Fixed
- Solved an issue where scipy.misc imports were relocated. (#182)
Tested against Nengo versions 2.2.0-2.8.0. Requires nengo<3.0
.
Fixed
- A variety of miscellaneous fixes were made to the documentation.
The
nengolib.networks.RollingWindow
documentation references the shifted Legendre polynomial equations forlegendre == True
. (#176)
Tested against Nengo versions 2.2.0-2.8.0.
We now require numpy>=1.13.0
, scipy>=0.19.0
, and nengo>=2.2.0
.
Added
- Added the
nengolib.RLS()
recursive least-squares (RLS) learning rule. This can be substituted fornengo.PES()
. Seenotebooks/examples/full_force_learning.ipynb
for an example that uses this to implement spiking FORCE in Nengo. (#133) - Added the
nengolib.stats.Rd()
method for quasi-random sampling of arbitrarily high-dimensional vectors. It is now the default method for scattered sampling of encoders and evaluation points. The method can be manually switched back tonengolib.stats.Sobol()
. (#153) - Added the
nengolib.neuron.init_lif(sim, ens)
helper function for initializing the neural state of aLIF
ensemble, from within a simulator block, to represent0
uniformly at the start. (#156) - Added
nengolib.synapses.LegendreDelay
as an alternative tonengolib.synapses.PadeDelay
-- it has an equivalent transfer function but a state-space realization corresponding to the shifted Legendre basis. The networknengolib.networks.RollingWindow
supportlegendre=True
to make this system the default realization. (#161)
Fixed
- Release no longer requires
pytest
. (#156)
Tested against Nengo versions 2.1.0-2.7.0.
Added
- Solving for connection weights by accounting for the neural
dynamics. To use, pass in
nengolib.Temporal()
tonengo.Connection
for thesolver
parameter. Requiresnengo>=2.5.0
. (#137)
Tested against Nengo versions 2.1.0-2.6.0.
Fixed
- Compatible with newest SciPy release (1.0.0). (#130)
Initial beta release of nengolib. Tested against Nengo versions 2.1.0-2.4.0.