ucimlrepo

To install the ucimlrepo package in a Jupyter notebook, use the following command. After installation, restart the kernel before importing the module.

!pip3 install -U ucimlrepo

The fetch_ucirepo function returns a dotdict, which is a specialized dictionary that allows accessing keys as attributes. This is particularly useful for navigating the nested metadata and data structures returned by the API.

Access Patterns

• Data Access: dataset.data.features or dataset.data.targets.
• Metadata Access: dataset.metadata.intro_paper or dataset.metadata.additional_info.
• Variable Access: dataset.variables returns a pandas DataFrame, which is ideal for tabular inspection of feature roles and descriptions.

The fetch_ucirepo function loads a dataset from the UCI ML Repository, including dataframes and metadata. You must provide either a dataset id or a name (or a substring of the name) as a keyword argument. You cannot provide both.

from ucimlrepo import fetch_ucirepo

# Fetch by ID
dataset = fetch_ucirepo(id=45)

# Alternatively, fetch by name
dataset = fetch_ucirepo(name='Heart Disease')

Use list_available_datasets() to print a list of datasets available for import. You can refine the list using the following optional keyword arguments:

• filter: Filter datasets based on a category (e.g., filter='aim-ahead').
• search: Search for datasets whose name contains the provided query string.

from ucimlrepo import list_available_datasets

# List all
list_available_datasets()

# Search for a specific name
list_available_datasets(search='Heart')

The object returned by fetch_ucirepo contains the following structured data:

### data (pandas DataFrames)
- ids
- features
- targets
- original
- headers (list)

### metadata
- uci_id
- name
- abstract
- area
- task
- characteristics
- num_instances
- num_features
- feature_types
- target_col
- index_col
- has_missing_values
- missing_values_symbol
- year_of_dataset_creation
- dataset_doi
- creators
- intro_paper
- repository_url
- data_url
- additional_info (dict with keys: summary, purpose, funding, instances_represent, recommended_data_splits, sensitive_data, preprocessing_description, variable_info, citation)
- external_url

### variables (DataFrame)
- name
- role
- type
- demographic
- description
- units
- missing_values

The fetch_ucirepo function allows you to import datasets from the UCI Machine Learning Repository using either a unique integer id or a string name.

from ucimlrepo import fetch_ucirepo

# Import by ID
sepsis = fetch_ucirepo(id=827)

# Import by Name
glioma = fetch_ucirepo(name='glioma')

Use list_available_datasets() to see which datasets are available for import. You can refine the list using the filter parameter for specific projects (e.g., filter='aim-ahead') or the search parameter to find datasets matching a specific string (e.g., search='diabe').

from ucimlrepo import list_available_datasets

# List all datasets
list_available_datasets()

# Filter by project
list_available_datasets(filter='aim-ahead')

# Search for specific keywords
list_available_datasets(search='diabe')

The .variables attribute provides detailed information about each variable in the dataset, including its type and role, presented in a tabular DataFrame format.

# Display variable information in a DataFrame format
print(sepsis.variables)

The data attribute of a fetched dataset object provides access to the actual data stored in pandas DataFrames.

Available data components:

• data.features: The feature set (independent variables).
• data.targets: The target set (dependent variables).
• data.ids: Dataset identifiers (if available).
• data.original: A combined DataFrame containing features, targets, and IDs.
• data.headers: The column headers for the data.

# Accessing data for model training
X = iris.data.features
y = iris.data.targets

# Accessing other components
print(iris.data.original)
print(iris.data.headers)

Once a dataset is fetched, you can access its metadata attribute. This contains information such as the target column name, the title of the introductory paper, and additional summaries.

Common metadata fields include:

• metadata.target_col: The name of the target column.
• metadata.intro_paper.title: The title of the paper associated with the dataset.
• metadata.additional_info.summary: A summary of the dataset.

import pprint

# Access full metadata
pprint.pp(sepsis.metadata)

# Access specific metadata fields
print(sepsis.metadata.target_col)
print(sepsis.metadata.intro_paper.title)
print(sepsis.metadata.additional_info.summary)

Use list_available_datasets() to print a formatted table of datasets available for import via Python. By default, it filters for datasets compatible with Python.

Parameters

• filter (str, optional): A label to filter datasets (e.g., 'python').
• search (str, optional): A substring to search for in dataset names.
• area (str, optional): A subject area to filter by.

from ucimlrepo import list_available_datasets

# List all python-compatible datasets
list_available_datasets()

# Search for specific datasets
list_available_datasets(search='Heart')

# Filter by area
list_available_datasets(area='medical')