Release Notes

This section discusses changes between versions, especially changes significant to the use and behavior of the library. This is not meant to be a comprehensive list of changes. For such a complete record, consult the lmfit GitHub repository.

Version 1.3.2 Release Notes (July 19, 2024)

Fixes:

  • fix typo in restoring a _buildmodel dict (PR #957, Issue #956)

  • fixes for Numpy2 support (PR #959, Issue #958)

  • ensure that correct initial params are used when re-fitting a ModeRresult (PR #961, Issue #960)

  • make sure that CompositeModels cannot have a prefix (PR #961, Issue #954)

Build, Maintenance:

  • update pre-commit hooks, adding codespell exceptions

  • update to latest SciPy/NumPy versions, including dependency versions for NumPy 2.

  • now require asteval>=1.0 and uncertainties>=3.2.2

Version 1.3.1 Release Notes (April 19, 2024)

Mostly fixes for bugs introduced in 1.3.0

  • allow Model.eval_uncertainty to be performed with single points for x independent variables (PR #952, Issue #951)

  • allow Model._parse_param to handle older-style passed-in ‘argnames’ and ‘kwargs’ as for variadic function, add test (PR #950)

  • better allow (or re-allow) Model function independent variables / keyword argumentss to be given non-default values at model creation time

  • add form as independent variable for builtin Step, Rectangle, Thermal Distribution models.

  • use a copy of sys.modules when iterating over it. (#949)

  • use Model._reprstring(long=True) for model Model.__repr__().

Version 1.3.0 Release Notes (April 4, 2024)

New features:

  • add 'min_rel_change' as optional variable in calculation of confidence intervals with

    Model.conf_interval(). (PR #937).

  • Model.eval_uncertainty now takes an optional dscale parameter (default value of 0.01) to

    set the step size for calculating derivatives (PR #933).

  • add calculation of predicted_interval to Model.eval_uncertainty (PR #933).

Bug fixes/enhancements:

  • restore best-fit parameter values for high accuracy values of constrained values (PR #907)

  • improvement to Model for the difference between Parameter, “independent variable”, and “option”. With this change, keyword arguments to model functions with non-numerice default values such as do_thing=True, or form='linear' has those arguments become clearly identified as independent variables,and use the provided values as default values. (PR #941)

  • better saving/loading saved states of Model now use dill, have several cleanups, and are now versioned for future-proofing. Also, propagate funcdets for Parameters when loading a Model. (PR #932, PR #934)

  • in the TNC method, maxfun is used instead of maxiter.

  • fix bug calculating r-squared for fits with weights (PR #921, PR #923)

  • fix bug in modelresult.eval_uncertainty() after load_modelresult() (PR #909)

  • use StringIO for pandas.read_json.

  • add test for MinimizerResult.uvars after successful fit (PR #913)

  • adding an example using basinhopping, can take other methods as command-line argument

Maintenance/Deprecations:

  • drop support for Python 3.7 that reached EOL on 2023-06-27 (PR #927)

  • fix tests for Python 3.12 and Python 3.13-dev

  • increase minimum numpy verstio to 1.23 and scipy to 1.8.

  • updates for compatibility with numpy 2.0

  • the dill package is now required. (#940)

  • build switchded to use pyproject.toml (#928)

  • fix broken links in Examples gallery

  • fix intersphinx mapping to scipy docs.

Version 1.2.2 Release Notes (July 14, 2023)

New features:

  • add ModelResult.uvars output to a ModelResult after a successful fit that contains ufloats from the uncertainties package which can be used for downstream calculations that propagate the uncertainties (and correlations) of the variable Parameters. (PR #888)

  • Outputs of residual functions, including Model._residual, are more explicitly coerced to 1d-arrays of datatype Float64. This decreases the expectation for the user-supplied code to return ndarrays, and increases the tolerance for more “array-like” objects or ndarrays that are not Float64 or 1-dimensional. (PR #899)

  • Model.fit now takes a coerce_farray option, defaulting to True to control whether to input data and independent variables that are “array-like” are coerced to ndarrays of datatype Float64 or Complex128. If set to False then independent data that “array-like” (pandas.Series, int32 arrays, etc) will be sent to the model function unaltered. The user may then use other features of these objects, but may also need to explicitly coerce the datatype of the result the change described above about coercing the result causes problems. (Discussion #873; PR #899)

Bug fixes/enhancements:

  • fixed bug in Model.make_params() for non-composite models that use a prefix (Discussion #892; Issue #893; PR #895)

  • fixed bug with aborted fits for several methods having incorrect or invalid fit statistics. (Discussion #894; Issue #896; PR #897)

  • Model.eval_uncertainty now correctly calculates complex (real/imaginary pairs) uncertainties for Models that generate complex results. (Issue #900; PR #901)

  • Model.eval now returns and array-like value. This adds to the coercion features above and fixes a bug for composite models that return lists (Issue #875; PR #901)

  • the HTML representation for a ModelResult or MinimizerResult are improved, and create fewer entries in the Table of Contents for Jupyter lab. (Issue #884; PR #883; PR #902)

Version 1.2.1 Release Notes (May 02, 2023)

Bug fixes/enhancements:

  • fixed bug in Model.make_params() for initial parameter values that were not recognized as floats such as np.Int64. (Issue #871; PR #872)

  • explicitly set maxfun for l-bfgs-b method when setting maxiter. (Issue #864; Discussion #865; PR #866)

Version 1.2.0 Release Notes (April 05, 2023)

New features:

  • add create_params function (PR #844)

  • add chi2_out and nsigma options to conf_interval2d()

  • add ModelResult.summary() to return many resulting fit statistics and attributes into a JSON-able dict.

  • add correl_table() function to lmfit.printfuncs and correl_mode option to fit_report() and ModelResult.fit_report() to optionally display a RST-formatted table of a correlation matrix.

Bug fixes/enhancements:

  • fix bug when setting param.vary=True for a constrained parameter (Issue #859; PR #860)

  • fix bug in reported uncertainties for constrained parameters by better propagating uncertainties (Issue #855; PR #856)

  • Coercing of user input data and independent data for Model to float64 ndarrays is somewhat less aggressive and will not increase the precision of numpy ndarrays (see Data Types for data and independent data with Model for details). The resulting calculation from a model or objective function is more aggressively coerced to float64. (Issue #850; PR #853)

  • the default value of epsfcn is increased to 1.e-10 to allow for handling of data with precision less than float64 (Issue #850; PR #853)

  • fix conf_interval2d to use “increase chi-square by sigma**2*reduced chi-square” to give the sigma-level probabilities (Issue #848; PR #852)

  • fix reading of older ModelResult (Issue #845; included in PR #844)

  • fix deepcopy of Parameters and user data (mguhyo; PR #837)

  • improve Model.make_params and create_params to take optional dict of Parameter attributes (PR #844)

  • fix reporting of nfev from least_squares to better reflect actual number of function calls (Issue #842; PR #844)

  • fix bug in Model.eval when mixing parameters and keyword arguments (PR #844, #839)

  • re-adds residual to saved Model result (PR #844, #830)

  • ConstantModel and ComplexConstantModel will return an ndarray of the same shape as the independent variable x (JeppeKlitgaard, Issue #840; PR #841)

  • update tests for latest versions of NumPy and SciPy.

  • many fixes of doc typos and updates of dependencies, pre-commit hooks, and CI.

Version 1.1.0 Release Notes (November 27, 2022)

New features:

  • add Pearson4Model (@lellid; PR #800)

  • add SplineModel (PR #804)

  • add R^2 rsquared statistic to fit outputs and reports for Model fits (Issue #803; PR #810)

  • add calculation of dely for model components of composite models (Issue #761; PR #826)

Bug fixes/enhancements:

  • make sure variable spercent is always defined in params_html_table functions (reported by @MySlientWind; Issue #768, PR #770)

  • always initialize the variables success and covar the MinimizerResult (reported by Marc W. Pound; PR #771)

  • build package following PEP517/PEP518; use pyproject.toml and setup.cfg; leave setup.py for now (PR #777)

  • components used to create a CompositeModel can now have different independent variables (@Julian-Hochhaus; Discussion #787; PR #788)

  • fixed function definition for StepModel(form='linear'), was not consistent with the other ones (@matpompili; PR #794)

  • fixed height factor for Gaussian2dModel, was not correct (@matpompili; PR #795)

  • for covariances with negative diagonal elements, we set the covariance to None (PR #813)

  • fixed linear mode for RectangleModel (@arunpersaud; Issue #815; PR #816)

  • report correct initial values for parameters with bounds (Issue #820; PR #821)

  • allow recalculation of confidence intervals (@jagerber48; PR #798)

  • include ‘residual’ in JSON output of ModelResult.dumps (@mac01021; PR #830)

  • supports and is tested against Python 3.11; updated minimum required version of SciPy, NumPy, and asteval (PR #832)

Deprecations:

  • remove support for Python 3.6 which reached EOL on 2021-12-23 (PR #790)

Version 1.0.3 Release Notes (October 14, 2021)

Potentially breaking change:

  • argument x is now required for the guess method of Models (Issue #747; PR #748)

To get reasonable estimates for starting values one should always supply both x and y values; in some cases it would work when only providing data (i.e., y-values). With the change above, x is now required in the guess method call, so scripts might need to be updated to explicitly supply x.

Bug fixes/enhancements:

  • do not overwrite user-specified figure titles in Model.plot() functions and allow setting with title keyword argument (PR #711)

  • preserve Parameters subclass in deepcopy (@jenshnielsen; PR #719)

  • coerce data and indepdent_vars to NumPy array with dtype=float64 or dtype=complex128 where applicable (Issues #723 and #728)

  • fix collision between parameter names in built-in models and user-specified parameters (Issue #710 and PR #732)

  • correct error message in PolynomialModel (@kremeyer; PR #737)

  • improved handling of altered JSON data (Issue #739; PR #740, reported by Matthew Giammar)

  • map max_nfev to maxiter when using differential_evolution (PR #749, reported by Olivier B.)

  • correct use of noise versus experimental uncertainty in the documentation (PR #751, reported by Andrés Zelcer)

  • specify return type of eval method more precisely and allow for plotting of (Complex)ConstantModel by coercing their float, int, or complex return value to a numpy.ndarray (Issue #684 and PR #754)

  • fix dho (Damped Harmonic Oscillator) lineshape (PR #755; @rayosborn)

  • reset Minimizer._abort to False before starting a new fit (Issue #756 and PR #757; @azelcer)

  • fix typo in guess_from_peak2d (@ivan-usovl; PR #758)

Various:

  • update asteval dependency to >= 0.9.22 to avoid DeprecationWarnings from NumPy v1.20.0 (PR #707)

  • remove incorrectly spelled DonaichModel and donaich lineshape, deprecated in version 1.0.1 (PR #707)

  • remove occurrences of OrderedDict throughout the code; dict is order-preserving since Python 3.6 (PR #713)

  • update the contributing instructions (PR #718; @martin-majlis)

  • (again) defer import of matplotlib to when it is needed (@zobristnicholas; PR #721)

  • fix description of name argument in Parameters.add (@kristianmeyerr; PR #725)

  • update dependencies, make sure a functional development environment is installed on Windows (Issue #712)

  • use setuptools_scm for version info instead of versioneer (PR #729)

  • transition to using f-strings (PR #730)

  • mark test_manypeaks_speed.py as flaky to avoid intermittent test failures (repeat up to 5 times; PR #745)

  • update scipy dependency to >= 1.14.0 (PR #751)

  • improvement to output of examples in sphinx-gallery and use higher resolution figures (PR #753)

  • remove deprecated functions lmfit.printfuncs.report_errors and asteval argument in Parameters class (PR #759)

Version 1.0.2 Release Notes (February 7, 2021)

Version 1.0.2 officially supports Python 3.9 and has dropped support for Python 3.5. The minimum version of the following dependencies were updated: asteval>=0.9.21, numpy>=1.18, and scipy>=1.3.

New features:

  • added two-dimensional Gaussian lineshape and model (PR #642; @mpmdean)

  • all built-in models are now registered in lmfit.models.lmfit_models; new Model class attribute valid_forms (PR #663; @rayosborn)

  • added a SineModel (PR #676; @lneuhaus)

  • add the run_mcmc_kwargs argument to Minimizer.emcee to pass to the emcee.EnsembleSampler.run_mcmc function (PR #694; @rbnvrw)

Bug fixes:

  • ModelResult.eval_uncertainty should use provided Parameters (PR #646)

  • center in lognormal model can be negative (Issue #644, PR #645; @YoshieraHuang)

  • restore best-fit values after calculation of covariance matrix (Issue #655, PR #657)

  • add helper-function not_zero to prevent ZeroDivisionError in lineshapes and use in exponential lineshape (Issue #631, PR #664; @s-weigand)

  • save last_internal_values and use to restore internal values if fit is aborted (PR #667)

  • dumping a fit using the lbfgsb method now works, convert bytes to string if needed (Issue #677, PR #678; @leonfoks)

  • fix use of callable Jacobian for scalar methods (PR #681; @mstimberg)

  • preserve float/int types when encoding for JSON (PR #696; @jedzill4)

  • better support for saving/loading of ExpressionModels and assure that init_params and init_fit are set when loading a ModelResult (PR #706)

Various:

  • update minimum dependencies (PRs #688, #693)

  • improvements in coding style, docstrings, CI, and test coverage (PRs #647, #649, #650, #653, #654; #685, #668, #689)

  • fix typo in Oscillator (PR #658; @flothesof)

  • add example using SymPy (PR #662)

  • allow better custom pool for emcee() (Issue #666, PR #667)

  • update NIST Strd reference functions and tests (PR #670)

  • make building of documentation cross-platform (PR #673; @s-weigand)

  • relax module name check in test_check_ast_errors for Python 3.9 (Issue #674, PR #675; @mwhudson)

  • fix/update layout of documentation, now uses the sphinx13 theme (PR #687)

  • fixed DeprecationWarnings reported by NumPy v1.2.0 (PR #699)

  • increase value of tiny and check for it in bounded parameters to avoid “parameter not moving from initial value” (Issue #700, PR #701)

  • add max_nfev to basinhopping and brute (now supported everywhere in lmfit) and set to more uniform default values (PR #701)

  • use Azure Pipelines for CI, drop Travis (PRs #696 and #702)

Version 1.0.1 Release Notes

Version 1.0.1 is the last release that supports Python 3.5. All newer version will require 3.6+ so that we can use formatting-strings and rely on dictionaries being ordered.

New features:

  • added thermal distribution model and lineshape (PR #620; @mpmdean)

  • introduced a new argument max_nfev to uniformly specify the maximum number of function evaluations (PR #610) Please note: all other arguments (e.g., ``maxfev``, ``maxiter``, …) will no longer be passed to the underlying solver. A warning will be emitted stating that one should use ``max_nfev``.

  • the attribute call_kws was added to the MinimizerResult class and contains the keyword arguments that are supplied to the solver in SciPy.

Bug fixes:

  • fixes to the load and __setstate__ methods of the Parameter class

  • fixed failure of ModelResult.dump() due to missing attributes (Issue #611, PR #623; @mpmdean)

  • guess_from_peak function now also works correctly with decreasing x-values or when using pandas (PRs #627 and #629; @mpmdean)

  • the Parameter.set() method now correctly first updates the boundaries and then the value (Issue #636, PR #637; @arunpersaud)

Various:

  • fixed typo for the use of expressions in the documentation (Issue #610; @jkrogager)

  • removal of PY2-compatibility and unused code and improved test coverage (PRs #619, #631, and #633)

  • removed deprecated isParameter function and automatic conversion of an uncertainties object (PR #626)

  • inaccurate FWHM calculations were removed from built-in models, others labeled as estimates (Issue #616 and PR #630)

  • corrected spelling mistake for the Doniach lineshape and model (Issue #634; @rayosborn)

  • removed unsupported/untested code for IPython notebooks in lmfit/ui/*

Version 1.0.0 Release Notes

Version 1.0.0 supports Python 3.5, 3.6, 3.7, and 3.8

New features:

  • no new features are introduced in 1.0.0.

Improvements:

  • support for Python 2 and use of the six package are removed. (PR #612)

Various:

  • documentation updates to clarify the use of emcee. (PR #614)

Version 0.9.15 Release Notes

Version 0.9.15 is the last release that supports Python 2.7; it now also fully supports Python 3.8.

New features, improvements, and bug fixes:

  • move application of parameter bounds to setter instead of getter (PR #587)

  • add support for non-array Jacobian types in least_squares (Issue #588, @ezwelty in PR #589)

  • add more information (i.e., acor and acceptance_fraction) about emcee fit (@j-zimmermann in PR #593)

  • “name” is now a required positional argument for Parameter class, update the magic methods (PR #595)

  • fix nvars count and bound handling in confidence interval calculations (Issue #597, PR #598)

  • support Python 3.8; requires asteval >= 0.9.16 (PR #599)

  • only support emcee version 3 (i.e., no PTSampler anymore) (PR #600)

  • fix and refactor prob_bunc in confidence interval calculations (PR #604)

  • fix adding Parameters with custom user-defined symbols (Issue #607, PR #608; thanks to @gbouvignies for the report)

Various:

  • bump requirements to LTS version of SciPy/ NumPy and code clean-up (PR #591)

  • documentation updates (PR #596, and others)

  • improve test coverage and Travis CI updates (PR #595, and others)

  • update pre-commit hooks and configuration in setup.cfg

To-be deprecated: - function Parameter.isParameter and conversion from uncertainties.core.Variable to value in _getval (PR #595)

Version 0.9.14 Release Notes

New features:

  • the global optimizers shgo and dual_annealing (new in SciPy v1.2) are now supported (Issue #527; PRs #545 and #556)

  • eval method added to the Parameter class (PR #550 by @zobristnicholas)

  • avoid ZeroDivisionError in printfuncs.params_html_table (PR #552 by @aaristov and PR #559)

  • add parallelization to brute method (PR #564, requires SciPy v1.3)

Bug fixes:

  • consider only varying parameters when reporting potential issues with calculating errorbars (PR #549) and compare value to both min and max (PR #571)

  • guard against division by zero in lineshape functions and FWHM and height expression calculations (PR #545)

  • fix issues with restoring a saved Model (Issue #553; PR #554)

  • always set result.method for emcee algorithm (PR #558)

  • more careful adding of parameters to handle out-of-order constraint expressions (Issue #560; PR #561)

  • make sure all parameters in Model.guess() use prefixes (PRs #567 and #569)

  • use inspect.signature for PY3 to support wrapped functions (Issue #570; PR #576)

  • fix result.nfev` for brute method when using parallelization (Issue #578; PR #579)

Various:

  • remove “missing” in the Model class (replaced by nan_policy) and “drop” as option to nan_policy (replaced by omit) deprecated since 0.9 (PR #565).

  • deprecate ‘report_errors’ in printfuncs.py (PR #571)

  • updates to the documentation to use jupyter-sphinx to include examples/output (PRs #573 and #575)

  • include a Gallery with examples in the documentation using sphinx-gallery (PR #574 and #583)

  • improve test-coverage (PRs #571, #572 and #585)

  • add/clarify warning messages when NaN values are detected (PR #586)

  • several updates to docstrings (Issue #584; PR #583, and others)

  • update pre-commit hooks and several docstrings

Version 0.9.13 Release Notes

New features:

  • Clearer warning message in fit reports when uncertainties should but cannot be estimated, including guesses of which Parameters to examine (#521, #543)

  • SplitLorenztianModel and split_lorentzian function (#523)

  • HTML representations for Parameter, MinimizerResult, and Model so that they can be printed better with Jupyter (#524, #548)

  • support parallelization for differential evolution (#526)

Bug fixes:

  • delay import of matplotlib (and so, the selection of its backend) as late as possible (#528, #529)

  • fix for saving, loading, and reloading ModelResults (#534)

  • fix to leastsq to report the best-fit values, not the values tried last (#535, #536)

  • fix synchronization of all parameter values on Model.guess() (#539, #542)

  • improve deprecation warnings for outdated nan_policy keywords (#540)

  • fix for edge case in gformat() (#547)

Project management:

  • using pre-commit framework to improve and enforce coding style (#533)

  • added code coverage report to github main page

  • updated docs, github templates, added several tests.

  • dropped support and testing for Python 3.4.

Version 0.9.12 Release Notes

Lmfit package is now licensed under BSD-3.

New features:

  • SkewedVoigtModel was added as built-in model (Issue #493)

  • Parameter uncertainties and correlations are reported for least_squares

  • Plotting of complex-valued models is now handled in ModelResult class (PR #503)

  • A model’s independent variable is allowed to be an object (Issue #492)

  • Added usersyms to Parameters() initialization to make it easier to add custom functions and symbols (Issue #507)

  • the numdifftools package can be used to calculate parameter uncertainties and correlations for all solvers that do not natively support this (PR #506)

  • emcee can now be used as method keyword-argument to Minimizer.minimize and minimize function, which allows for using emcee in the Model class (PR #512; see examples/example_emcee_with_Model.py)

(Bug)fixes:

  • asteval errors are now flushed after raising (Issue #486)

  • max_time and evaluation time for ExpressionModel increased to 1 hour (Issue #489)

  • loading a saved ModelResult now restores all attributes (Issue #491)

  • development versions of scipy and emcee are now supported (Issue #497 and PR #496)

  • ModelResult.eval() do no longer overwrite the userkws dictionary (Issue #499)

  • running the test suite requires pytest only (Issue #504)

  • improved FWHM calculation for VoigtModel (PR #514)

Version 0.9.10 Release Notes

Two new global algorithms were added: basinhopping and AMPGO. Basinhopping wraps the method present in scipy, and more information can be found in the documentation (basinhopping() and scipy.optimize.basinhopping). The Adaptive Memory Programming for Global Optimization (AMPGO) algorithm was adapted from Python code written by Andrea Gavana. A more detailed explanation of the algorithm is available in the AMPGO paper and specifics for lmfit can be found in the ampgo() function.

Lmfit uses the external uncertainties (https://github.com/lebigot/uncertainties) package (available on PyPI), instead of distributing its own fork.

An AbortFitException is now raised when the fit is aborted by the user (i.e., by using iter_cb).

Bugfixes:

  • all exceptions are allowed when trying to import matplotlib

  • simplify and fix corner-case errors when testing closeness of large integers

Version 0.9.9 Release Notes

Lmfit now uses the asteval (https://github.com/newville/asteval) package instead of distributing its own copy. The minimum required asteval version is 0.9.12, which is available on PyPI. If you see import errors related to asteval, please make sure that you actually have the latest version installed.

Version 0.9.6 Release Notes

Support for SciPy 0.14 has been dropped: SciPy 0.15 is now required. This is especially important for lmfit maintenance, as it means we can now rely on SciPy having code for differential evolution and do not need to keep a local copy.

A brute force method was added, which can be used either with Minimizer.brute() or using the method='brute' option to Minimizer.minimize(). This method requires finite bounds on all varying parameters, or that parameters have a finite brute_step attribute set to specify the step size.

Custom cost functions can now be used for the scalar minimizers using the reduce_fcn option.

Many improvements to documentation and docstrings in the code were made. As part of that effort, all API documentation in this main Sphinx documentation now derives from the docstrings.

Uncertainties in the resulting best-fit for a model can now be calculated from the uncertainties in the model parameters.

Parameters have two new attributes: brute_step, to specify the step size when using the brute method, and user_data, which is unused but can be used to hold additional information the user may desire. This will be preserved on copy and pickling.

Several bug fixes and cleanups.

Versioneer was updated to 0.18.

Tests can now be run either with nose or pytest.

Version 0.9.5 Release Notes

Support for Python 2.6 and SciPy 0.13 has been dropped.

Version 0.9.4 Release Notes

Some support for the new least_squares routine from SciPy 0.17 has been added.

Parameters can now be used directly in floating point or array expressions, so that the Parameter value does not need sigma = params['sigma'].value. The older, explicit usage still works, but the docs, samples, and tests have been updated to use the simpler usage.

Support for Python 2.6 and SciPy 0.13 is now explicitly deprecated and will be dropped in version 0.9.5.

Version 0.9.3 Release Notes

Models involving complex numbers have been improved.

The emcee module can now be used for uncertainty estimation.

Many bug fixes, and an important fix for performance slowdown on getting parameter values.

ASV benchmarking code added.

Version 0.9.0 Release Notes

This upgrade makes an important, non-backward-compatible change to the way many fitting scripts and programs will work. Scripts that work with version 0.8.3 will not work with version 0.9.0 and vice versa. The change was not made lightly or without ample discussion, and is really an improvement. Modifying scripts that did work with 0.8.3 to work with 0.9.0 is easy, but needs to be done.

Summary

The upgrade from 0.8.3 to 0.9.0 introduced the MinimizerResult class (see MinimizerResult – the optimization result) which is now used to hold the return value from minimize() and Minimizer.minimize(). This returned object contains many goodness of fit statistics, and holds the optimized parameters from the fit. Importantly, the parameters passed into minimize() and Minimizer.minimize() are no longer modified by the fit. Instead, a copy of the passed-in parameters is made which is changed and returns as the params attribute of the returned MinimizerResult.

Impact

This upgrade means that a script that does:

my_pars = Parameters()
my_pars.add('amp', value=300.0, min=0)
my_pars.add('center', value=5.0, min=0, max=10)
my_pars.add('decay', value=1.0, vary=False)

result = minimize(objfunc, my_pars)

will still work, but that my_pars will NOT be changed by the fit. Instead, my_pars is copied to an internal set of parameters that is changed in the fit, and this copy is then put in result.params. To look at fit results, use result.params, not my_pars.

This has the effect that my_pars will still hold the starting parameter values, while all of the results from the fit are held in the result object returned by minimize().

If you want to do an initial fit, then refine that fit to, for example, do a pre-fit, then refine that result different fitting method, such as:

result1 = minimize(objfunc, my_pars, method='nelder')
result1.params['decay'].vary = True
result2 = minimize(objfunc, result1.params, method='leastsq')

and have access to all of the starting parameters my_pars, the result of the first fit result1, and the result of the final fit result2.

Discussion

The main goal for making this change were to

  1. give a better return value to minimize() and Minimizer.minimize() that can hold all of the information about a fit. By having the return value be an instance of the MinimizerResult class, it can hold an arbitrary amount of information that is easily accessed by attribute name, and even be given methods. Using objects is good!

  2. To limit or even eliminate the amount of “state information” a Minimizer holds. By state information, we mean how much of the previous fit is remembered after a fit is done. Keeping (and especially using) such information about a previous fit means that a Minimizer might give different results even for the same problem if run a second time. While it’s desirable to be able to adjust a set of Parameters re-run a fit to get an improved result, doing this by changing an internal attribute (Minimizer.params) has the undesirable side-effect of not being able to “go back”, and makes it somewhat cumbersome to keep track of changes made while adjusting parameters and re-running fits.