CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[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

[2.2.1] - 2022-05-07

Added

• You can now perform fit and fitfast for one line segment!

[2.2.0] - 2022-05-01

Added

• Now you can specify a numpy.random.seed to use on init to get reproducible results from the fit and fitfast methods. Simply specify an integer seed number like so pwlf.PiecewiseLinFit(x, y, seed=123). Note this hijacks your current random seed. By default, not random seed is specified.
• Python 3.9 is now part of the ci

Changed

• Add flake8 checks to tests
• Two tests were not being checked because the method did not start with test

[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!

[2.0.5] - 2021-12-04

Added

• conda forge installs are now officially mentioned on readme and in documentation

[2.0.4] - 2020-08-27

Deprecated

• Python 2.7 will not be supported in future releases
• Python 3.5 reaches end-of-life on 2020-09-13, and will not be supported in future releases

Added

• python_requires in setup.py

[2.0.3] - 2020-06-26

Changed

• version handling does not require any dependencies

Removed

• importlib requirement

[2.0.2] - 2020-05-25

Changed

• Fixed an encoding bug that would not let pwlf install on windows. Thanks to h-vetinari for the PR!

[2.0.1] - 2020-05-24

Changed

• Removed setuptools for importlib single source versioning

Added

• Requirement for importlib-metadata if Python version is less than 3.8

Removed

• Requriement for setuptools

[2.0.0] - 2020-04-02

Added

• Added supports for pwlf to fit to weighted data sets! Check out this example.

Changed

• Setup.py now grabs markdown file for long description

Removed

• Tensorflow support has been removed. It hasn't been updated in a long time. If you still require this object, check out pwlftf

[1.1.7] - 2020-02-05

Changed

• Minimum SciPy version is now 1.2.0 because of issues with MacOS and the old SciPy versions. See issue #40 and thanks to bezineb5 !

[1.1.6] - 2020-01-22

Changed

• Single source version now found in setup.py instead of pwlf/VERSION see issue #53
• New setuptools requirement to handle new version file
• Fix bug where forcing pwlf through points didn't work with higher degrees. See issue #54

[1.1.5] - 2019-11-21

Changed

• Fix minor typo in docstring of calc_slopes
• Initialized all attributes in the __init__ funciton
• All attributes are now documented in the __init__ function. To view this docstring, use pwlf.PiecewiseLinFit?.

[1.1.4] - 2019-10-24

Changed

• TensorFlow 2.0.0 is not (and most probably will not) be supported. DepreciationWarning is displayed when using the PiecewiseLinFitTF object. Setup.py checks for this optional requirement. Tests are run on Tensorflow<2.0.0.
• TravisCi now checks Python version 3.7 in addition to 3.6, 3.5, 2.7.
• TravisCi tests should now be run daily.

[1.1.3] - 2019-09-14

Changed

• Make .ssr stored with fit_with_break* functions

[1.1.2] - 2019-08-19

Changed

• Bug fix in non-linear standard error, predict was calling y instead of x. #46 Thanks to @tcanders

[1.1.1] - 2019-08-18

Changed

• Raise the correct AttributeError when a fit has not yet been performed

[1.1.0] - 2019-06-16

Added

• Now you can calculate standard errors for non-linear regression using the Delta method! Check out this example.

[1.0.1] - 2019-06-15

Added

• Now you can fit constants and continuous polynomials with pwlf! Just specify the keyword degree= when initializing the PiecewiseLinFit object. Note that degree=0 for constants, degree==1 for linear (default), degree==2 for quadratics, etc.
• You can manually specify the optimization bounds for each breakpoint when calling the fit functions by using the bounds= keyword. Check out the related example.

Changed

• n_parameters is now calculated based on the shape of the regression matrix
• assembly of the regression matrix now considers which degree polynomial
• n_segments calculated from break points...
• Greatly reduce teststf.py run time

[1.0.0] - 2019-05-16

Changed

• Numpy matrix assembly is now ~100x times faster, which will translate to much faster fits! See this comment about the speed up. There should no longer be any performance benefits with using the PiecewiseLinFitTF (TensorFlow) object, so the only reason to use PiecewiseLinFitTF is if you want access to TensorFlow's optimizers.

Removed

• There are no sort or order optional parameters in PiecewiseLinFit. The new matrix assembly method doesn't need sorted data. This may break backwards compatibility with your code.

[0.5.1] - 2019-05-05

Changed

• Fixed PiecewiseLinFitTF for Python 2.

[0.5.0] - 2019-04-15

Added

• New PiecewiseLinFitTF class which uses TensorFlow to accelerate pwlf. This class is nearly identical to PiecewiseLinFit, with the exception of the removed sorted_data options. If you have TensorFlow installed you'll be able to use PiecewiseLinFitTF. If you do not have TensorFlow installed, importing pwlf will issue a warning that PiecewiseLinFitTF is not available. See this blog post for more information and benchmarks. The new class includes an option to use float32 or float64 data types.
• lapack_driver option to choose between the least squares backend. For more see https://docs.scipy.org/doc/scipy/reference/generated/scipy.linalg.lstsq.html and http://www.netlib.org/lapack/lug/node27.html

Changed

• Now use scipy.linalg instead of numpy.linalg because scipy is always compiled with lapack
• least squares fit now defaults to scipy instead of numpy

Removed

• rcond optional parameter; was not necessary with scipy.linalg

[0.4.3] - 2019-04-02

Changed

• You can now manually specify rcond for the numpy least squares solver. For more see #21 .

[0.4.2] - 2019-03-22

Changed

• assemble_regression_matrix() now checks if breaks is a numpy array

[0.4.1] - 2019-03-18

Changed

• p_values() now uses the correct degrees of freedom (Thanks to Tong Qiu for pointing this out!)
• p_values() now returns a value that can be compared to alpha; previous values would have been compared to alpha/2

[0.4.0] - 2019-03-14

Added

• new assemble_regression_matrix() function that returns the linear regression matrix A which can allow you to do some more complicated fits
• test function for the linear regression matrix
• new fit_guess() function to perform a fit when you have an estimate of the breakpoint locations

Changed

• consolidated the assembly of the linear regression matrix to a single function (and removed the duplicate code)

[0.3.5] - 2019-02-25

Changed

• minor correction to r_squared docstring example

[0.3.4] - 2019-02-06

Added

• Uploaded paper and citation information
• Added example of what happens when you have more unknowns than data

Changed

• Examples now include figures

[0.3.3] - 2019-01-32

Added

• Documentation and link to documentation

Changed

• Minor changes to docstrings (spelling, formatting, attribute fixes)

[0.3.2] - 2019-01-24

Added

• y-intercepts are now calculated when running the calc_slopes() function. The Y-intercept for each line is stored in self.intercepts.

[0.3.1] - 2018-12-09

Added

• p_values() function to calculate the p-value of each beta parameter (WARNING! you may only want to use this if you specify the break point locations, as this does not account for uncertainty in break point locations)

Changed

• Now changes stored in CHANGELOG.md

[0.3.0] - 2018-12-05

Added

• r_squared() function to calculate the coefficent of determination after a fit has been performed

Changed

• complete docstring overhaul to match the numpydoc style
• fix issues where sum-of-squares of residuals returned array_like, now should always return float

Removed

• legacy piecewise_lin_fit object has been removed

[0.2.10] - 2018-12-04

Changed

• Minor docstring changes
• Fix spelling mistakes throughout
• Fix README.rst format for PyPI

[0.0.X - 0.2.9] - from 2017-04-01 to 2018-10-03

• 2018/10/03 Add example of bare minimum model persistance to predict for new data (see examples/model_persistence_prediction.py). Bug fix in predict function for custom parameters. Add new test function to check that predict works with custom parameters.
• 2018/08/11 New function which calculates the predication variance for given array of x locations. The predication variance is the squared version of the standard error (not to be confused with the standard errors of the previous change). New example prediction_variance.py shows how to use the new function.
• 2018/06/16 New function which calculates the standard error for each of the model parameters (Remember model parameters are stored as my_pwlf.beta). Standard errors are calculated by calling se = my_pwlf.standard_errors() after you have performed a fit. For more information about standard errors see this. Fix docstrings for all functions.
• 2018/05/11 New sorted_data key which can be used to avoided sorting already ordered data. If your data is already ordered as x[0] < x[1] < ... < x[n-1], you may consider using sorted_data=True for a slight performance increase. Additionally the predict function can take the sorted_data key if the data you want to predict at is already sorted. Thanks to V-Kh for the idea and PR.
• 2018/04/15 Now you can find piecewise linear fits that go through specified data points! Read this post for the details.
• 2018/04/09 Intelligently converts your x, y, or breaks to be numpy array.
• 2018/04/06 Speed! pwlf just got better and faster! A vast majority of this library has been entirely rewritten! New naming convention. The class piecewise_lin_fit() is being depreciated, now use the class PiecewiseLinFit(). See this post for details on the new formulation. New test function that tests predict().
• 2018/03/25 Default now hides optimization results. Use disp_res=True when initializing piecewise_lin_fit to change. The multi-start fitfast() function now defaults to the minimum population of 2.
• 2018/03/11 Added try/except behavior for fitWithBreaks function such that the function could be used in an optimization routine. In general when you have a singular matrix, the function will now return np.inf.
• 2018/02/16 Added new fitfast() function which uses multi-start gradient optimization instead of Differential Evolution. It may be substantially faster for your application. Also it would be a good candidate if you don't need the best solution, but just a reasonable fit. Fixed bug in tests function where assert was checking bound, not SSr. New requirement, pyDOE library. New 0.1.0 Version.
• 2017/11/03 add setup.py, new tests folder and test scripts, new version tracking, initialize break0 breakN in the beginning
• 2017/10/31 bug fix related to the case where break points exactly equal to x data points ( as per issue #1 ) and added attributes .sep_data_x, .sep_data_y, .sep_predict_data_x for troubleshooting issues related to the separation of data points to their respective regions
• 2017/10/20 remove determinant calculation and use try-except instead, this will offer a larger performance boost for big problems. Change library name to something more Pythonic. Add version attribute.
• 2017/08/03 gradients (slopes of the line segments) now stored as piecewise_lin_fit.slopes (or myPWLF.slopes) after they have been calculated by performing a fit or predicting
• 2017/04/01 initial release