Closed ebergstr closed 4 months ago
Thanks for filling this out Erik. After additional testing was done, the tolerance value is properly being passed and the maximum number of iterations was being run due to tolerance criteria not being met. This was confirmed by running two different tests with the same seed for 21BRCA data with the default tolerance value and 1e-2 and comparing the iterations for each replicate. The number of iterations for the 1e-2 tolerance were less than or equal to that of the respective replicate with default tolerance.
This issue is being closed because this issue arose from a non-standard matrix size that may benefit from more iterations or a larger tolerance.
Just as a placeholder for us to look into further...
After several tests, it appears that the extractor always runs the maximum number of NMF iterations independent of the precision and tolerance. For instance, I have even increased the tolerance as high as 1e-2 with single precision and it will still run the maximum of 1 million NMF iterations each time.
This should not effect the final results, but will drastically increase the total run time for each extraction.
Best, Erik