Eric Hunsberger

signals_allclose doesn't plot allclose_tolerances

2

**Describe the bug** `allclose_tolerances` allow backends to override default `allclose` tolerances for specific tests. These can work with `signals_allclose` by making it use the `allclose` fixture rather than `np.allclose`, but...

NNLS solvers are slow/broken

2

The non-negative least squares (NNLS) solvers have problems. The plain `Nnls` solver is slow, especially when solving for weights. The `NnlsL2` solver is considerably faster when solving for weights, but...

test_eval_point_decoding requires new Matplotlib

4

``utils/test_connection.py:test_eval_point_decoding`` requires Matplotlib > 1.4.3. I was using that version, and got the following error: ``` nengo/utils/tests/test_connection.py:62: in contour zi = griddata(xy[:, 0], xy[:, 1], z.ravel(), xi, yi, interp='linear') ../../venv/full2/local/lib/python2.7/site-packages/matplotlib-1.4.3-py2.7-linux-x86_64.egg/matplotlib/mlab.py:3508:...

Discuss/test `network=None` in `Simulator.__init__`

2

`Simulator` has traditionally supported having `network=None` and just passing in a `Model`. This can be used if a user has a pre-built model (e.g. from a previous simulator), and wants...

Functools.partial does not work properly for node outputs

1

**Describe the bug** When using a `functools.partial` instance as a node output, we internally check `inspect.isroutine` on the output and if this is not true, we subtract 1 from the...

Tiny lowpass synapses break

2

**Describe the bug** A tiny value for `Lowpass.tau` results in an error in Scipy's `expm`. **To reproduce** ```python import nengo with nengo.Network() as net: a = nengo.Ensemble(1, 1) b =...

Sigmoid breaks with many values of max_rates

2

**Describe the bug** Using `Sigmoid` with `max_rates` exactly half of `1 / tau_rc` results in gains of zero and totally flat tuning curves. Using `max_rates` less than this results in...

Test tolerances for backends

3

I often find that different backends require different tolerances on tests. Does anyone have good ideas about how to change that easily in the backend code? Right now, I'm monkey-patching...

Allow encoders to be Transforms

2

With #1481, we're adding a new `Transform` class to represent other types of transforms than full weight matrices. For now, this is just Convolution, but it could include e.g. general...

I briefly looked into using codespell for this, but I couldn't figure out how to get it to look at comments (which is what markdown cells get converted to when...