These are chat archives for thunder-project/thunder

30th
Mar 2015
Gilles Vanwalleghem
@Yassum
Mar 30 2015 03:00
@jwittenbach Thanks, that works perfectly. I am trying to use the direction tuning as described in the paper. I am trying to see direction selectivity in my neuronal GCaMP data, I have a bar moving horizontally in one direction, the in the opposite direction (half the stimulus time for each direction). From what I understood of the paper, I do bilinear regression based on X1 for the stimulus and X2for the directions, then I do circular tuning model on the betas. I am trying this code, but the spread I get back is zero for all my neurons. Am I doing something wrong or are my neurons not direction sensitive ?
L = len(data.index)
X1= np.zeros((1, L),dtype=np.int)
X1[0,0:30] = 1
X2= np.zeros((2, L),dtype=np.int)
X2[0,0:15] = 1
X2[1,15:30] = 1
bilinreg = RegressionModel.load((X1,X2), "bilinear")
results = bilinreg.fit(data)
circ = np.array([-pi, pi])
model = TuningModel.load(circ , "circular")
params = model.fit(results.select('betas'))
plt.plot(params.select('spread').values().collect())
Jason Wittenbach
@jwittenbach
Mar 30 2015 15:21
@Yassum I'm not 100% sure about how to construct the matrices for the bilinear model. However, I do know that, for the circular version of the TuningModel, you need more than just two directions. The tuning model will try to fit a function of response vs angle, and just having forward/backward won't be enough to get a meaningful fit there.