Error when running skbio.stats.ordination.CA: 'LinAlgErr: SVD did not converge'

Question

I would like to use Emperor to create an interactive PCoA plot outside the context of QIIME. To do this I need to generate an ordination file from my data matrix, like the results provided by skbio.stats.ordination.CA.

My data is a pandas DataFrame of bacterial strains (rows) and the COGs found in their genome (columns). Each column is a different COG, and each strain has between 0 and 4 copies of each COG. There are a lot of zeros in the matrix but no rows or columns that are populated entirely by zeros.

When I try to run stats.ordination.CA as follows:

res_ord = stats.ordination.CA(matrix_dm, row_ids=matrix_dm.index, column_ids=cogs)

where cogs is a pandas Series consisting of the matrix_dm column headers.

I get a very long error, ending with: 'LinAlgError: SVD did not converge'

I've verified that my matrix does not contain any NaNs or Infs. According to a couple other threads, this might be a bug in scikit-bio. Any other thoughts on what might be happening?

Are you confident there are no nan's? Note that `nan != nan`, so you need to use eg. `np.isnan` to check for them. — jorgeca, Dec 05 '15 at 04:08
To triple check this, I ran df.isnull().values.any() as suggested at http://stackoverflow.com/questions/29530232/python-pandas-check-if-any-value-is-nan-in-dataframe and got the output 'False.' — Nastassia Patin, Dec 05 '15 at 15:41
Also ran df.apply(np.isnan).values.any() and got the output 'False'. — Nastassia Patin, Dec 06 '15 at 16:42
Aaaand finally this; df.apply(np.isinf).values.any() - also False! — Nastassia Patin, Dec 06 '15 at 16:49
Thanks, if you can provide a reproducible example it'd be great if you opened an issue on [github](https://github.com/biocore/scikit-bio/issues). — jorgeca, Dec 06 '15 at 21:50

Error when running skbio.stats.ordination.CA: 'LinAlgErr: SVD did not converge'

0 Answers0