I did iterative clustering with nmf package.
nmf
In detail, I ran nmf function 100 times every time with 500 random observations (sampled from ~1500