Release History¶

An index with links to the documentation of the latest and former emgfit versions can be found here.

v0.3.2 (2020-12-04)¶

Resolved some incompatibility issues observed in notebooks for certain ipython and ipykernel versions.
Fixed failing notebook start-up due to pywin32 ImportError.

Fixed a bug causing crashes of parallelized fitting with get_errors_from_resampling() & get_MC_peakshape_errors() in Python3.8.
Fixed some deprecation warnings.
Added additional wait time to prevent Travis CI build from timing out prematurely.

Added emgfit.sample module for easy generation of simulated spectra with Gaussian and hyper-EMG line shapes.
Incorporated the option to perform blind analysis via the new set_blinded_peaks() method. The latter hides the obtained mass values and positions of user-defined peaks-of-interest.
Implemented get_errors_from_resampling() method which can yield refined estimates of the statistical and peak area errors by performing a parametric bootstrap for each fitted peak.
Added a Markov-Chain Monte Carlo sampling method (_get_MCMC_par_samples()) for mapping out the posterior distributions and correlations of model parameters. This method can be called with the map_par_covar option in the peak-shape determination.
Added a method ( get_MC_peakshape_errors()) for obtaining refined peak-shape error estimates that account for correlations and non-normal posterior distributions of shape parameters. This method relies on shape parameter sets obtained via Markov-Chain Monte Carlo sampling.
Added peak_indeces argument to fit_peaks() to enable automatic fit range selection from the specified indeces of interest.
Added fit_kws argument to peakfit method to enable more control over the underlying scipy optimization algorithms.
Updated emgfit tutorial with new uncertainty estimation methods.
Add concept articles and apply various edits to the documentation.

Changed bounding of Pearson weights to addition of small number eps = 1e-10 in the denominator of the Pearson chi-square residuals. This ensures that the cost function asymptotically converges to a chi-squared distribution while still avoiding convergence issues due to overweighting of bins whose predicted number of counts approach zero.
Changed automatic tail order determination in determine_peak_shape() method. Now tail orders are excluded if either the corresponding eta or tau parameter agrees with zero within 1-sigma confidence.
Extended peak-shape error evaluation methods to also estimate the corresponding peak area uncertainties and automatically add them in quadrature to the statistical peak area uncertainties.
Updated formatting of peak properties table for more clarity including color coding to indicate the way uncertainties have been estimated.

Fixed bug in mass re-scaling in peak-shape error evaluation.
Relevant for developers only: Further automatized the deployment of new releases.

Improved numerical robustness and speed of Hyper-EMG functions in emg_funcs module. The improved routines avoid arithmetic overflow of exp() or underflow of erfc().
Improved calculation of effective mass shifts in peak-shape error determination (_eval_peak_shape_errors() method). The IOI mass shifts are now corrected for shifts of the calibrant mass using shifted mass re-calibration factors instead of taking the simple mass difference between shifted IOI and calibrant centroids.
Added remove_peaks() method to spectrum class to allow removing multiple peaks at once, the remove_peak() method is deprecated but still supported.
Added upper bound of 1 to Pearson weights for increased numerical stability in fits with ‘chi-square’ cost function. Now Pearson_weights = 1./np.maximum(1.,np.sqrt(y_m)) where y_m is the model y-value in the foregoing fit iteration.
Improved handling of NaN values in calculation of negative log-likelihood ratio for MLE fit residuals.
Made determine_A_stat_emg() method more robust (better handling of ValueErrors due to NaNs in fit model y-values).

Initial Release