On Tue, Feb 14, 2012 at 2:25 AM, Lance Norskog wrote:
> ...
> OnlineLogisticRegression allocates DenseVector/DenseMatrix objects- if
> it used RandomSparse Vector/Matrix could it operate on million-term
> sparse arrays?
>
Not likely.
The feature vectors that come in are sparse and the updates t
This is in the context of playing with the example classification
scripts in mahout/examples/bin.
OnlineLogisticRegression allocates DenseVector/DenseMatrix objects- if
it used RandomSparse Vector/Matrix could it operate on million-term
sparse arrays?
The problem is that seq2sparse has several te
If you don't use hashed encoding you lose the single pass nature of the
example. Also many real applications require huge vocabularies which make non
hashed representations infeasible due to memory use in the logistic regression
models.
Sent from my iPhone
On Feb 12, 2012, at 20:53, Lance No
Ah! Ok. The SGD examples in examples/bin/asf-examples.sh and
examples/bin/classify-twentynewsgroups.sh both use hash vectorization.
Should they use the sparse term vectors instead? The "new" Bayes
examples (nbtrain and nbtest) in asf-examples.sh use sparse.
On Sun, Feb 12, 2012 at 7:00 AM, Ted Dun
Hash coded vectorization *is* a random projection. It is just one that
preserves some degree of sparsity. It definitely loses information when
you use it to decrease dimension of the input. It does not "add bogus
information".
SGD doesn't like dense vectors, actually. In fact, one of the nice
Does hash-coded vectorization add bogus information compared to sparse
term vectors? A more concrete question: would a random projection on
the sparse vector give a "better quality" dense vector? (This is in
the context of SGD classification, which "likes" dense vectors.)
--
Lance Norskog
goks...
