v2.4.0

[2.4.0] - 2024-12-31

Added

• Added support for mixing degree fits. You may now specify degree=[0,1,0] to fit a constant, then linear, then constant line. These are discontinuous piecewise linear. Currently this only supports mixing degrees of 0 and 1. Thanks to wonch002 for resurrecting an old branch.

Changed

• fixed a bug where seed=0 would net get set. Thanks to filippobistaffa from #118 .

v2.3.0

[2.3.0] - 2024-10-26

Changed

• You can now force fits of one line segment to go through a particular point mypwlf.fit(1, x_c, y_c). Thanks to fredrikhellman.
• Update docs to sphinx 8 (from 4)
• Update ci version to python==3.9, python==3.10, python==3.11, and python==3.12

v2.2.1

You can now perform fit and fitfast for one line segment! Both routines do the same linear regression in this case.

v2.2.0

You can now manually specify a random seed on init. This will give you reproducible results with the fit and fitfast methods which are stochastic.

pwlf.PiecewiseLinFit(x, y, seed=123)

v2.1.0

[2.1.0] - 2022-03-31

Changed

• All instances of linalg.inv now use linalg.pinv. All APIs are still the same, but this is potentially a backwards breaking change as previous results may be different from new results. This will mainly affect standard error calculations.
• Previously calc_slopes was called after every least squares fit in optimization routines trying to find breakpoints. This would occasionally raise a numpy RuntimeWarning if two breakpoints were the same, or if a breakpoint was on the boundary. Now calc_slopes is not called during experimental breakpoint calculation, which should no longer raise this warning for most users. Slopes will still be calculated once optimal breakpoints are found!

v2.0.5

conda force installs are now officially mentioned on readme and in documentation

v2.0.4

This is version 2.0.4 of pwlf. It will be the last version to support Python<3.6.