Skip to content

Latest commit

 

History

History
303 lines (209 loc) · 9.54 KB

CHANGELOG.rst

File metadata and controls

303 lines (209 loc) · 9.54 KB

Changes

Version 3.0.5

2024.03.14 Remove usage of deprecated and removed np.complex.

Version 3.0.4

2023.08.08 Build on numpy 1.24 for compatability with other pennmem packages

Version 3.0.3

2023.08.03 Update SWIG version to support python 3.11

Version 3.0.2

2022.11.18 Added support for hdf5 > 2.9.0

Version 3.0.1

2021.11.08 Bug fixes, automated conda deployment

Version 3.0.0

2021.11.08 Version 3.0 of PTSA is largely a cleanup release. Several deprecated features have been removed and things have been reorganized.

What's New

  • Filter objects no longer accept a TimeSeries object as a positional argument. Rather, these are passed to the object's .filter method instead. This allows for reusing filters on multiple time series as well as adding filter pipeline capabilities.
  • Added a class decorator to TimeSeries that coerces the return type of xarray methods to be TimeSeries rather than xarray.DataArray.
  • to_hdf and from_hdf now use xarray's recommended file saving method, netCDF. We use xarray built-ins to do I/O rather than in-house code via h5py.
  • to_hdf and from_hdf now support pandas MultiIndexes as coordinates

Removals

  • TimeSeriesX backwards compatibility has been removed. Use TimeSeries instead.
  • the .filtered method has been removed. Initialize a ButterworthFilter and use .filter_with() instead

Version 2.0.10

2020-08-21

  • Drop extra conda dependencies.

Version 2.0.9

2020-05-07

  • Updates to conform with new xarray API.

Version 2.0.8

2019-10-17

  • Resolved several warnings and a test failure in EEGReader.

Version 2.0.7

2019-09-09

  • Identical to 2.0.6 but fixes a file upload error for the Python 3.7 version.

Version 2.0.6

2019-08-15

  • MorletWaveletFilter now supports a complete parameter, defaulting to True, which uses a complete Morlet wavelet with a zero mean, and with its frequency corrected to be equivalent to a standard Morlet. This significantly improves power and phase accuracy for small wavelet widths.

Version 2.0.4

2018-08-31

  • TimeSeries.to_hdf once again uses a human-readable output format

Version 2.0.3

2018-08-17

  • TimeSeries.to_hdf and from_hdf have been reverted to the previous behavior of base 64-encoding Numpy recarrays before writing to disk. This makes the stored files less human-readable, but offers the advantage of being much more robust.

Version 2.0.2

2018-08-08

New features:

  • TimeSeries now has a filter_with method to make applying filters slightly easier (#211)

Fixes:

  • Resource warnings were resolved by ensuring files get closed (#212)
  • ButterworthFilter's freq_range argument is now coerced to a list (#214)

Other:

  • Filter modules have been renamed to conform with PEP8 (#213). This should not affect normal usage since all filters are imported to the top-level filter namespace by default.

Version 2.0.1

2018-07-18

This is a minor bugfix release.

Issues addressed:

  • Require a newer version of xarray and fix HDF5 serialization (#204)
  • Optionally coerce datatypes to double for filters (#205)

Version 2.0.0

2018-07-13

Version 2.0 of PTSA is a major update which includes several breaking changes including the removal of several packages and modules, name changes, and deprecations.

PTSA 2.0 is the last major version to support loading of lab-specific data. In future releases, all of this functionality will be removed in favor of using the cmlreaders package which includes the ability to load timeseries data and transform into a PTSA-compatible format.

Removals

The following packages and modules have been removed:

  • ptsa.plotting - see https://github.com/pennmem/ptsa_plot instead (#147)
  • ptsa.stats - most functionality now exists in SciPy (#147)
  • ptsa.emd, ptsa.iwasobi, ptsa.ica and ptsa.wica (#147)
  • ptsa.filt - filtering is now contained primarily in ptsa.data.filters (#158)
  • Old time series class (fb08e6c2)
  • All modules within ptsa.data that worked with the deprecated time series implementation

Renames

  • TimeSeriesX has been renamed to TimeSeries and is now found in the ptsa.data.timeseries module (#141). TimeSeriesX still exists as an alias but will give a DeprecationWarning.
  • Most modules have been renamed to conform to PEP8 module naming conventions. These renames are generally not noticed by users if loading readers from the ptsa.data.readers namespace.

Other changes

  • Reformatted output of TalReader.read

    • The atlases field is a nested structured array, rather than an array of dictionaries
    • Most idioms for getting coordinates and atlas labels for an electrode should continue to work in the new format
    • The channels field has been changed from an object array of lists to an array of two integers

  • Added IPython notebook demonstrating reading localization information with TalReader (thanks @LoganJF !)

  • Replaced custom typing system with the traits package (b2f4609d)

  • Improved TimeSeries.to_hdf (#138)

Version 1.1.5

2018-2-1

Summary of changes:

  • Removed further debug printing from Morlet filters (#111)
  • Cleaned up ptsa.data.readers to be more in line with PEP8 naming (#112)
  • Added support for reading EDF files (#113)
  • Included h5py in conda requirements (#118)
  • Suppressed unhelpful messages by default (#121). These can be "re-enabled" by adding a non-null log handler.
  • Fixed incompatibility between TalReader and pairs.json (#116)
  • Added a LocReader class that produces a flat view on localization.json files

Version 1.1.4

2017-12-01

Summary of changes:

  • Removed cerr debugging output from compiled extension modules (#93)
  • Changed to use h5py instead of pytables whenever HDF5 files are involved (#94)
  • Fixed behavior of EEG reader to warn when removing "bad" events and optionally disable this behavior with a keyword argument (#95)
  • Updated conversion from structured arrays to avoid potential issues in numpy 1.13 (#103)

Version 1.1.3

2017-09-28

New Features

  • Added H5RawReader as a subclass of BaseRawReader, to read raw EEG data stored in HDF5 format - H5RawReader dataroots should have a file extension (e.g. '.h5'), as opposed to BaseRawReader dataroots

    which should not have a file extension

    • H5RawReader (and EEGReader, when reading data from HDF5 files) allow one to pass an empty list of channels to read, in which case data from all channels will be read, similar to passing -1 as the read_size to read an entire session.

  • EEGReader returns TimeSeriesX with 'bipolar_pairs' axis instead of 'channels' axis when loading data recorded using bipolar referencing scheme.

Bug Fixes

  • Fixed bug in JsonIndexReader in which passing two conditions with the same value (e.g "session=0,montage=0") would cause the reader to not return any values.
  • BaseEventReader.as_dataframe() excludes the 'stim_params' field from the DataFrame it returns, since Pandas doesn't support nested DataFrames.

Version 1.1.2

2017-08-29

  • Added support for monopolar structures to TalReader
  • Added 'float32', 'float64' as alternatives to 'single','double' in BaseRawReader.file_format_dict
  • Added as_dataframe methods to BaseEventReader and JsonIndexReader to simplify usage.
  • Saving timeseries to HDF5 now includes attributes describing the PTSA version and creation time.

Version 1.1.1

2017-06-20

  • Patched MorletWaveletFilter, ResampleFilter, ButterworthFilter classes to work with the new TimeSeriesX constructor.
  • Fixed bug in which filtering on the value of a field could fail if that field was not consistently present.
  • Added support for recarray coordinates with unicode in the to_hdf and from_hdf methods of TimeSeriesX.
  • Simplified importing JsonIndexReader.

Version 1.1.0

2017-06-06

  • Added new demo suite (anotated ipython notebook examples)
  • Improved documentation (currently still under development)
  • Added conda installer for easy deployment
  • Expanded test suite
  • Cleaned up docstring documentation
  • Provided support for both Python 2.x and 3.x on Windows, Linux, OSX
  • Added Continuous Integration system to the development pipeline
  • Added CMLEventsReader (CML stands for Computational Memory Lab) that by default reads events data "as-is" without doing any pre-processing
  • Serialization of TimeSeriesX object to HDF5

Bug Fixes

Fixed bugs related to missing samplerate in the TimeSeriesX. As of now TimeSeriesX by default will include samplerate attribute

Known Issues

  • BaseEventReader and CMLEventReader are not "fool-proof" and may misinterpret types of certain columns and replace NaN with random integers This is due to the fact that numpy does not allow marking NaN in sht array of integers. Suggested solution is to use curate events files and replace NaNs with sentinel values (as was done for RAM dataset)
  • to_hdf function of the TimeSeriesX does not work when elements of the structured array it tries to save are unicode. This is a known limitation of the h5py library. The temporary workaround it to replace all unicode strings with ASCII based equivalents