Releases: quantling/pyndl
Releases · quantling/pyndl
pyndl 0.8.2
- improves the documentation on Widrow-Hoff learning
pyndl 0.8.1
- adds Widrow-Hof learning on Linux (real_to_real, real_to_binary, binary_to_real)
- adds fast correlation code
- drop python 3.6 support and add python 3.9 support
- changes API from
number_of_threads / number_of_processes
ton_jobs
- exposes make_data_array as ndl.data_array(...) and adds tests
- adds learning events from generators in ndl.ndl
- changes tuples into namedtuples for return values
- adds sanity check for too many cues or outcomes
- checks if outcomes_vectors, cue_vectors and weights are c_contiguous before passing it to cython functions
- adds scipy, packaging as dependencies
Co-authored-by: Elnaz Shafaei [email protected]
and Marc Weitz [email protected]
pyndl 0.7.2
- changes tests to be run on mac with threading
- drops support for python 3.5 due to the usage of format strings
- adds testing for python 3.8
- corrects grand information in zenodo and adds Acknowledgements in readme
pyndl 0.7.0
- pyndl should now be installable on MacOS without openmp support
- splits
ndl_parallel.pyx
into several files to separate the openmp code from the rest - some maintenance improvements
pyndl 0.6.4
-
fixes math.ceil with explicit cast to double before ceiling
-
makes error codes in ndl_parallel more human readable
pyndl 0.6.3
- fixes error code -166... (issue #85 on github)
- the (now fixed) error lead to pyndl not running under specific circumstances
- general code improvements
pyndl 0.6.2
- fixes encoding error in long description of setup.py
- a lot of small code improvements
- improvements in the maintenance code
pyndl 0.6.1
- fixes integer overflow in uni-dimensional index of weights in
ndl.ndl
that results in wrong learning in weight matrices with more than 4294967295 cells - please relearn any weights where number of cues times numbers of outcomes exceeds 4294967295 (sorry for this bug)
- thanks to Yuying for finding it
pyndl 0.6.0
- Moves code to read and write events into
pyndl.io
module. This changes the existing API! Deprecation warnings are added accordingly. - improves documentation
pyndl 0.5.1
- removes duplicate outcomes, when remove_duplicates=True was given in preprocess.py (thanks to Christian Adam @kuchenrolle for spotting and fixing the problem)