Jon Bratseth
Jon Bratseth
Today we take the idf value from the current index when needing it for relevance scoring of a document. With uncommon terms and small memory indexes this can lead to...
We currently take the median number of douments in the other groups to decide if a given group should be in rotation. This does not work when a majority of...
If clients feed very long tokens, or long text in fields without tokenization (match:exact/word), Vespa will produce an index containing them which is wasteful and possibly damaging. - Add a...
If we are searching a stemmed index, it's probably better to stem terms also when we are searching for prefixes.
Less work to support it than to keep explaining how to do it in your app.
If you have lots of summary-enabled fields and you want a summary containing most, but not all of them (e.g not vectors), you need to repeat lots of field names....
@geirst please review. Sorry for the many commits, this was a bit exploratory. Now that we have contextual type resolving, I want to replace the earlier half-baked attempts at type...
Vespa provides the feature elementwise(bm25(field),dimension,cell_type) to calculate a text match score for each element in an array string field. Generalize this to support elementwise over any rank function, where features...
Today, indexed Vespa documents have a special field `documentid` to which the summary transform `documentid` is applied, such that the field is populated with the full string document id, when...