[tests]"
 
 cd eda/
-python3 -m tox -e lint,typecheck,format,testenv
+# python3 -m tox -e lint,typecheck,format,testenv   # redundant with pytest below
+
 if [ -n "$ADDITIONAL_TEST_ARGS" ]
 then
     python3 -m pytest --junitxml=results.xml --runslow "$ADDITIONAL_TEST_ARGS" tests

VERSION

@@ -1 +1 @@
-0.8.2
+0.8.3

core/src/autogluon/core/_setup_utils.py

@@ -20,12 +20,12 @@
     "boto3": ">=1.10,<2",  # <2 because unlikely to introduce breaking changes in minor releases. >=1.10 because 1.10 is 3 years old, no need to support older
     "numpy": ">=1.21,<1.27",  # "<{N+3}" upper cap, where N is the latest released minor version, assuming no warnings using N
     "pandas": ">=1.4.1,<1.6",  # "<{N+1}" upper cap
-    "scikit-learn": ">=1.0,<1.3",  # "<{N+1}" upper cap
+    "scikit-learn": ">=1.1,<1.4.1",  # "<{N+1}" upper cap
     "scipy": ">=1.5.4,<1.12",  # "<{N+2}" upper cap
     "psutil": ">=5.7.3,<6",  # Major version cap
     "networkx": ">=3.0,<4",  # Major version cap
     "tqdm": ">=4.38,<5",  # Major version cap
-    "Pillow": ">=9.3,<9.6",  # "<{N+2}" upper cap
+    "Pillow": ">=10.2,<11",  # "<{N+2}" upper cap
 }
 if LITE_MODE:
     DEPENDENT_PACKAGES = {package: version for package, version in DEPENDENT_PACKAGES.items() if package not in ["psutil", "Pillow", "timm"]}

core/src/autogluon/core/calibrate/_decision_threshold.py

@@ -61,7 +61,7 @@ def calibrate_decision_threshold(
         problem_type=problem_type,
         decision_threshold=0.5,
     )
-    # TODO: Avoid calling like this, re-use logic that works with weights + extra args
+    # TODO: Avoid calling like this, reuse logic that works with weights + extra args
     score_val_baseline = metric(y, y_pred_val, **metric_kwargs)
 
     if verbose:
@@ -73,7 +73,7 @@ def calibrate_decision_threshold(
             problem_type=problem_type,
             decision_threshold=decision_threshold,
         )
-        # TODO: Avoid calling like this, re-use logic that works with weights + extra args
+        # TODO: Avoid calling like this, reuse logic that works with weights + extra args
         score_val = metric(y, y_pred_val, **metric_kwargs)
 
         if best_score_val is None or score_val > best_score_val:

core/src/autogluon/core/models/ensemble/bagged_ensemble_model.py

@@ -84,7 +84,7 @@ def __init__(self, model_base: Union[AbstractModel, Type[AbstractModel]], model_
         self._bagged_mode = None
         # _child_oof currently is only set to True for KNN models, that are capable of LOO prediction generation to avoid needing bagging.
         # TODO: Consider moving `_child_oof` logic to a separate class / refactor OOF logic.
-        # FIXME: Avoid unnecessary refit during refit_full on `_child_oof=True` models, just re-use the original model.
+        # FIXME: Avoid unnecessary refit during refit_full on `_child_oof=True` models, just reuse the original model.
         self._child_oof = False  # Whether the OOF preds were taken from a single child model (Assumes child can produce OOF preds without bagging).
         self._cv_splitters = []  # Keeps track of the CV splitter used for each bagged repeat.
         self._params_aux_child = None  # aux params of child model

core/src/autogluon/core/trainer/abstract_trainer.py

@@ -1251,7 +1251,7 @@ def refit_single_full(self, X=None, y=None, X_val=None, y_val=None, X_unlabeled=
                 model_name = model.name
                 reuse_first_fold = False
                 if isinstance(model, BaggedEnsembleModel):
-                    # Re-use if model is already _FULL and no X_val
+                    # Reuse if model is already _FULL and no X_val
                     if X_val is None:
                         reuse_first_fold = not model._bagged_mode
                 if not reuse_first_fold:

docs/conf.py

@@ -4,7 +4,7 @@
 sys.path = ['.', '..'] + sys.path
 
 project = 'AutoGluon'
-release = '0.8.2'
+release = '0.8.3'
 copyright = '2023, All authors. Licensed under Apache 2.0.'
 author = 'AutoGluon contributors'
 

docs/tutorials/cloud_fit_deploy/index.md

@@ -39,7 +39,7 @@ hidden: true
 
 AutoGluon Cloud <autogluon-cloud>
 AutoGluon Tabular on SageMaker AutoPilot <autopilot-autogluon>
-Deploy AutoGluon Models on Severless Templates <cloud-aws-lambda-deployment>
+Deploy AutoGluon Models on Serverless Templates <cloud-aws-lambda-deployment>
 Cloud Training and Deployment with Amazon SageMaker <cloud-aws-sagemaker-train-deploy>
 ```
 

docs/tutorials/multimodal/object_detection/index.md

@@ -10,12 +10,6 @@
 
   How to train high quality object detection model with MultiModalPredictor in under 5 minutes on COCO format dataset.
 :::
-
-:::{grid-item-card} Quick Start on Open Voccabulary Detection
-  :link: quick_start/quick_start_ovd.html
-
-  How to use a foundation model in object detection to detect novel classes defined by an unbounded (open) vocabulary.
-:::
 ::::
 
 ## Data Preparation

docs/tutorials/multimodal/object_detection/quick_start/index.md

@@ -8,12 +8,6 @@
 
   How to train high quality object detection model with MultiModalPredictor in under 5 minutes on COCO format dataset.
 :::
-
-:::{grid-item-card} Quick Start on Open Voccabulary Detection
-  :link: quick_start_ovd.html
-
-  How to use a foundation model in object detection to detect novel classes defined by an unbounded (open) vocabulary.
-:::
 ::::
 
 ```{toctree}
@@ -24,5 +18,4 @@ hidden: true
 ---
 
 quick_start_coco
-quick_start_ovd
 ```