PCA and ICA return ValueError

[bainzo]

Running a simple test,

import thunder as td
from factorization import PCA,ICA
ts = td.series.fromexample('fish')
algo = PCA()
w,t = algo.fit(ts)

fails with,

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-7-350246bb024c> in <module>()
----> 1 w,t = algopca.fit(ts)

/home/ec2-user/anaconda2/lib/python2.7/site-packages/factorization/base.pyc in fit(self, X)
     15 
     16         if data.mode == "local":
---> 17             results = self._fit_local(data)
     18 
     19         if data.mode == "spark":

/home/ec2-user/anaconda2/lib/python2.7/site-packages/factorization/algorithms/PCA.pyc in _fit_local(self, X)
     14 
     15     def _fit_local(self, X):
---> 16         t, v = self._fit_spark(X)
     17         return t.toarray(), v.toarray()
     18 

/home/ec2-user/anaconda2/lib/python2.7/site-packages/factorization/algorithms/PCA.pyc in _fit_spark(self, X)
     23         from thunder.series import fromarray
     24 
---> 25         X = toseries(X).center(1)
     26 
     27         svd = SVD(k=self.k, method=self.svd_method, max_iter=self.max_iter, tol=self.tol, seed=self.seed)

/home/ec2-user/anaconda2/lib/python2.7/site-packages/factorization/utils.pyc in toseries(y)
     11 
     12     if len(y.shape) != 2:
---> 13         raise ValueError("factorization on for 2-dimensional arrays")
     14 
     15     return y

ValueError: factorization on for 2-dimensional arrays

In both PySpark and regular python.

I've also had similar errors for ICA with this type of test and my own tif data.

I'm running Thunder 1.0.0 and Spark 1.6.0.

[jwittenbach]

@bainzo This is one of the aspects that we want to address before we publish the package. With the original design, all of the factorization algorithms only work with two-dimensional objects. It's easy enough of the user to ravel/unravel higher-dimensional data to on passing it into/out of the algorithms, but we realized that we should probably just do that for them automatically.

[bainzo]

@jwittenbach Ahh, that makes sense.

Is there a method like flatten that can reshape the Series object?

[bainzo]

I've opened this as issue #335 on the main thunder repo as suggested by @jwittenbach

[jwittenbach]

Just merged a big update that address this (#5). Now factorization can be done direction on Images and Series objects from thunder and shapes will be conserved. I'm going to close this, but definitely open a new issue if there are any problems with the new functionality~