hsbc/tslumen

A library for Time Series EDA (exploratory data analysis)

A library for exploratory analysis of Time Series data

tslumen helps bring to light the key characteristics of your time series data with rich, pre-canned artifacts, packed with charts and statistical information. The primary goal of tslumen is to expedite and bring consistency to how time series EDA is performed, allowing you to uncover the fundamental aspects in seconds rather than hours or days.

Key features

• Platform agnostic, integrates nicely with your datascience workspace
• Built on open source technology and research
• Highly customizable and extensible
• Data (profiling results) completely detached from the visuals
• Can be executed from the command line
• Efficient execution using parallel processing
• Includes a great number of statistical information, including descriptive statistics statistical tests like KPSS or ADF, correlation, tsfeatures, etc.
• Various plots specifically tailored to time series analysis
• Self-contained HTML report that can easily be shared with interested parties
• Fully interactive dashboard for a richer experience and detailed exploration

See https://hsbc.github.io/tslumen/ for the complete documentation.

Installation

From PyPI:

pip install -U tslumen

From source:

# cd into tslumen after cloning the repo
make install

Examples

Refer to the Quick Start page of the documentation for a brief tour of the package.

Complete example notebooks can be found on the User Guide section of the documentation.

Contributing

Contributions to tslumen are welcome. Please see our contribution guide for more details.

tslumen's Issues

Hierarchical time series summary

Describe the feature or enhancement you want to enable

Basically a selection option for when the data is arranged hierarchically according to a collection of categories. These hierarchies sometimes come from nested geographical features (e.g. country > province > region > district). They can also come from organizational structures, or parts of a facility, or some biological taxomonies, etc.

Being able to select from different levels of this hierarchies would be helpful in navigating hierarchical times series data.

Describe your proposed solution

Have an input to the HTML report constructor that receives a sequence of column names which are understood to be the hierarchical categories from top-to-bottom of the hierarchy. The implementation could ignore this option when it is None by default. But when it is given the HTML report constructs elements to have multiple selectors displayed in some order to specify the subset of the data of interest.

Describe alternatives you've considered, if relevant

Alternatively you can combine the different levels of categorical variables into tuples and use each tuple as label for each time series. This would create a very long list of options that may be less readible.

Additional context