Warning in svm-scale libsvm -


i'm using libsvm-3.21 epsilon-svr. have training data many non zeros (sparse format). when use svm-scale scale features range [0, 1], i'm getting warning

warning: original #nonzeros 503981        > new      #nonzeros 6450944 if feature values non-negative , sparse, use -l 0 rather default -l -1 

should ignore warning, affect predictions ?

sparse inputs can processed more efficiently. if can scale data in way preserve zeros training model might quite bit faster.

and model takes less time train might end giving better results since have more time optimising parameters.


Comments

Popular posts from this blog

javascript - Hide toolbar of pdf file opened inside iframe using firefox -

Copy range with conditional formatting -

Ansible - ERROR! the field 'hosts' is required but was not set -