Version 2.3.1

• accelerate rescale cluster periods
• update documentation and include autodeployment

Version 2.3.0

• Fix depreciated pandas functions
• fix sum for distribution representation incl. min max vals - now mean value of representation equals mean value of original time series
• add possibility to define segment representation
• extend the default example
• switch from travis to github workflow for ci

Version 2.2.2

• Fix Hypertuning class
• Set high as new default MILP solver
• Rework README

Version 2.1.0

Following functionality was added:

• a hyperparameter tuning meta class which is able to identify the optimal combination of typical periods and segments for a given time series dataset

Version 2.0.1

tsam release (2.0.1) includes the following new functionalities:

• Changed dependency of scikit-learn to make tsam conda-forge runnable.

Version 2.0.0

In tsam’s latest release (2.0.0) the following functionalities were included:

• A new comprehensive structure that allows for free cross-combination of clustering algorithms and cluster representations, e.g. centroids or medoids.
• A novel cluster representation method that precisely replicates the original time series value distribution in the aggregated time series based on “Hoffmann, Kotzur and Stolten (2021): The Pareto-Optimal Temporal Aggregation of Energy System Models (https://arxiv.org/abs/2111.12072)”
• Maxoids as representation algorithm which represents time series by outliers only based on “Sifa and Bauckhage (2017): Online k-Maxoids clustering”
• K-medoids contiguity: An algorithm based on “Oehrlein and Hauner (2017): A cutting-plane method for adjacency-constrained spatial aggregation” that accounts for contiguity constraints to e.g. cluster only time series in neighboring regions

Version 1.1.2

This tsam release (1.1.2) includes the following new functionalities

• Added first version of the k-medoid contiguity algorithm

Version 1.1.1

This tsam release (1.1.1) includes

• Significantly increased test coverage
• Separation between clustering and representation, i.e. for clustering algorithms like Ward’s hierarchical clustering algorithm the representation by medoids or centroids can now freely be chosen.

Include build testing

v1.01

prepare new release including testing

First pypi release

v0.9.9

finalize pypi release