All notable changes to this project will be documented in this file. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Relaxed the requirements for
numpy
,pandas
, andh5py
. (#27)
- Added
set_train_domain
,set_valid_domain
, andset_test_domain
methods toData
to set the domain and split masks at once. (#21)
- Changed the
keys
method tokeys()
to be consistent with other packages. (#22)
- Fixed a bug where a
LazyData
object is instanitated, but the class does not exist, andData
should be used instead. (#17) - Fixed a bug where
is_dijoint
callssort
incorrectly causing an error when evaluating unsorted intervals. (#20)
- Added a
domain_start
attribute to theRegularTimeSeries
object to simplify the creation of the domain. (#8) - Added an automated way of resolving
domain
forData
objects by infering it from the domains of its attributes. (#7) - Added documentation. (#6)
- Added special keys with the
_domain
suffix. These keys are exluded fromadd_split_mask
and_check_for_data_leakage
. (#2) - Added warning when
timestamps
orstart
andend
are not innp.float64
precision. (#5) - Added
materialize
method to lazy objects to load them directly to memory. (#3)
- Changed slicing behavior in the
RegularTimeSeries
to make it more consistent with theIrregularTimeSeries
object. (#4 #12) - Changed
repr
method of all objects to exclude split masks anddomain
attributes. (#10)
- Deprecated
trials
as a special key that is not checked for data leakage. (#2)
- Fixed a bug where
absolute_start
was not saved to hdf5 files. (#9)
- Initial release of the package.