Datasets¶

For a list of currently available datasets and usage instructions, see the datasets page.

License¶

To be considered for inclusion in statsmodels, a dataset must be in the public domain, distributed under a BSD-compatible license, or we must obtain permission from the original author.

Adding a dataset: An example¶

The Nile River data measures the volume of the discharge of the Nile River at Aswan for the years 1871 to 1970. The data are copied from the paper of Cobb (1978).

Step 1: Create a directory datasets/nile/

Step 2: Add datasets/nile/nile.csv and a new file datasets/__init__.py which contains

from data import *

Step 3: If nile.csv is a transformed/cleaned version of the original data, create a nile/src directory and include the original raw data there. In the nile case, this step is not necessary.

Step 4: Copy datasets/template_data.py to nile/data.py. Edit nile/data.py by filling-in strings for COPYRIGHT, TITLE, SOURCE, DESCRSHORT, DESCLONG, and NOTE.

COPYRIGHT   = """This is public domain."""
TITLE       = """Nile River Data"""
SOURCE      = """
Cobb, G.W. 1978. The Problem of the Nile: Conditional Solution to a Changepoint
    Problem. Biometrika. 65.2, 243-251,
"""

DESCRSHORT  = """Annual Nile River Volume at Aswan, 1871-1970""

DESCRLONG   = """Annual Nile River Volume at Aswan, 1871-1970. The units of
measurement are 1e9 m^{3}, and there is an apparent changepoint near 1898."""

NOTE        = """
Number of observations: 100
Number of variables: 2
Variable name definitions:
    year - Year of observation
    volume - Nile River volume at Aswan

The data were originally used in Cobb (1987, See SOURCE). The author
acknowledges that the data were originally compiled from various sources by
Dr. Barbara Bell, Center for Astrophysics, Cambridge, Massachusetts. The data
set is also used as an example in many textbooks and software packages.
"""

Step 5: Edit the docstring of the load function in data.py to specify which dataset will be loaded. Also edit the path and the indices for the endog and exog attributes. In the nile case, there is no exog, so everything referencing exog is not used. The year variable is also not used.

Step 6: Edit the datasets/__init__.py to import the directory.

That’s it! The result can be found here for reference.