Adrien Grand

I like the idea, but this seems to come with greater heap requirements as well?

Sorry for the very long delay. This sort of detection of loop invariants is indeed handled by C1/C2 compilers, not javac. Without a benchmark suggesting that there is no performance...

Have you checked if this actually made things faster? We're saving calls to `advance` in some cases but also adding conditions to some very tight loops. E.g. we recently got...

I believe that PISA does not require changes though it would be nice to make the BM25 configuration more explicit in the query logic, what do you think @amallia? I...

It does not indeed. I would like to change it but I am not familiar with Rust and unsure how to do it. I could use some help.

As a counterpoint, @rmuir pointed me to the [DFR paper](https://theses.gla.ac.uk/1570/1/2003amatiphd.pdf) which shows that BM25 with k1=1.2/b=0.75 happens to closely match the parameter-free `I(n)L2` model, giving further evidence that k1=1.2/b=0.75 are...

Separately I checked more search engines and IR toolkits: - PISA, Anserini, JASS, ATIRE use 0.9/0.4 - Terrier, Vespa, Lucene, Tantivy use 1.2/0.75 So there doesn't really seem to be...

To get a sense of the influence of these parameters on query performance, I compared Lucene-9.8 with 1.2/0.75 against 0.9/0.4 on the `TOP_100` command. I'm getting: - 4.6% better latency...

For the record, I tested how Rally behaved with these stats removed and it looks like it handled it gracefully. The output was a bit smaller than usual as Rally...

This is how I started, but I would really like to see how all queries perform when a filter is applied, not just a few of them, and duplicating all...