29339 (1) [Avatar] Offline
#1
Hi,

I tested the code in "Listing 8.9 Indexing a feature vector as a point", but failed in step 3.

It shows that the dimension count of a Lucene FloatPoint must be <= 8, but that of the image representation is 1024 in this case. How can I fixed this issue?
al.krinker (16) [Avatar] Offline
#2
Same here... 8 seems to be the max according to Lucene Doc API. Not sure how the author got around this


/**
* Enables points indexing.
*/
public void setDimensions(int dimensionCount, int dimensionNumBytes) {
if (dimensionCount < 0) {
throw new IllegalArgumentException("dimensionCount must be >= 0; got " + dimensionCount);
}
if (dimensionCount > PointValues.MAX_DIMENSIONS) {
throw new IllegalArgumentException("dimensionCount must be <= " + PointValues.MAX_DIMENSIONS + "; got " + dimensionCount);
}
43632 (4) [Avatar] Offline
#3
you're both correct, the latest MEAP version doesn't contain the full code yet.
before you go with indexing features as float points you need to perform dimensionality reduction, e.g. via PCA and the likes.
I've incorporated such missing bits and you'll find them in the final revision.