session_info

session_info outputs version information for modules loaded in the current session, Python, the OS, and the CPU. It is designed as a minimum measure to increase reproducibility and provides similar information as sessionInfo and devtools::session_info in R.

Motivation

session_info is particularly useful when conducting exploratory data analysis in Jupyter notebooks. Listing the version numbers of all loaded modules after importing them is a simple way to ensure a minimum level of reproducibility while requiring little additional effort. This practice is useful both when revisiting notebooks and when sharing them with colleagues. session_info is meant to complement more robust practices such as frozen virtual environments, containers, and binder.

Installation

session_info can be installed via pip install session-info or conda install -c conda-forge session-info. It does not depend on a package manager to find version numbers since it fetches them from the module's version string. Its only dependency is stdlib_list, which is used to distinguish between standard library and third party modules.

Usage

session_info can be used from a script like so:

import math

import natsort
import pandas
import session_info


session_info.show()

Output:

Session information:
-----
natsort             7.1.1
pandas              1.2.2
session_info        1.0.0
-----
IPython             7.23.0
jupyter_client      6.1.12
jupyter_core        4.7.1
-----
Python 3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:02:46) [GCC 9.3.0]
Linux-5.11.13-arch1-1-x86_64-with-glibc2.33
-----
Session information updated at 2021-05-06 09:59

The default behavior is to only output modules not in the standard library, which is why the math module is omitted above (it can be included by specifying std_lib=True). To include not only the explicitly imported modules, but also any dependencies they import internally, specify dependencies=True.

When session_info is invoked from a Jupyter Notebook, the output is concealed in <details> tags and will only show when clicked. Since this saves visual real estate, any modules imported indirectly as dependencies will be included by default and it looks like this:

-----
natsort             7.1.1
pandas              1.2.2
session_info        1.0.0
-----

backcall            0.2.0
cython_runtime      NA
dateutil            2.8.1
decorator           5.0.7
ipykernel           5.5.3
ipython_genutils    0.2.0
jedi                0.18.0
numpy               1.20.2
parso               0.8.2
pexpect             4.8.0
pickleshare         0.7.5
prompt_toolkit      3.0.18
ptyprocess          0.7.0
pygments            2.8.1
pytz                2021.1
six                 1.15.0
storemagic          NA
tornado             6.1
traitlets           5.0.5
wcwidth             0.2.5
zmq                 22.0.3

-----
IPython             7.23.0
jupyter_client      6.1.12
jupyter_core        4.7.1
-----
Python 3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:02:46) [GCC 9.3.0]
Linux-5.11.13-arch1-1-x86_64-with-glibc2.33
-----
Session information updated at 2021-05-06 09:59

If you prefer to show the session information without the HTML tags, you can use session_info.show(html=False) in the notebook to get the same output as in the first example above.

session_info can also write the module dependencies to a requirements.txt file, which is compatible with pip install -r /path/to/requirements.txt.

View the docstring for complete parameter information.

Background

session_info started as minor modifications of py_session, and as it grew it became convenient to create a new package. session_info was built with the help of information provided in stackoverflow answers and existing similar packages, including

• https://github.com/fbrundu/py_session
• https://github.com/jrjohansson/version_information
• https://github.com/rasbt/watermark
• https://stackoverflow.com/a/4858123/2166823
• https://stackoverflow.com/a/40690954/2166823
• https://stackoverflow.com/a/52187331/2166823