Is there a way where I could use the neuron models to learn the optimal weights using spike trains of an entire dataset? For example, I want to perform classification on the Iris Dataset using Spiking Neural Networks. I have my input and output encoded as spike trains, is there a way to use your model for training the entire dataset, instead of just one point? Thank you!

When I try to plot a psth diagram as in the Jeffress tutorial, I get the attached error.

psth = plotting.PSTH(spiketrain, binsize=5)
psth.show_plot(neuron_indices=[8, 9, 10])
plt.show()

Machine: Windows 64 bit, Python3

<class 'numpy.ndarray'>
Traceback (most recent call last):
File "jeffress_test1.py", line 41, in
psth.show_plot(neuron_indices=[8, 9, 10])
File \Python34\site-packages\neurons\plo
tting.py", line 72, in show_plot
axis.hist(times, bins, histtype='bar', stacked=True, fill=True, facecolor='g
reen', alpha=0.5, zorder=0)
File \site-packages\matplotlib\axes_axes.py", l
ine 5597, in hist
raise ValueError("x must have at least one data point")
ValueError: x must have at least one data point

Is it possible to accomplish some application such as pattern recognition using this framework?

Would be nice to have an unsymmetric learning window for STDP learning. At the moment there is only the paramater tau, which influences both the left (x<0) and the right (x>0) side.

See title.

The Izhikevich model in 'spiking.py' doesn't seem to work right.

I suspect this line: 'I = I + np.sum(w[fired, :], axis=0, keepdims=True)' to be false.

I don't like the way that I implemented the pytests.

Rewrite them in a better way, so that I can use them more easily.

Ideas: Learning and Spiking models as fixtures

Also, I notice a mistake in your implementation. I'm not sure, but you could check. At any timestep, if the potential of a neuron exceeds the threshold you let it spike. But it has to spike only once per threshold crossing (previous potential < threshold and current_potential > threshold). Don't you think so?

s = np.array([[1, 1, 0, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 1, 1, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])

Output:
[[1 1 0 1 1 1 1 0 0 0 1 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 1 1 1 1 0 1 0 1 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 1 1 1 1 1 1 1 0 0 0 0 0 0 0]]

You get a series of consecutive spikes instead of just one.

In the SRM class, please recall the function nu to eta, to be consistent with the documentation for the model.

can we model the 3d sound perception of birds?

The idea is to use PyCuda to bring the spiketrain and weight arrays on the GPU and to manipulate it there.

The easiest solution would be to make two classes: One 'SRM_Cuda' in spiking, and 'STDP_Cuda' in learning.py, that take cuda memory references instead of numpy arrays as first argument.

At the moment the plotting api is pretty inconsistent.

Maybe we can make a unified toolbox out of it, which is easy to handle?

A Hinton Plot for weights would be cool. If time allows, maybe also animated?

Until now, the threshhold value for the neurons is unique among all neurons in the SRM model.

Maybe we can define a threshold for each neuron, similar to the way we can define a $t_s$ for each neuron.

Implement a Peri Stimulus Time Histogram plot.

What value should the kernel return in case of a spike?

Idea: a tuple (True, 0) iff spike, (False, value) else.