search

dipenps / declassification

0 stars 3 forks source link

DSA implementation #1

Open oganm opened 5 years ago

oganm commented 5 years ago

I have been looking into the results from the DSA package and the implementation here and the results seem to be fairly different, though I didn't have time to examine the code in detail to figure out why. Was wondering if you could offer some insights into the origin of this difference.

oganm commented 5 years ago

Looking at the code, the difference seems to be your use of non negative matrix factorization vs the original DSAs unrestricted factorization. Honestly this one performs better. Don't know why original dsa went with unrestricted factorization