[CLI]: No module named 'pkg_resources' while import wandb

[YiranJing]

Describe the bug 🐛

without setuptools installation, I get error No module named 'pkg_resources' at line import wandb

Reason: pkg_resources belongs to setuptools, but wandb library missed setuptools dependency in requrienment.txt

To reproduce the error, run

pip uninstall setuptools
pip uninstall wandb
pip install wandb
python3 example.py

example.py:

import wandb
import uuid


from sklearn import datasets
from sklearn import ensemble
from sklearn import preprocessing
from sklearn import model_selection
from sklearn.metrics import f1_score

ENTITY = "entity"
PROJECT = "sweep-test"

def train():

    sweep_config = {
        "name": f"sweep-{str(uuid.uuid4())[:8]}",
        "method": "bayes",
        "metric": {"name": f"accuracy", "goal": "maximize"},
        "parameters": {
            "max_depth": {
                "distribution": "int_uniform",
                "min": 2,
                "max": 5,
            },
        },
    }

    sweep_id = wandb.sweep(sweep_config, project=PROJECT, entity=ENTITY)
    print("Sweep:", sweep_id)

    def _func() -> None:
        with wandb.init() as run:

            estimator_params = {k: v for k, v in wandb.config.items()}

            clf = ensemble.RandomForestClassifier(
                max_depth=wandb.config["max_depth"],
                n_estimators=3,
                criterion="gini",
                min_samples_split=5,
            )
            X, y = datasets.make_circles(noise=0.2, factor=0.5, random_state=0)
            X = preprocessing.StandardScaler().fit_transform(X)
            X_train, X_test, y_train, y_test = model_selection.train_test_split(
                X, y, test_size=0.4, random_state=0
            )
            clf.fit(X_train, y_train)
            
            accuracy = clf.score(X_test, y_test)
            print("Accuracy:", accuracy)

            y_pred = clf.predict(X_test)
            f1 = f1_score(y_test, y_pred, zero_division=1)
            print("f1:", f1)

            wandb.log({"accuracy": accuracy, "f1": f1})

    wandb.agent(sweep_id, function=_func, count=3)


   
if __name__ == "__main__":
    
    train()

full error message using example.py

Traceback (most recent call last):
  File "example.py", line 2, in <module>
    import wandb
  File "foo/python/.venv/lib/python3.9/site-packages/wandb/__init__.py", line 27, in <module>
    from wandb import sdk as wandb_sdk
  File "foo/python/.venv/lib/python3.9/site-packages/wandb/sdk/__init__.py", line 7, in <module>
    from .wandb_artifacts import Artifact  # noqa: F401
  File "foo/python/.venv/lib/python3.9/site-packages/wandb/sdk/wandb_artifacts.py", line 30, in <module>
    from wandb.apis import InternalApi, PublicApi
  File "foo/python/.venv/lib/python3.9/site-packages/wandb/apis/__init__.py", line 32, in <module>
    from .internal import Api as InternalApi  # noqa
  File "foo/python/.venv/lib/python3.9/site-packages/wandb/apis/internal.py", line 1, in <module>
    from wandb.sdk.internal.internal_api import Api as InternalApi
  File "foo/python/.venv/lib/python3.9/site-packages/wandb/sdk/internal/internal_api.py", line 12, in <module>
    from pkg_resources import parse_version  # type: ignore
ModuleNotFoundError: No module named 'pkg_resources'

Environment

WandB version: 0.12.11

Python version: 3.9.9

pip version: 21.3

OS

OSX and Linux

[thundergolfer]

Can confirm the reproduction instructions produce the same error trace for me. The setuptools package is a dependency of pip so it's almost always available in typical Python environments, but in more hermetic Python environments/deployments Weights & Biases' undeclared dependency on setuptools creates an import error.

[thundergolfer]

Related issue in another project: snowflakedb/snowflake-connector-python#546

[armanhar]

Thanks everyone for flagging. I'll file a ticket so our team can address this.

[exalate-issue-sync]

Arman Harutyunyan commented:
Hey everyone, this should be fixed now. Please let me know if you encounter the issue again.

[thundergolfer]

Thanks, looks like the fix landed here: #3516.

It will show up in the next release. cc @YiranJing

[thundergolfer]

Closeable as fixed by #3516