	SPECIAL NOTES ON k-NN

	Query batch size:

When using kNN, Set queryBatch as high as possible; preferably, to be
as large as the total number of topics you have. In kNN, many
expensive things -- such as threshold setting -- cost the same per
batch no matter how large the batch is; so doing it with 5 batches of
10 topics each will cost you almost 5 times as much as with a single
50-topic batch.


	 Parameter file (filter_param):

To choose k-NN as the filter method, set
filterModel = 2; 

You can set 
updateFreq = -1;
for batch filtering; but adaptive filtering is probably still not
supported anyway when you are reading this.

Set queryBatch to a high value (see above).

The names of the configuration parameters specific to k-NN start with
"knn.". Most of them have been documented in David Lewis' 10-21-2002
note; a few have been added for completeness. Please refer to the
sample config file in data/filterparam/filter_param for the up-to-date
list.

Besides the k-NN specific parameters, you should keep doc.tfMethod=1,
to ensure the usual LOG(TF) transformation of document
vectors. 

The usual IDF-based weighting, as in Rocchio, is always used in dot
products.
