Changelog entries for the development version are available at

Hyperspy 1.7.5 (2023-05-04)

Bug Fixes

  • Fix plotting boolean array with plot_images() (#3118)

  • Fix test with scipy1.11 and update deprecated scipy.interpolate.interp2d in the test suite (#3124)

  • Use intersphinx links to fix links to scikit-image documentation (#3125)


  • Improve performance of model.multifit by avoiding axes.is_binned repeated evaluation (#3126)


  • Simplify release workflow and replace deprecated actions/create-release action with softprops/action-gh-release. (#3117)

  • Add support for python 3.11 (#3134)

  • Pin imageio to <2.28 (#3138)

Hyperspy 1.7.4 (2023-03-16)

Bug Fixes

  • Fixes an array indexing bug when loading a .sur file format spectra series. (#3060)

  • Speed up to_numpy() function to avoid slow down when used repeatedly, typically during fitting (#3109)

Improved Documentation

  • Replace sphinx.ext.imgmath by sphinx.ext.mathjax to fix the math rendering in the ReadTheDocs build (#3084)


  • Add support for Phenom .elid revision 3 and 4 formats (#3073)


  • Add pooch as test dependency, as it is required to use scipy.dataset in latest scipy (1.10) and update plotting test. Fix warning when plotting non-uniform axis (#3079)

  • Fix matplotlib 3.7 and scikit-learn 1.4 deprecations (#3102)

  • Add support for new pattern to generate random numbers introduced in dask 2023.2.1. Deprecate usage of numpy.random.RandomState in favour of numpy.random.default_rng(). Bump scipy minimum requirement to 1.4.0. (#3103)

  • Fix checking links in documentation for domain, which aren’t compatible with sphinx linkcheck (#3108)

Hyperspy 1.7.3 (2022-10-29)

Bug Fixes

  • Fix error when reading Velox containing FFT with odd number of pixels (#3040)

  • Fix pint Unit for pint>=0.20 (#3052)


  • Fix deprecated import of scipy ascent in docstrings and the test suite (#3032)

  • Fix error handling when trying to convert a ragged signal to non-ragged for numpy >=1.24 (#3033)

  • Fix getting random state dask for dask>=2022.10.0 (#3049)

Hyperspy 1.7.2 (2022-09-17)

Bug Fixes

  • Fix some errors and remove unnecessary code identified by LGTM. (#2977)

  • Fix error which occurs when guessing output size in the map() function and using dask newer than 2022.7.1 (#2981)

  • Fix display of x-ray lines when using log norm and the intensity at the line is 0 (#2995)

  • Fix handling constant derivative in spikes_removal_tool() (#3005)

  • Fix removing horizontal or vertical line widget; regression introduced in hyperspy 1.7.0 (#3008)

Improved Documentation

  • Add a note in the user guide to explain that when a file contains several datasets, load() returns a list of signals instead of a single signal and that list indexation can be used to access a single signal. (#2975)


  • Fix extension test suite CI workflow. Enable workflow manual trigger (#2982)

  • Fix deprecation warning and time zone test failing on windows (locale dependent) (#2984)

  • Fix external links in the documentation and add CI build to check external links (#3001)

  • Fix hyperlink in bibliography (#3015)

  • Fix matplotlib SpanSelector import for matplotlib 3.6 (#3016)

Hyperspy 1.7.1 (2022-06-18)

Bug Fixes

  • Fixes invalid file chunks when saving some signals to hspy/zspy formats. (#2940)

  • Fix issue where a TIFF image from an FEI FIB/SEM navigation camera image would not be read due to missing metadata (#2941)

  • Respect show_progressbar parameter in map() (#2946)

  • Fix regression in set_signal_range() which was raising an error when used interactively (#2948)

  • Fix SpanROI regression: the output of interactive() was not updated when the ROI was changed. Fix errors with updating limits when plotting empty slice of data. Improve docstrings and test coverage. (#2952)

  • Fix stacking signals that contain their variance in metadata. Previously it was raising an error when specifying the stacking axis. (#2954)

  • Fix missing API documentation of several signal classes. (#2957)

  • Fix two bugs in decomposition():

    • The poisson noise normalization was not applied when giving a signal_mask

    • An error was raised when applying a signal_mask on a signal with signal dimension larger than 1. (#2964)

Improved Documentation


  • Minor refactor of the EELS subshells in the elements dictionary. (#2868)

  • Fix packaging of test suite and tweak tests to pass on different platform of blas implementation (#2933)

Hyperspy 1.7.0 (2022-04-26)

New features

Bug Fixes

  • Signals with 1 value in the signal dimension will now be BaseSignal (#2773)

  • density_of_mixture() now throws a Value error when the density of an element is unknown (#2775)

  • Improve error message when performing Cliff-Lorimer quantification with a single line intensity (#2822)

  • Fix bug for the hydrogenic gdos k edge (#2859)

  • Fix bug in axes.UnitConversion: the offset value was initialized by units. (#2864)

  • Fix bug where the map() function wasn’t operating properly when an iterating signal was larger than the input signal. (#2878)

  • In case the Bruker defined XML element node at SpectrumRegion contains no information on the specific selected X-ray line (if there is only single line available), suppose it is ‘Ka’ line. (#2881)

  • When loading Bruker Bcf, cutoff_at_kV=None does no cutoff (#2898)

  • Fix bug where the map() function wasn’t operating properly when an iterating signal was not an array. (#2903)

  • Fix bug for not saving ragged arrays with dimensions larger than 2 in the ragged dimension. (#2906)

  • Fix bug with importing some spectra from eelsdb and add progress bar (#2916)

  • Fix bug when the spikes_removal_tool would not work interactively for signal with 0-dimension navigation space. (#2918)



  • Region of Interest (ROI) can now be created without specifying values (#2341)

  • mpfit cleanup (#2494)

  • Document reading Attolight data with the sur/pro format reader (#2559)

  • Lazy signals now caches the current data chunk when using multifit and when plotting, improving performance. (#2568)

  • Read cathodoluminescence metadata from digital micrograph files, amended in PR #2894 (#2590)

  • Add possibility to search/access nested items in DictionaryTreeBrowser (metadata) without providing full path to item. (#2633)

  • Improve map() function in BaseSignal by utilizing dask for both lazy and non-lazy signals. This includes adding a lazy_output parameter, meaning non-lazy signals now can output lazy results. See the user guide for more information. (#2703)

  • NeXus file with more options when reading and writing (#2725)

  • Add dtype argument to rebin() (#2764)

  • Add option to set output size when exporting images (#2791)

  • Add switch_iterpath() context manager to switch iterpath (#2795)

  • Add options not to close file (lazy signal only) and not to write dataset for hspy file format, see HSpy - HyperSpy’s HDF5 Specification for details (#2797)

  • Add Github workflow to run test suite of extension from a pull request. (#2824)

  • Add ragged attribute to BaseSignal to clarify when a signal contains a ragged array. Fix inconsistency caused by ragged array and add a ragged array section to the user guide (#2842)

  • Import hyperspy submodules lazily to speed up importing hyperspy. Fix autocompletion signals submodule (#2850)

  • Add support for JEOL SightX tiff file (#2862)

  • Add new markers arrow, ellipse and filled rectangle. (#2871)

  • Add metadata about the file-reading and saving operations to the Signals produced by load() and save() (see the metadata structure section of the user guide) (#2873)

  • expose Stage coordinates and rotation angle in metada for sem images in bcf reader. (#2911)

API changes

  • metadata.Signal.binned is replaced by an axis parameter, e. g. axes_manager[-1].is_binned (#2652)

    • when loading Bruker bcf, cutoff_at_kV=None (default) applies no more automatic cutoff.

    • New acceptable values "zealous" and "auto" do automatic cutoff. (#2910)

  • Deprecate the ability to directly set metadata and original_metadata Signal attributes in favor of using set_item() and add_dictionary() methods or specifying metadata when creating signals (#2913)


  • Fix warning when build doc and formatting user guide (#2762)

  • Drop support for python 3.6 (#2839)

  • Continuous integration fixes and improvements; Bump minimal version requirement of dask to 2.11.0 and matplotlib to 3.1.3 (#2866)

  • Tweak tests tolerance to fix tests failure on aarch64 platform; Add python 3.10 build. (#2914)

  • Add support for matplotlib 3.5, simplify maintenance of RangeWidget and some signal tools. (#2922)

  • Compress some tiff tests files to reduce package size (#2926)

v1.6.5 (2021-10-28)

Bug Fixes

  • Suspend plotting during smart_fit() call (#2796)

  • make add_marker() also check if the plot is not active before plotting signal (#2799)

  • Fix irresponsive ROI added to a signal plot with a right hand side axis (#2809)

  • Fix plot_histograms() drawstyle following matplotlib API change (#2810)

  • Fix incorrect map() output size of lazy signal when input and output axes do not match (#2837)

  • Add support for latest h5py release (3.5) (#2843)



  • add_widget() can now take a string or integer instead of tuple of string or integer (#2809)

v1.6.4 (2021-07-08)

Bug Fixes

  • Fix parsing EELS aperture label with unexpected value, for example ‘Imaging’ instead of ‘5 mm’ (#2772)

  • Lazy datasets can now be saved out as blockfiles (blo) (#2774)

  • ComplexSignals can now be rebinned without error (#2789)

  • Method estimate_parameters() of the Polynomial component now supports order greater than 10 (#2790)

  • Update minimal requirement of dependency importlib_metadata from >= 1.6.0 to >= 3.6 (#2793)


  • When saving a dataset with a dtype other than uint8 to a blockfile (blo) it is now possible to provide the argument intensity_scaling to map the intensity values to the reduced range (#2774)


  • Fix image comparison failure with numpy 1.21.0 (#2774)

Hyperspy 1.6.3 (2021-06-10)

Bug Fixes

  • Fix ROI snapping regression (#2720)

  • Fix shift1D(), align1D() and align_zero_loss_peak() regression with navigation dimension larger than one (#2729)

  • Fix disconnecting events when closing figure and remove_background() is active (#2734)

  • Fix map() regression of lazy signal with navigation chunks of size of 1 (#2748)

  • Fix unclear error message when reading a hspy file saved using blosc compression and hdf5plugin hasn’t been imported previously (#2760)

  • Fix saving navigator of lazy signal (#2763)


  • Use importlib_metadata instead of pkg_resources for extensions registration to speed up the import process and making it possible to install extensions and use them without restarting the python session (#2709)

  • Don’t import hyperspy extensions when registering extensions (#2711)

  • Improve docstrings of various fitting methods (#2724)

  • Improve speed of shift1D() (#2750)

  • Add support for recent EMPAD file; scanning size wasn’t parsed. (#2757)


  • Add drone CI to test arm64 platform (#2713)

  • Fix latex doc build on github actions (#2714)

  • Use towncrier to generate changelog automatically (#2717)

  • Fix test suite to support dask 2021.4.1 (#2722)

  • Generate changelog when building doc to keep the changelog of the development doc up to date on (#2758)

  • Use mamba and conda-forge channel on azure pipeline (#2759)


This is a maintenance release that adds support for python 3.9 and includes numerous bug fixes and enhancements. See the issue tracker for details.

Bug Fixes

  • Fix disconnect event when closing navigator only plot (fixes #996), (#2631)

  • Fix incorrect chunksize when saving EMD NCEM file and specifying chunks (#2629)

  • Fix find_peaks() GUIs call with laplacian/difference of gaussian methods (#2622 and #2647)

  • Fix various bugs with CircleWidget and Line2DWidget (#2625)

  • Fix setting signal range of model with negative axis scales (#2656)

  • Fix and improve mask handling in lazy decomposition; Close #2605 (#2657)

  • Plot scalebar when the axis scales have different sign, fixes #2557 (#2657)

  • Fix align1D() returning zeros shifts (#2675)

  • Fix finding dataset path for EMD NCEM file containing more than one dataset in a group (#2673)

  • Fix squeeze function for multiple zero-dimensional entries, improved docstring, added to user guide. (#2676)

  • Fix error in Cliff-Lorimer quantification using absorption correction (#2681)

  • Fix navigation_mask bug in decomposition when provided as numpy array (#2679)

  • Fix closing image contrast tool and setting vmin/vmax values (#2684)

  • Fix range widget with matplotlib 3.4 (#2684)

  • Fix bug in interactive() with function returning None. Improve user guide example. (#2686)

  • Fix broken events when changing signal type #2683

  • Fix setting offset in rebin: the offset was changed in the wrong axis (#2690)

  • Fix reading XRF bruker file, close #2689 (#2694)


  • Widgets plotting improvement and add pick_tolerance to plot preferences (#2615)

  • Pass keyword argument to the image IO plugins (#2627)

  • Improve error message when file not found (#2597)

  • Add update instructions to user guide (#2621)

  • Improve plotting navigator of lazy signals, add navigator setter to lazy signals (#2631)

  • Use 'dask_auto' when rechunk=True in change_dtype() for lazy signal (#2645)

  • Use dask chunking when saving lazy signal instead of rechunking and leave the user to decide what is the suitable chunking (#2629)

  • Added lazy reading support for FFT and DPC datasets in FEI emd datasets (#2651).

  • Improve error message when initialising SpanROI with left >= right (#2604)

  • Allow running the test suite without the pytest-mpl plugin (#2624)

  • Add Releasing guide (#2595)

  • Add support for python 3.9, fix deprecation warning with matplotlib 3.4.0 and bump minimum requirement to numpy 1.17.1 and dask 2.1.0. (#2663)

  • Use native endianess in numba jitted functions. (#2678)

  • Add option not to snap ROI when calling the interactive() method of a ROI (#2686)

  • Make DictionaryTreeBrowser lazy by default - see #368 (#2623)

  • Speed up setting CI on azure pipeline (#2694)

  • Improve performance issue with the map method of lazy signal (#2617)

  • Add option to copy/load original metadata in hs.stack and hs.load to avoid large original_metadata which can slowdown processing. Close #1398, #2045, #2536 and #1568. (#2691)


  • Fix warnings when building documentation (#2596)

  • Drop support for numpy<1.16, in line with NEP 29 and fix protochip reader for numpy 1.20 (#2616)

  • Run test suite against upstream dependencies (numpy, scipy, scikit-learn and scikit-image) (#2616)

  • Update external links in the loading data section of the user guide (#2627)

  • Fix various future and deprecation warnings from numpy and scikit-learn (#2646)

  • Fix iterpath VisibleDeprecationWarning when using fit_component() (#2654)

  • Add integration test suite documentation in the developer guide. (#2663)

  • Fix SkewNormal component compatibility with sympy 1.8 (#2701)


This is a maintenance release that adds compatibility with h5py 3.0 and includes numerous bug fixes and enhancements. See the issue tracker for details.




  • The get_histogram() now uses numpy’s np.histogram_bin_edges() and supports all of its bins keyword values.

  • Further improvements to the contrast adjustment tool. Test it by pressing the h key on any image.

  • The following components have been rewritten using Expression, boosting their speeds among other benefits.

  • The model fitting fit() and multifit() methods have been vastly improved. See Fitting the model to the data and the API changes section below.

  • New serpentine iteration path for multi-dimensional fitting. See Fitting multidimensional datasets.

  • The plot_spectra() function now listens to events to update the figure automatically. See this example.

  • Improve thread-based parallelism. Add max_workers argument to the map() method, such that the user can directly control how many threads they launch.

  • Many improvements to the decomposition() and blind_source_separation() methods, including support for scikit-learn like algorithms, better API and much improved documentation. See Machine learning and the API changes section below.

  • Add option to calculate the absolute thickness to the EELS estimate_thickness() method. See Thickness estimation.

  • Vastly improved performance and memory footprint of the estimate_shift2D() method.

  • The remove_background() method can now remove Doniach, exponential, Lorentzian, skew normal, split Voigt and Voigt functions. Furthermore, it can return the background model that includes an estimation of the reduced chi-squared.

  • The performance of the maximum-likelihood PCA method was greatly improved.

  • All ROIs now have a __getitem__ method, enabling e.g. using them with the unpack * operator. See Slicing using ROIs for an example.

  • New syntax to set the contrast when plotting images. In particular, the vmin and vmax keywords now take values like vmin="30th" to clip the minimum value to the 30th percentile. See Fast Fourier Transform (FFT) for an example.

  • The plot() and plot() methods take a new keyword argument autoscale. See Customising image plot for details.

  • The contrast editor and the decomposition methods can now operate on complex signals.

  • The default colormap can now be set in preferences.

API changes

  • The plot() keyword argument saturated_pixels is deprecated. Please use vmin and/or vmax instead.

  • The load() keyword argument dataset_name has been renamed to dataset_path.

  • The set_signal_type() method no longer takes None. Use the empty string "" instead.

  • The get_histogram() bins keyword values have been renamed as follows for consistency with numpy:

    • "scotts" -> "scott",

    • "freedman" -> "fd"

  • Multiple changes to the syntax of the fit() and multifit() methods:

  • The fitter keyword has been renamed to optimizer.

  • The values that the optimizer keyword take have been renamed for consistency with scipy:

    • "fmin" -> "Nelder-Mead",

    • "fmin_cg" -> "CG",

    • "fmin_ncg" -> "Newton-CG",

    • "fmin_bfgs" -> "BFGS",

    • "fmin_l_bfgs_b" -> "L-BFGS-B",

    • "fmin_tnc" -> "TNC",

    • "fmin_powell" -> "Powell",

    • "mpfit" -> "lm" (in combination with "bounded=True"),

    • "leastsq" -> "lm",

  • Passing integer arguments to parallel to select the number of workers is now deprecated. Use parallel=True, max_workers={value} instead.

  • The method keyword has been renamed to loss_function.

  • The loss_function value "ml" has been renamed to "ML-poisson".

  • The grad keyword no longer takes boolean values. It takes the following values instead: "fd", "analytical", callable or None.

  • The ext_bounding keyword has been deprecated and will be removed. Use bounded=True instead.

  • The min_function keyword argument has been deprecated and will be removed. Use loss_function instead.,

  • The min_function_grad keyword arguments has been deprecated and will be removed. Use grad instead.

  • The iterpath default will change from 'flyback' to 'serpentine' in HyperSpy version 2.0.

  • The following BaseModel methods are now private:

    • set_boundaries()

    • set_mpfit_parameters_info()

    • set_boundaries()

  • The comp_label keyword of the machine learning plotting functions has been renamed to title.

  • The orpca constructor’s learning_rate keyword has been renamed to subspace_learning_rate

  • The orpca constructor’s momentum keyword has been renamed to subspace_momentum

  • The svd_pca constructor’s centre keyword values have been renamed as follows:

    • "trials" -> "navigation"

    • "variables" -> "signal"

  • The bounds keyword argument of the decomposition() is deprecated and will be removed.

  • Several syntax changes in the decomposition() method:

    • Several algorithm keyword values have been renamed as follows:

      • "svd": "SVD",

      • "fast_svd": "SVD",

      • "nmf": "NMF",

      • "fast_mlpca": "MLPCA",

      • "mlpca": "MLPCA",

      • "RPCA_GoDec": "RPCA",

    • The polyfit argument has been deprecated and will be removed. Use var_func instead.


This is a maintenance release that adds compatibility with Numpy 1.17 and Dask 2.3.0 and fixes a bug in the Bruker reader. See the issue tracker for details.


This is a maintenance release that fixes some regressions introduced in v1.5. Follow the following links for details on all the bugs fixed.




For developers


This is a maintenance release. Among many other fixes and enhancements, this release fixes compatibility issues with Matplotlib v 3.1. Follow the following links for details on all the bugs fixed and enhancements.


This is a maintenance release. Follow the following links for details on all the bugs fixed and enhancements.

This release fixes compatibility issues with Python 3.7.


This is a minor release. Follow the following links for details on all the bugs fixed, enhancements and new features.



  • plot_images() improvements (see Plotting several images for details):

    • The cmap option of plot_images() supports iterable types, allowing the user to specify different colormaps for the different images that are plotted by providing a list or other generator.

    • Clicking on an individual image updates it.

  • New customizable keyboard shortcuts to navigate multi-dimensional datasets. See Data visualization.

  • The remove_background() method now operates much faster in multi-dimensional datasets and adds the options to interatively plot the remainder of the operation and to set the removed background to zero. See Background removal for details.

  • The plot() method now takes a norm keyword that can be “linear”, “log”, “auto” or a matplotlib norm. See Customising image plot for details. Moreover, there are three new extra keyword arguments, fft_shift and power_spectrum, that are useful when plotting fourier transforms. See Fast Fourier Transform (FFT).

  • The align2D() and estimate_shift2D() can operate with sub-pixel accuracy using skimage’s upsampled matrix-multiplication DFT. See Signal registration and alignment.


This is a maintenance release. Follow the following links for details on all the bugs fixed and enhancements.


This is a maintenance release. Follow the following links for details on all the bugs fixed and enhancements.

Starting with this version, the HyperSpy WinPython Bundle distribution is no longer released in sync with HyperSpy. For HyperSpy WinPython Bundle releases see


This is a minor release. Follow the following links for details on all the bugs fixed, feature and documentation enhancements, and new features.


  • rebin() supports upscaling and rebinning to arbitrary sizes through linear interpolation. See Rebinning. It also runs faster if numba is installed.

  • signal_extent and navigation_extent properties to easily get the extent of each space.

  • New IPywidgets Graphical User Interface (GUI) elements for the Jupyter Notebook. See the new hyperspy_gui_ipywidgets package. It is not installed by default, see Installing HyperSpy for details.

  • All the Region Of Interest (ROI) now have a gui() method to display a GUI if at least one of HyperSpy’s GUI packgages are installed.


  • Creating many markers is now much faster.

  • New “Stage” metadata node. See Metadata structure for details.

  • The Brucker file reader now supports the new version of the format. See Bruker composite file.

  • HyperSpy is now compatible with all matplotlib backends, including the nbagg which is particularly convenient for interactive data analysis in the Jupyter Notebook in combination with the new hyperspy_gui_ipywidgets package. See Starting Python in Windows.

  • The vmin and vmax arguments of the plot_images() function now accept lists to enable setting these parameters for each plot individually.

  • The plot_decomposition_results() and plot_bss_results() methods now makes a better guess of the number of navigators (if any) required to visualise the components. (Previously they were always plotting four figures by default.)

  • All functions that take a signal range can now take a SpanROI.

  • The following ROIs can now be used for indexing or slicing (see here for details):

API changes

  • Permanent markers (if any) are now displayed when plotting by default.

  • HyperSpy no longer depends on traitsui (fixing many installation issues) and ipywidgets as the GUI elements based on these packages have now been splitted into separate packages and are not installed by default.

  • The following methods now raise a ValueError when not providing the number of components if output_dimension was not specified when performing a decomposition. (Previously they would plot as many figures as available components, usually resulting in memory saturation):

    • plot_decomposition_results().

    • plot_decomposition_factors().

  • The default extension when saving to HDF5 following HyperSpy’s specification is now hspy instead of hdf5. See HSpy - HyperSpy’s HDF5 Specification.

  • The following methods are deprecated and will be removed in HyperSpy 2.0

    • show(). Use gui() instead.

    • All notebook_interaction() method. Use the equivalent gui() method instead.

    • integrate_in_range(). Use integrate1D() instead.

  • The following items have been removed from preferences:

    • General.default_export_format

    • General.lazy

    • Model.default_fitter

    • Machine_learning.multiple_files

    • Machine_learning.same_window

    • Plot.default_style_to_compare_spectra

    • Plot.plot_on_load

    • Plot.pylab_inline

    • EELS.fine_structure_width

    • EELS.fine_structure_active

    • EELS.fine_structure_smoothing

    • EELS.synchronize_cl_with_ll

    • EELS.preedge_safe_window_width

    • EELS.min_distance_between_edges_for_fine_structure

  • New Preferences.GUIs section to enable/disable the installed GUI toolkits.

For developers

  • In addition to adding ipywidgets GUI elements, the traitsui GUI elements have been splitted into a separate package. See the new hyperspy_gui_traitsui package.

  • The new ui_registry enables easy connection of external GUI elements to HyperSpy. This is the mechanism used to split the traitsui and ipywidgets GUI elements.


This is a minor release. Follow the following links for details on all the bugs fixed, enhancements and new features.



For developers


This is a maintenance release. Follow the following links for details on all the bugs fixed and enhancements.


This is a maintenance release. Follow the following link for details on all the bugs fixed.


  • Prettier X-ray lines labels.

  • New metadata added to the HyperSpy metadata specifications: magnification, frame_number, camera_length, authors, doi, notes and quantity. See Metadata structure for details.

  • The y-axis label (for 1D signals) and colorbar label (for 2D signals) are now taken from the new metadata.Signal.quantity.

  • The time and date metadata are now stored in the ISO 8601 format.

  • All metadata in the HyperSpy metadata specification is now read from all supported file formats when available.


This is a minor release. Follow the following links for details on all the bugs fixed.



  • fit() takes a new algorithm, the global optimizer differential evolution.

  • fit() algorithm, leastsq, inherits SciPy’s bound constraints support (requires SciPy >= 0.17).

  • fit() algorithm names changed to be consistent scipy.optimze.minimize() notation.


This is a maintenance release. Follow the following links for details on all the bugs fixed.


This is a major release. Here we only list the highlist. A detailed list of changes is available in github.







API changes

  • Split components into components1D and components2D.

  • Remove record_by from metadata.

  • Remove simulation classes.

  • The Signal1D, Signal2D and BaseSignal classes deprecated the old Spectrum Image and Signal classes.


This is a maintenance release. Follow the following links for details on all the bugs fixed, feature and documentation enhancements.

It also includes a new feature and introduces an important API change that will be fully enforced in Hyperspy 1.0.

New feature

  • Widgets to interact with the model components in the Jupyter Notebook. See here and #1007 .

API changes

The new BaseSignal, Signal1D and Signal2D deprecate Signal, Signal1D and Signal2D respectively. Also as_signal1D, as_signal2D, to_signal1D and to_signal2D deprecate as_signal1D, as_signal2D, to_spectrum and to_image. See #963 and #943 for details.


This release adds support for Python 3 and drops support for Python 2. In all other respects it is identical to v0.8.3.


This is a maintenance release that includes fixes for multiple bugs, some enhancements, new features and API changes. This is set to be the last HyperSpy release for Python 2. The release (HyperSpy 0.8.4) will support only Python 3.

Importantly, the way to start HyperSpy changes (again) in this release. Please read carefully Starting Python in Windows for details.

The broadcasting rules have also changed. See Signal operations for details.

Follow the following links for details on all the bugs fixed, documentation enhancements, enhancements, new features and API changes


This is a maintenance release that fixes an issue with the Python installers. Those who have successfully installed v0.8.1 do not need to upgrade.


This is a maintenance release. Follow the following links for details on all the bugs fixed, feature and documentation enhancements.

Importantly, the way to start HyperSpy changes in this release. Read Starting Python in Windows for details.

It also includes some new features and introduces important API changes that will be fully enforced in Hyperspy 1.0.

New features

  • Support for IPython 3.0.

  • %hyperspy IPython magic to easily and transparently import HyperSpy, matplotlib and numpy when using IPython.

  • Expression model component to easily create analytical function components. More details here.

  • unfolded() context manager.

  • derivative() method.

  • syntax to access the components in the model that includes pretty printing of the components.

API changes

  • hspy is now deprecated in favour of the new api. The new API renames and/or move several modules as folows:

    • hspy.components -> api.model.components

    • hspy.utils-> api

    • hspy.utils.markers api.plot.markers

    • hspy.utils.example_signals -> api.datasets.example_signals

    In HyperSpy 0.8.1 the full content of hspy is still imported in the user namespace, but this can now be disabled in hs.preferences.General.import_hspy. In Hyperspy 1.0 it will be disabled by default and the hspy module will be fully removed in HyperSpy 0.10. We encourage all users to migrate to the new syntax. For more details see Starting Python in Windows.

  • Indexing the Signal class is now deprecated. We encourage all users to use isig and inav instead for indexing.

  • create_model() is now deprecated in favour of the new equivalent create_model() Signal method.

  • unfold_if_multidim() is deprecated.


New features


  • spikes_removal_tool() displays derivative max value when used with GUI.

  • Progress-bar can now be suppressed by passing show_progressbar argument to all functions that generate it.


  • HDF5 file format now supports saving lists, tuples, binary strings and signals in metadata.



API changes

  • Component and Parameter now inherit traits.api.HasTraits that enable traitsui to modify these objects.

  • attrsetter() is added, behaving as the default python setattr() with nested attributes.

  • Several widget functions were made internal and/or renamed:
    • add_patch_to -> _add_patch_to

    • set_patch -> _set_patch

    • onmove -> _onmousemove

    • update_patch_position -> _update_patch_position

    • update_patch_size -> _update_patch_size

    • add_axes -> set_mpl_ax


This is a maintenance release. A list of fixed issues is available in the 0.7.3 milestone in the github repository.


This is a maintenance release. A list of fixed issues is available in the 0.7.2 milestone in the github repository.


This is a maintenance release. A list of fixed issues is available in the 0.7.1 milestone in the github repository.

New features


New features


  • New syntax to index the AxesManager.

  • New Signal methods to transform between Signal subclasses. More information here.

    • set_signal_type()

    • set_signal_origin()

    • as_signal2D()

    • as_signal1D()

  • The string representation of the Signal class now prints the shape of the data and includes a separator between the navigation and the signal axes e.g (100, 10| 5) for a signal with two navigation axes of size 100 and 10 and one signal axis of size 5.

  • Add support for RGBA data. See Changing the data type.

  • The default toolkit can now be saved in the preferences.

  • Added full compatibility with the Qt toolkit that is now the default.

  • Added compatibility witn the the GTK and TK toolkits, although with no GUI features.

  • It is now possible to run HyperSpy in a headless system.

  • Added a CLI to remove_background().

  • New estimate_peak_width() method to estimate peak width.

  • New methods to integrate over one axis: integrate1D() and integrate_in_range().

  • New metadata attribute, Signal.binned. Several methods behave differently on binned and unbinned signals. See Binned and unbinned signals.

  • New map() method to easily transform the data using a function that operates on individual signals. See Iterating over the navigation axes.

  • New get_histogram() and print_summary_statistics() methods.

  • The spikes removal tool has been moved to the Signal1D class so that it is available for all its subclasses.

  • The split() method now can automatically split back stacked signals into its original part. See Splitting and stacking.


  • Improved support for FEI’s emi and ser files.

  • Improved support for Gatan’s dm3 files.

  • Add support for reading Gatan’s dm4 files.


Curve fitting

Machine learning

  • The PCA scree plot can now be easily obtained as a Signal. See Scree plots.

  • The decomposition and blind source separation components can now be obtained as Signal instances. See Clustering plots.

  • New methods to plot the decomposition and blind source separation results that support n-dimensional loadings. See Visualizing results.

Dielectric function



API changes

  • hyperspy.misc has been reorganized. Most of the functions in misc.utils has been rellocated to specialized modules. misc.utils is no longer imported in hyperspy.hspy. A new hyperspy.utils module is imported instead.

  • Objects that have been renamed

    • hspy.elements -> utils.material.elements.

    • Signal.navigation_indexer -> inav.

    • Signal.signal_indexer -> isig.

    • Signal.mapped_parameters -> Signal.metadata.

    • Signal.original_parameters -> Signal.original_metadata.

  • The metadata has been reorganized. See Metadata structure.

  • The following signal methods now operate out-of-place:

    • swap_axes()

    • rebin()


New features

  • Signal now supports indexing and slicing. See Indexing.

  • Most arithmetic and rich arithmetic operators work with signal. See Signal operations.

  • Much improved EELSSpectrum methods: estimate_zero_loss_peak_centre(), estimate_elastic_scattering_intensity() and estimate_elastic_scattering_threshold().

  • The axes can now be given using their name e.g. s.crop("x", 1,10)

  • New syntax to specify position over axes: an integer specifies the indexes over the axis and a floating number specifies the position in the axis units e.g. s.crop("x", 1, 10.) crops over the axis x (in meters) from index 1 to value 10 meters. Note that this may make your old scripts behave in unexpected ways as just renaming the old *_in_units and *_in_values methods won’t work in most cases.

  • Most methods now use the natural order i.e. X,Y,Z.. to index the axes.

  • Add padding to fourier-log and fourier-ratio deconvolution to fix the wrap-around problem and increase its performance.

  • New get_fine_structure_as_spectrum() EELSCLEdge method.

  • New Arctan model component.

  • New enable_adjust_position() and disable_adjust_position() to easily change the position of components using the mouse on the plot.

  • New Model methods set_parameters_value(), set_parameters_free() and set_parameters_not_free() to easily set several important component attributes of a list of components at once.

  • New stack() function to stack signals.

  • New Signal methods: integrate_simpson(), max(), min(), var(), and std().

  • New sliders window to easily navigate signals with navigation_dimension > 2.

  • The Ripple (rpl) reader can now read rpl files produced by INCA.

API changes

  • The following functions has been renamed or removed:

    • components.EELSCLEdge

      • knots_factor -> fine_structure_smoothing

      • edge_position -> onset_energy

      • energy_shift removed

    • components.Voigt.origin -> centre

    • signals.Signal1D

      • find_peaks_1D -> Signal.find_peaks1D_ohaver

      • align_1D -> Signal.align1D

      • shift_1D -> Signal.shift1D

      • interpolate_1D -> Signal.interpolate1D

    • signals.Signal2D.estimate_2D_translation -> Signal.estimate_shift2D

    • Signal

      • split_in -> split

      • crop_in_units -> crop

      • crop_in_pixels -> crop

  • Change syntax to create Signal objects. Instead of a dictionary Signal.__init__ takes keywords e.g with a new syntax . >>> s = signals.Signal1D(np.arange(10)) instead of >>> s = signals.Signal1D({'data' : np.arange(10)})


New features

  • New Signal method get_current_signal proposed by magnunor.

  • New Signal save method keyword extension to easily change the saving format while keeping the same file name.

  • New EELSSpectrum methods: estimate_elastic_scattering_intensity, fourier_ratio_deconvolution, richardson_lucy_deconvolution, power_law_extrapolation.

  • New Signal1D method: hanning_taper.

Major bugs fixed

  • The print_current_values Model method was raising errors when fine structure was enabled or when only_free = False.

  • The load function signal_type keyword was not passed to the readers.

  • The spikes removal tool was unable to find the next spikes when the spike was detected close to the limits of the spectrum.

  • load was raising an UnicodeError when the title contained non-ASCII characters.

  • In Windows HyperSpy Here was opening in the current folder, not in the selected folder.

  • The fine structure coefficients were overwritten with their std when charging values from the model.

  • Storing the parameters in the maps and all the related functionality was broken for 1D spectrum.

  • Remove_background was broken for 1D spectrum.

API changes

  • EELSSpectrum.find_low_loss_centre was renamed to estimate_zero_loss_peak_centre.

  • EELSSpectrum.calculate_FWHM was renamed to estimate_FWHM.


New features

  • The documentation was thoroughly revised, courtesy of M. Walls.

  • New user interface to remove spikes from EELS spectra.

  • New align2D signals.Signal2D method to align image stacks.

  • When loading image files, the data are now automatically converted to grayscale when all the color channels are equal.

  • Add the possibility to load a stack memory mapped (similar to ImageJ virtual stack).

  • Improved hyperspy starter script that now includes the possibility to start HyperSpy in the new IPython notebook.

  • Add “HyperSpy notebook here” to the Windows context menu.

  • The information displayed in the plots produced by Signal.plot have been enhanced.

  • Added Egerton’s sigmak3 and sigmal3 GOS calculations (translated from matlab by I. Iyengar) to the EELS core loss component.

  • A browsable dictionary containing the chemical elements and their onset energies is now available in the user namespace under the variable name elements.

  • The ripple file format now supports storing the beam energy, the collection and the convergence angle.

Major bugs fixed

  • The EELS core loss component had a bug in the calculation of the relativistic gamma that produced a gamma that was always approximately zero. As a consequence the GOS calculation was wrong, especially for high beam energies.

  • Loading msa files was broken when running on Python 2.7.2 and newer.

  • Saving images to rpl format was broken.

  • Performing BSS on data decomposed with poissonian noise normalization was failing when some columns or rows of the unfolded data were zero, what occurs often in EDX data for example.

  • Importing some versions of scikits learn was broken

  • The progress bar was not working properly in the new IPython notebook.

  • The constrast of the image was not automatically updated.

API changes

  • spatial_mask was renamed to navigation_mask.

  • Signal1D and Signal2D are not loaded into the user namespace by default. The signals module is loaded instead.

  • Change the default BSS algorithm to sklearn fastica, that is now distributed with HyperSpy and used in case that sklearn is not installed e.g. when using EPDFree.

  • _slicing_axes was renamed to signal_axes.

  • _non_slicing_axes to navigation_axes.

  • All the Model *_in_pixels methods were renamed to to _*_in_pixel.

  • EELSCLEdge.fs_state was renamed to fine_structure_active.

  • EELSCLEdge.fslist was renamed to fine_structure_coeff.

  • EELSCLEdge.fs_emax was renamed to fine_structure_width.

  • EELSCLEdge.freedelta was renamed to free_energy_shift.

  • was renamed to energy_shift.

  • A value of True in a mask now means that the item is masked all over HyperSpy.


New features

  • Added TIFF 16, 32 and 64 bits support by using (and distributing) Christoph Gohlke’s tifffile library.

  • Improved UTF8 support.

  • Reduce the number of required libraries by making mdp and hdf5 not mandatory.

  • Improve the information returned by __repr__ of several objects.

  • DictionaryBrowser now has an export method, i.e. mapped parameters and original_parameters can be exported.

  • New _id_name attribute for Components and Parameters. Improvements in their __repr__ methods.

  • can now be overwriten by the user.

  • New Signal.__str__ method.

  • Include HyperSpy in The Python Package Index.

Bugs fixed

  • Non-ascii characters breaking IO and print features fixed.

  • Loading of multiple files at once using wildcards fixed.

  • Remove broken hyperspy-gui script.

  • Remove unmantained and broken 2D peak finding and analysis features.

Syntax changes

  • In EELS automatic background feature creates a PowerLaw component, adds it to the model an add it to a variable in the user namespace. The variable has been renamed from bg to background.

  • pes_gaussian Component renamed to pes_core_line_shape.


New features

  • Add a slider to the filter ui.

  • Add auto_replot to sum.

  • Add butterworth filter.

  • Added centring and auto_transpose to the svd_pca algorithm.

  • Keep the mva_results information when changing the signal type.

  • Added sparse_pca and mini_batch_sparse_pca to decomposition algorithms.

  • Added TV to the smoothing algorithms available in BSS.

  • Added whitening to the mdp ICA preprocessing.

  • Add explained_variance_ratio.

  • Improvements in saving/loading mva data.

  • Add option to perform ICA on the scores.

  • Add orthomax FA algorithm.

  • Add plot methods to Component and Parameter.

  • Add plot_results to Model.

  • Add possibility to export the decomposition and bss results to a folder.

  • Add Signal method change_dtype.

  • Add the possibility to pass extra parameters to the ICA algorithm.

  • Add the possibility to reproject the data after a decomposition.

  • Add warning when decomposing a non-float signal.

  • adds a method to get the PCs as a Signal1D object and adds smoothing to the ICA preprocessing.

  • Add the possibility to select the energy range in which to perform spike removal operations.

  • the smoothings guis now offer differentiation and line color option. Smoothing now does not require a gui.

  • Fix reverse_ic which was not reversing the scores and improve the autoreversing method.

  • Avoid cropping when is not needed.

  • Changed criteria to reverse the ICs.

  • Changed nonans default to False for plotting.

  • Change the whitening algorithm to a svd based one and add sklearn fastica algorithm.

  • Clean the ummixing info after a new decomposition.

  • Increase the chances that similar independent components will have the same indexes.

  • Make savitzky-golay smoothing work without raising figures.

  • Make plot_decomposition* plot only the number of factors/scores determined by output_dimension.

  • make the Parameter __repr__ method print its name.

  • New contrast adjustment tool.

  • New export method for Model, Component and Parameter.

  • New Model method: print_current_values.

  • New signal, spectrum_simulation.

  • New smoothing algorithm: total variance denoising.

  • Plotting the components in the same or separate windows is now configurable in the preferences.

  • Plotting the spikes is now optional.

  • Return an error message when the decomposition algorithm is not recognised.

  • Store the masks in mva_results.

  • The free parameters are now automically updated on chaning the free attribute.

Bugs fixed

  • Added missing keywords to plot_pca_factors and plot_ica_factors.

  • renamed incorrectly named exportPca and exportIca functions.

  • an error was raised when calling generate_data_from_model.

  • a signal with containing nans was failing to plot.

  • attempting to use any decomposition plotting method after loading with mva_results.load was raising an error.

  • a typo was causing in error in pca when normalize_variance = True.

  • a typo was raising an error when cropping the decomposition dimension.

  • commit 5ff3798105d6 made decomposition and other methods raise an error.

  • BUG-FIXED: the decomposition centering index was wrong.

  • ensure_directory was failing for the current directory.

  • model data forced to be 3D unnecessarily.

  • non declared variable was raising an error.

  • plot naming for peak char factor plots were messed up.

  • plot_RGB was broken.

  • plot_scores_2D was using the transpose of the shape to reshape the scores.

  • remove background was raising an error when the navigation dimension was 0.

  • saving the scores was sometimes transposing the shape.

  • selecting indexes while using the learning export functions was raising an error.

  • the calibrate ui was calculating wrongly the calibration the first time that Apply was pressed.

  • the offset estimation was summing instead of averaging.

  • the plot_explained_variance_ratio was actually plotting the cumulative, renamed.

  • the signal mask in decomposition and ica was not being raveled.

  • the slice attribute was not correctly set at init in some scenarios.

  • the smoothing and calibrabrion UIs were freezing when the plots where closed before closing the UI window.

  • to_spectrum was transposing the navigation dimension.

  • variance2one was operating in the wrong axis.

  • when closing the plots of a model, the UI object was not being destroyed.

  • when plotting an image the title was not displayed.

  • when the axis size was changed (e.g. after cropping) the set_signal_dimension method was not being called.

  • when using transform the data was being centered and the resulting scores were wrong.

Syntax changes

  • in decomposition V rename to explained_variance.

  • In FixedPattern, default interpolation changed to linear.

  • Line and parabole components deleted + improvements in the docstrings.

  • pca_V = variance.

  • mva_result renamed to learning_results.

  • pca renamed to decomposition.

  • pca_v and mva_results.v renamed to scores pc renamed to factors . pca_build_SI renamed to get_pca_model ica_build_SI renamed to get_ica_model.

  • plot_explained_variance renamed to plot_explained_variance_ratio.

  • principal_components_analysis renamed to decomposition.

  • rename eels_simulation to eels_spectrum_simulation.

  • Rename the output parameter of svd_pca and add scores.

  • Replace plot_lev by plot_explained_variance_ratio.

  • Scores renamed to loadings.

  • slice_bool renamed to navigate to make its function more explicit.

  • smoothing renamed to pretreatment and butter added.

  • variance2one renamed to normalize_variance.

  • w renamed to unmixing matrix and fixes a bug when loading a mva_result in which output_dimension = None.

  • ubshells are again availabe in the interactive session.

  • Several changes to the interface.

  • The documentation was updated to reflex the last changes.

  • The microscopes.csv file was updated so it no longer contains the Orsay VG parameters.