Upgrade numpy 2.0

.github/workflow_scripts/env_setup.sh

@@ -16,10 +16,11 @@ function setup_build_contrib_env {
 
 function setup_benchmark_env {
     pip install -U autogluon.bench
-    pip install pyarrow # TODO: Remove once AG-Bench v0.4.4 is released
+    pip install pyarrow
     git clone https://github.com/autogluon/autogluon-dashboard.git
     pip install -e ./autogluon-dashboard
     pip install yq
+    pip install s3fs
 }
 
 function setup_hf_model_mirror {

.github/workflow_scripts/lint_check.sh

@@ -22,6 +22,6 @@ function lint_check_all {
     # lint_check tabular
 }
 
-bandit -r multimodal/src -ll
+bandit -r multimodal/src -ll --exclude "multimodal/src/autogluon/multimodal/configs/pretrain/*"
 lint_check_all
 ruff check timeseries/

.github/workflow_scripts/setup_mmcv.sh

@@ -1,4 +1,6 @@
 function setup_mmcv {
+  # Install MMEngine from git with the fix for torch 2.5
+  python3 -m pip install "git+https://github.com/open-mmlab/mmengine.git@2e0ab7a92220d2f0c725798047773495d589c548"
   mim install "mmcv==2.1.0" --timeout 60
   python3 -m pip install "mmdet==3.2.0"
 }

.github/workflows/benchmark_master.yml

@@ -5,7 +5,7 @@ on:
     branches:
       - master
   schedule:
-    - cron: '00 09 * * SUN' #  UTC 9:00(2:00 PST Time) every Sunday
+    - cron: '00 02 * * *' #  UTC 2:00 AM every day
 
 env:
   AG_BRANCH_NAME: master

CI/batch/docker/Dockerfile.cpu

@@ -1,10 +1,10 @@
-FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.2.0-cpu-py310-ubuntu20.04-ec2
+FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.4.0-cpu-py311-ubuntu22.04-ec2
 
 RUN apt-get update \
  && apt-get -y upgrade \
  && apt-get install -y --no-install-recommends \
     pandoc \
-    python3.8-venv \
+    python3.11-venv \
     graphviz \
     graphviz-dev \
  && apt-get autoremove -y \

CI/batch/docker/Dockerfile.gpu

@@ -1,12 +1,12 @@
-FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.2.0-gpu-py310-cu121-ubuntu20.04-ec2
+FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.4.0-gpu-py311-cu124-ubuntu22.04-ec2
 
 ARG AWSCLI_VER=1.22.45
 
 RUN apt-get update \
  && apt-get -y upgrade \
  && apt-get install -y --no-install-recommends \
     pandoc \
-    python3.8-venv \
+    python3.11-venv \
     graphviz \
     graphviz-dev \
  && apt-get autoremove -y \

CI/bench/generate_bench_config.sh

@@ -27,7 +27,7 @@ if [ $MODULE == "tabular" ] || [ $MODULE == "timeseries" ]; then
     --amlb-benchmark $BENCHMARK \
     --amlb-constraint $TIME_LIMIT \
     --amlb-user-dir $(dirname "$0")/amlb_user_dir \
-    --git-uri-branch https://github.com/openml/automlbenchmark.git#stable
+    --git-uri-branch https://github.com/Innixma/automlbenchmark.git#autogluon_switch_to_uv
 else
     FRAMEWORK=AutoGluon_$PRESET
     aws s3 cp --recursive s3://autogluon-ci-benchmark/configs/$MODULE/$USER_DIR_S3_PREFIX/latest/ $(dirname "$0")/custom_user_dir/

CI/docker/Dockerfile.cpu-inference

@@ -1,4 +1,4 @@
-FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-inference:2.2.0-cpu-py310-ubuntu20.04-sagemaker
+FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-inference:2.4.0-cpu-py311-ubuntu22.04-sagemaker
 
 RUN apt-get update \
  && apt-get -y upgrade \

CI/docker/Dockerfile.cpu-training

@@ -1,4 +1,4 @@
-FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.2.0-cpu-py310-ubuntu20.04-sagemaker
+FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.4.0-cpu-py311-ubuntu22.04-sagemaker
 
 RUN apt-get update \
  && apt-get -y upgrade \

CI/docker/Dockerfile.gpu-inference

@@ -1,4 +1,4 @@
-FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-inference:2.2.0-gpu-py310-cu118-ubuntu20.04-sagemaker
+FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-inference:2.4.0-gpu-py311-cu124-ubuntu22.04-sagemaker
 
 RUN apt-get update \
  && apt-get -y upgrade \

CI/docker/Dockerfile.gpu-training

@@ -1,4 +1,4 @@
-FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.2.0-gpu-py310-cu121-ubuntu20.04-sagemaker
+FROM 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.4.0-gpu-py311-cu124-ubuntu22.04-sagemaker
 
 RUN apt-get update \
  && apt-get -y upgrade \

common/setup.py

@@ -28,14 +28,12 @@
         "tqdm",  # version range defined in `core/_setup_utils.py`
         # s3fs is removed due to doubling install time due to version range resolution
         # "s3fs",  # version range defined in `core/_setup_utils.py`
-        "setuptools",
     ]
     if not ag.LITE_MODE
     else {
         "numpy",  # version range defined in `core/_setup_utils.py`
         "pandas",  # version range defined in `core/_setup_utils.py`
         "tqdm",  # version range defined in `core/_setup_utils.py`
-        "setuptools",
     }
 )
 

common/src/autogluon/common/__init__.py

@@ -1,3 +1,4 @@
+from .dataset import TabularDataset
 from .features.feature_metadata import FeatureMetadata
 from .utils.log_utils import _add_stream_handler
 from .utils.log_utils import fix_logging_if_kaggle as __fix_logging_if_kaggle

common/src/autogluon/common/dataset.py

@@ -0,0 +1,36 @@
+import pandas as pd
+
+from .loaders import load_pd
+
+__all__ = ["TabularDataset"]
+
+
+class TabularDataset:
+    """
+    A dataset in tabular format (with rows = samples, columns = features/variables).
+    This class returns a :class:`pd.DataFrame` when initialized and all existing pandas methods can be applied to it.
+    For full list of methods/attributes, see pandas Dataframe documentation: https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.html
+
+    The purpose of this class is to provide an easy-to-use shorthand for loading a pandas DataFrame to use in AutoGluon.
+
+    Parameters
+    ----------
+    data : str, :class:`pd.DataFrame`, :class:`np.ndarray`, Iterable, or dict
+        If str, path to data file (CSV or Parquet format).
+        If you already have your data in a :class:`pd.DataFrame`, you can specify it here. In this case, the same DataFrame will be returned with no changes.
+
+    Examples
+    --------
+    >>> import pandas as pd
+    >>> from autogluon.common import TabularDataset
+    >>> train_data = TabularDataset("https://autogluon.s3.amazonaws.com/datasets/Inc/train.csv")
+    >>> train_data_pd = pd.read_csv("https://autogluon.s3.amazonaws.com/datasets/Inc/train.csv")
+    >>> assert isinstance(train_data, pd.DataFrame)  # True
+    >>> assert train_data.equals(train_data_pd)  # True
+    >>> assert type(train_data) == type(train_data_pd)  # True
+    """
+
+    def __new__(cls, data, **kwargs) -> pd.DataFrame:
+        if isinstance(data, str):
+            data = load_pd.load(data)
+        return pd.DataFrame(data, **kwargs)

common/src/autogluon/common/loaders/load_pd.py

@@ -16,7 +16,6 @@
 
 # TODO: v1.0 consider renaming function so it isn't 'load'. Consider instead 'load_pd', or something more descriptive.
 # TODO: Add full docstring
-# TODO: Add full docstring for usage within TabularDataset
 def load(
     path,
     delimiter=None,

common/src/autogluon/common/utils/s3_utils.py

@@ -86,7 +86,7 @@ def upload_file(*, file_name: str, bucket: str, prefix: Optional[str] = None):
     import boto3
 
     object_name = os.path.basename(file_name)
-    if len(prefix) == 0:
+    if prefix is not None and len(prefix) == 0:
         prefix = None
     if prefix is not None:
         object_name = prefix + "/" + object_name

common/tests/unittests/test_dataset.py

@@ -0,0 +1,15 @@
+import pandas as pd
+
+from autogluon.common import TabularDataset
+
+
+def test_tabular_dataset():
+    data = {"col1": [1, 2, 3, 4], "col2": ["a", "b", "b", "c"]}
+
+    df_1 = pd.DataFrame(data)
+    df_2 = TabularDataset(data)
+
+    assert isinstance(df_1, pd.DataFrame)
+    assert df_1.equals(df_2)
+    assert type(df_1) == pd.DataFrame
+    assert type(df_1) == type(df_2)

core/src/autogluon/core/__init__.py

@@ -1,7 +1,8 @@
+# noinspection PyUnresolvedReferences
+from autogluon.common.dataset import TabularDataset
 from autogluon.common.utils.log_utils import _add_stream_handler
 
 from . import constants, metrics
-from .dataset import TabularDataset
 from .version import __version__
 
 _add_stream_handler()

core/src/autogluon/core/_setup_utils.py

@@ -17,22 +17,21 @@
 # Only put packages here that would otherwise appear multiple times across different module's setup.py files.
 DEPENDENT_PACKAGES = {
     "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,<1.29",  # "<{N+3}" upper cap, where N is the latest released minor version, assuming no warnings using N
+    "numpy": ">=1.25.0,<2.1.4",  # "<{N+3}" upper cap, where N is the latest released minor version, assuming no warnings using N
     "pandas": ">=2.0.0,<2.3.0",  # "<{N+3}" upper cap
     "scikit-learn": ">=1.4.0,<1.5.3",  # capping to latest version
-    "scipy": ">=1.5.4,<1.13",  # "<{N+2}" upper cap
+    "scipy": ">=1.5.4,<1.16",  # "<{N+2}" upper cap
     "matplotlib": ">=3.7.0,<3.11",  # "<{N+2}" upper cap
     "psutil": ">=5.7.3,<7.0.0",  # Major version cap
     "s3fs": ">=2023.1,<2025",  # Yearly cap
     "networkx": ">=3.0,<4",  # Major version cap
     "tqdm": ">=4.38,<5",  # Major version cap
     "Pillow": ">=10.0.1,<12",  # Major version cap
-    "torch": ">=2.2,<2.5",  # Major version cap, sync with common/src/autogluon/common/utils/try_import.py
-    "lightning": ">=2.2,<2.4",  # Major version cap
-    "pytorch_lightning": ">=2.2,<2.4",  # Major version cap, capping `lightning` does not cap `pytorch_lightning`!
-    "async_timeout": ">=4.0,<5",  # Major version cap
-    "transformers[sentencepiece]": ">=4.38.0,<4.41.0",
-    "accelerate": ">=0.21.0,<0.22.0",
+    "torch": ">=2.2,<2.6",  # Major version cap, sync with common/src/autogluon/common/utils/try_import.py
+    "lightning": ">=2.2,<2.6",  # Major version cap
+    "async_timeout": ">=4.0,<6",  # Major version cap
+    "transformers[sentencepiece]": ">=4.38.0,<5",
+    "accelerate": ">=0.32.0,<1.0",
 }
 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/metrics/__init__.py

@@ -17,6 +17,7 @@
 from ..constants import BINARY, MULTICLASS, QUANTILE, REGRESSION, SOFTCLASS
 from . import classification_metrics, quantile_metrics
 from .classification_metrics import confusion_matrix
+from .score_func import compute_metric
 
 
 class Scorer(object, metaclass=ABCMeta):
@@ -180,26 +181,31 @@ def __repr__(self) -> str:
     @property
     @abstractmethod
     def needs_pred(self) -> bool:
+        """If True, metric requires predictions rather than prediction probabilities"""
         raise NotImplementedError
 
     @property
     @abstractmethod
     def needs_proba(self) -> bool:
+        """If True, metric requires prediction probabilities rather than predictions"""
         raise NotImplementedError
 
     @property
     @abstractmethod
     def needs_class(self) -> bool:
+        """If True, metric requires class label predictions rather than prediction probabilities"""
         raise NotImplementedError
 
     @property
     @abstractmethod
     def needs_threshold(self) -> bool:
+        """If True, metric requires prediction probabilities rather than predictions"""
         raise NotImplementedError
 
     @property
     @abstractmethod
     def needs_quantile(self) -> bool:
+        """If True, metric requires quantile predictions rather than predictions or prediction probabilities"""
         raise NotImplementedError
 
     score = __call__
@@ -685,32 +691,32 @@ def customized_roc_auc(y_true, y_pred, **kwargs):
     _add_scorer_to_metric_dict(metric_dict=BINARY_METRICS, scorer=scorer)
 
 
-for name, metric in [("precision", sklearn.metrics.precision_score), ("recall", sklearn.metrics.recall_score), ("f1", sklearn.metrics.f1_score)]:
-    globals()[name] = make_scorer(name, metric, needs_class=True)
-    _add_scorer_to_metric_dict(metric_dict=BINARY_METRICS, scorer=globals()[name])
+for _name, _metric in [("precision", sklearn.metrics.precision_score), ("recall", sklearn.metrics.recall_score), ("f1", sklearn.metrics.f1_score)]:
+    globals()[_name] = make_scorer(_name, _metric, needs_class=True)
+    _add_scorer_to_metric_dict(metric_dict=BINARY_METRICS, scorer=globals()[_name])
     for average in ["macro", "micro", "weighted"]:
-        qualified_name = "{0}_{1}".format(name, average)
-        globals()[qualified_name] = make_scorer(qualified_name, partial(metric, pos_label=None, average=average), needs_class=True)
+        qualified_name = "{0}_{1}".format(_name, average)
+        globals()[qualified_name] = make_scorer(qualified_name, partial(_metric, pos_label=None, average=average), needs_class=True)
         _add_scorer_to_metric_dict(metric_dict=BINARY_METRICS, scorer=globals()[qualified_name])
         _add_scorer_to_metric_dict(metric_dict=MULTICLASS_METRICS, scorer=globals()[qualified_name])
 
 
-for name, metric, kwargs in [
+for _name, _metric, _kwargs in [
     ("roc_auc_ovo", customized_roc_auc, dict(multi_class="ovo")),
     ("roc_auc_ovr", customized_roc_auc, dict(multi_class="ovr")),
 ]:
     scorer_kwargs = dict(greater_is_better=True, needs_proba=True, needs_threshold=False)
-    globals()[name] = make_scorer(name, partial(metric, average="macro", **kwargs), **scorer_kwargs)
-    macro_name = "{0}_{1}".format(name, "macro")
-    globals()[name].add_alias(macro_name)
-    _add_scorer_to_metric_dict(metric_dict=MULTICLASS_METRICS, scorer=globals()[name])
-    if name == "roc_auc_ovo":
+    globals()[_name] = make_scorer(_name, partial(_metric, average="macro", **_kwargs), **scorer_kwargs)
+    macro_name = "{0}_{1}".format(_name, "macro")
+    globals()[_name].add_alias(macro_name)
+    _add_scorer_to_metric_dict(metric_dict=MULTICLASS_METRICS, scorer=globals()[_name])
+    if _name == "roc_auc_ovo":
         averages = ["weighted"]
     else:
         averages = ["micro", "weighted"]
     for average in averages:
-        qualified_name = "{0}_{1}".format(name, average)
-        globals()[qualified_name] = make_scorer(qualified_name, partial(metric, average=average, **kwargs), **scorer_kwargs)
+        qualified_name = "{0}_{1}".format(_name, average)
+        globals()[qualified_name] = make_scorer(qualified_name, partial(_metric, average=average, **_kwargs), **scorer_kwargs)
         _add_scorer_to_metric_dict(metric_dict=MULTICLASS_METRICS, scorer=globals()[qualified_name])