speed of initialization for large datasets

[stephens999]

for n=1000, p=10000 we find the current initialization using softImpute is very slow.
Changing type="als" (the default) would be much faster for these large data

[willwerscheid]

This change has been implemented. The issue can be closed.

[stephens999]

do we have a rule (if (n,p) small enought then use svd else use als)? or do we just make the user decide this manually?

[willwerscheid]

Default is "als", user must specify "svd". Should we look into this some more?