internal ExtractorWordShapeClassifier(int position, string wsc)
: base(position, false)
{
// end class ExtractorFrames
// This cache speeds things up a little bit. I used
// -Xrunhprof:cpu=samples,interval=1 when using the "distsim" tagger
// on the training set to measure roughly how much time was spent in
// this method. I concluded that with the cache, 1.24% of the time
// is spent here, and without the cache, 1.26% of the time is spent
// here. This is a very small savings, which would be even smaller
// if we make the cache thread safe. It turns out that, as written,
// the cache is not thread safe for various reasons. In particular,
// it assumes only one wordshape classifier is ever used, which
// might not be true even with just one tagger, and has an even
// higher chance of not being true if there are multiple taggers.
// Furthermore, access to the cache should really be synchronized
// regardless. The easiest solution is to comment out the cache and
// note that if you want to bring it back, make it a map from wsc to
// cache rather than just a single cache. -- horatio
//private static final Map<String, String> shapes =
// Generics.newHashMap();
// --- should be:
//private static final Map<String, Map<String, String>> ...
wordShaper = WordShapeClassifier.LookupShaper(wsc);
name = "ExtractorWordShapeClassifier(" + position + ',' + wsc + ')';
}
internal ExtractorWordShapeConjunction(int left, int right, string wsc)
: base()
{
this.left = left;
this.right = right;
wordShaper = WordShapeClassifier.LookupShaper(wsc);
name = "ExtractorWordShapeConjunction(" + left + ',' + right + ',' + wsc + ')';
}
