Version 0.4.0

[nirum]

Listing updates that I want to include in a 0.4.0 release. :beer:

Containers

:package: Adds a containers module that contains two classes, and Experiment and a Filter class, for managing stimuli and spikes (Experiment) and spike-triggered averages (Filter).

Example usage:

>>> ex = Experiment(stimulus, stimulus_time, spikes, dt)
>>> sta_cell0 = ex.sta(0, history)
>>> sta_cell0.plot()

New ellipse and contour fitting code

• pyret.filtertools.get_ellipse - gets the ellipse parameters (centroid, widths, and rotation angle) from a 2D image of the spatial RF profile
• pyret.filtertools.rfsize - returns the lengths of each axis of the ellipse (rough measure of the size of the RF), given a spatial RF and a spatial sampling rate
• pyret.filtertools.get_contours - wrapper for scikit-image.measure.find_contours
• pyret.filtertools.get_regionprops - wrapper for scikit-image.measure.regionprops
• pyret.visualizations.ellipse - plots an elliptical contour of an ellipse fit to the given 2D spatial RF profile
• pyret.filtertools.normalize_spatial - a helper utility for normalizing a given 2D spatial receptive field by mean subtracting and flipping the sign such that the RF is an 'on' feature (useful for fitting 2D gaussians to the spatial RF)

  Breaking changes

• :seedling: New function filtertools.resample, which given a 1-D or 2-D array and a scaling factor (n), resamples the filter to have n times the sampling rate of the original filter (based on scipy.signal.resample)
• :watch: Changes the default value of the argument in rolling_window to time_axis=0, to be consistent with the rest of pyret (after the flipped dimensions switch)
• :recycle: Removes the prinangles function (does not really belong in the filtertools module, or even in pyret at all)
• :chart_with_upwards_trend: Updated pyret.plotsta function

[bnaecker]

Everything looks fine, but I have a question about the new filtertools.resample method. SciPy's resample is a Fourier-based method, which means it really will not work well on things like spike trains (binned or not). There will be a ton of high-frequency ringing. Run this and see:

from scipy import signal
t = np.arange(1000)
sig = np.zeros(t.size)
sig[np.random.rand(sig.size) > 0.98] = 1
sig_res, t_res = signal.resample(sig, 2 * sig.size, t=t)
plt.plot(t, sig)
plt.plot(t_res, sig_res)

If we only plan to use this on stimuli or smoothed spike trains, then it's fine. We just need to bin spikes using resampled time bins, rather than resampling the binned spikes.

[nirum]

Oh yes! good point. I meant for the resample method to only be used on filters, which is why it is in filtertools. I'll add a note in the docstring mentioning that it is to be used for filters and not spike trains

[bnaecker]

Cool, sounds good. I'm on board with the list otherwise.

[nirum]

Ok I'll update the docs and merge