yadda yadda yadda

Results 55 comments of yadda yadda yadda

Was hit by this issue in production, so it's not a theoretic problem. We use zstd as a compression format for a large batch of data files using Apache Flink...

And also add response examples! :)

Hey @ravichoudhary33 , We even have a ticket for this: https://github.com/metarank/metarank/issues/575 While the "how it works" guide is not yet written, we can share a general overview of the data...

As an idea, store API can be also used not only to dump feature store state, but also to compute feature values without running the ML inference. Some sort of...

Inspired by this talk: [slides](https://mices.co/mices_eu_2022/slides/nate-day_ranking-factors-going-beyond-just-relevance.pdf) and [youtube](https://www.youtube.com/watch?v=K6CYAWef3SU&list=PLDF9_EdANJIsTnFUpgFDaNZHuo8VpHIcV&index=3)

The root cause is: https://github.com/actions/virtual-environments/issues/672 So LightGBM native library uses too much enthropy on training, and GHA VM has some issues with randomness sources. Right now we don't train the...

with the flink removal the issue seems to be somehow fixed. classloader and jni issue?

So the upstream blockers are https://github.com/dmlc/xgboost/issues/7501, https://github.com/dmlc/xgboost/issues/6408 and https://github.com/microsoft/LightGBM/issues/3606. As we use our own native lib wrappers (see https://github.com/metarank/xgboost-java and https://github.com/metarank/lightgbm4j), we in theory can build these libs manually if...

@ChakshuGautam if you're OK with it, we would like to reuse parts of your solution, but make it a documented way of running Metarank

I guess the most problematic part of this feature is being unable to bundle the maxmind geoip database with the app: even geolite2 does not allow bundling. So user should...