These are chat archives for thunder-project/thunder

Mar 2015
Gilles Vanwalleghem
Mar 24 2015 06:22
Hi, after installing thunder on our cluster I started playing around a bit with it, our data are image time series (I use 'loadImagesAsSeries'). It looks good and with the help of the tutorials I managed to use most of the functions. But I am starting to try and make the regressions to work and it is not as clear.
As far as I understand it, the design matrices and stimparams (for tuning) are all binary matrices, right ?
In our case we have "simple" stimuli which are just on/off, I was guessing I should then just use 1D matrix for my design matrix, but it doesn't seem to be working. Looking at the paper, should I make a s*t matrix with a 1 in column s[0]t[i], then 1 in s[1]t[i+1] and so on ? (t = length of time series)
Mar 24 2015 06:59

anyone who runs the code ">> from thunder.factorization import PCA

data = tsc.loadBinary('data')
pca = PCA(k=3)

successfully.Because I can't got the data
who can help me ?
Mar 24 2015 12:52
This message was deleted
Mar 24 2015 12:53
and I can changed to 0.3.2
Jeremy Freeman
Mar 24 2015 15:20
@wolfbill sorry, that website needs to be updated to explain where to get the data for the examples, will do so soon
Jason Wittenbach
Mar 24 2015 16:59
@Yassum The Regression analysis assumes a model y=βX where y represents an arbitrary record in your Series and β is a row-vector of regression coefficients. So if the records in your Series are of length T and you have S different regressors (each also of length T), then X should be a S-by-T matrix where each row contains the values of one of your regressors.
Jason Wittenbach
Mar 24 2015 17:04
Assuming that records in your Series are vectors representing time, then each row of X will be the time-course of one of your stimuli/regressors.
Gilles Vanwalleghem
Mar 24 2015 23:52
I see, that's what I gathered, but from the implementation, it doesn't look like you can only have one stimulus in your X, although I'm not sure if that would make sense. For the BilinearRegressionModel, if I understand well the "betas" you get are based on the dimensionality of X2 ?