tomtung

Unfortunately using pre-defined tree structure is not yet supported. I won't have time to implement it soon, but contributions are welcome :) Tree-building is half of parabel's job, and you...

> Hmm, but cbor files keep not only tree architecture information, but also the classifier themselves, right? That's correct. > Creating trees is an intermediate step in parabel, but can...

It does [beam search](https://en.wikipedia.org/wiki/Beam_search), as described in the paper as well.

Yeah it should be possible in principle: as it walks down the trees during prediction, it should be possible to keep track of, e.g., how much each feature contributed to...

Thanks for the suggestion! @rabitwhte I think LIME should be general enough, although it might be a bit slow depending on your use case. If it's useful I can also...

I tried confirming the speed regression with a quick & scrappy benchmark. I first cloned sprs locally, and patch it to add back the previous `csr_mulacc_dense_colmaj` implementation with a suffix...

Yep, the difference seems to widen. With: ```diff --- a/Cargo.toml +++ b/Cargo.toml @@ -29,6 +29,7 @@ smallvec = "1.4.0" rayon = { version = "1.3.0", optional = true } num_cpus...

So I had a bit more fun debugging the cause of the speed regression. This is a tricky one because there turned out to be 2 causes for the slowdown:...

Hi @aujxn, thanks for looking into this! Sorry about the delay; I've been a bit busy at work lately. I cloned your fork and ran the benchmark. The full result...

> Is there any reason other than the potential efficiency improvement of using mul_add? I guess since `mul_add` is performed in one step with a single rounding, while doing `mul`...