OpenML Python API

0.15.1 verified Fri May 01 auth: no python

OpenML Python API for downloading, uploading, and managing datasets, tasks, runs, and flows on OpenML.org. Current version 0.15.1, requires Python >=3.8. Released under BSD license. Active development with periodic releases.

pip install openml

Common errors

error openml.exceptions.OpenMLServerError: The request has not been authorized (401) ↓

cause Missing or invalid OpenML API key.

fix

Set API key via environment variable OPENML_API_KEY or openml.config.apikey = '...'

error AttributeError: 'int' object has no attribute 'get' (when calling get_dataset with wrong ID) ↓

cause Dataset ID does not exist or is not an integer. Function expects integer ID but may receive something else.

fix

Ensure dataset ID is valid integer. Use list_datasets() to check existing IDs.

error ModuleNotFoundError: No module named 'openml' ↓

cause OpenML not installed or installed in different environment.

fix

Run pip install openml and ensure correct Python environment is active.

Warnings

breaking In version 0.14, the `get_dataset` method changed from returning a tuple to returning a `OpenMLDataset` object. Calling `get_data()` now returns four values (X, y, categorical_indicator, attribute_names) instead of previously different structure. ↓

fix Update code to use `X, y, categorical, names = dataset.get_data(target=...)`.

deprecated `openml.datasets.list_datasets()` is deprecated; use `openml.datasets.get_dataset_list()` or pass parameters to `list_datasets()`. ↓

fix Replace `list_datasets()` with `get_dataset_list()` or use `list_datasets(size=...)` etc.

gotcha API key configuration requires either setting OPENML_API_KEY env var or creating a ~/.openml/config file. Without it, many functions raise OpenMLServerError (401). ↓

fix Set env var or file: `openml.config.apikey = 'your_apikey'`. Get key from openml.org.

Imports

openml
wrong
```
from openml import *
```
correct
```
import openml
```
Wildcard imports discouraged; use explicit submodules like openml.datasets.
openml.datasets
wrong
```
from openml.datasets import list_datasets
```
correct
```
from openml.datasets import get_dataset
```
list_datasets is deprecated, use openml.datasets.list_datasets() or openml.datasets.get_dataset_list().
openml.tasks
wrong
```
import openml.tasks as tasks
```
correct
```
from openml.tasks import get_task
```
That works, but ensure using correct function names.

Quickstart

Setup OpenML, list datasets, download and inspect the Iris dataset.

import openml
openml.config.apikey = openml.config.get_api_key()  # uses ~/.openml/config or environment
# List datasets
datasets = openml.datasets.list_datasets()
print(f"Number of datasets: {len(datasets)}")
# Download a dataset (iris)
dataset = openml.datasets.get_dataset(61)
X, y, categorical_names, attribute_names = dataset.get_data(target=dataset.default_target_attribute)
print(X.head())