Custom Metrics and Distribution Function Information

[Kerri Rapes]

Hi,

I'm Kerri, I'm a Data Scientist and lead of the Fraud team at Sendwave. We would like to implement H2O into our pipeline but require specific prioritization of our False Negatives. I believe this would be a great case for the custom_metric_func and custom_distribuition_func but I'm having trouble finding enough documentation to completely understand what is happening is these classes, how the methods are being called, what the different variables are. 

I've worked through the tutorials in the github repo but still struggle to 100% comprehend. Do you have any documentation, source code, examples, webinars, etc you can point me to? Or is there someone I can talk to to clear up the gray areas?

Thanks in advance,

Kerri 

--

--

Kerri Rapes (she/her) Risk Team Lead | Sendwave +56 9 4907 6529 ke...@sendwave.com www.sendwave.com

[Tom Kraljevic]

hi,

here is a good doc to look at.

it contains some references to other resources in the github repo.

https://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/custom_distribution_func.html

the examples show the custom metric being written as a python class, and underneath the hood h2o-3 is evaluating it using jython.

thanks,

tom

--
You received this message because you are subscribed to the Google Groups "H2O Open Source Scalable Machine Learning - h2ostream" group.
To unsubscribe from this group and stop receiving emails from it, send an email to h2ostream+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/h2ostream/CAL2kNcL8xQPiou%3DVtwNRZ87QyO3WTuZVxpuT34QbUBYsOuHEDw%40mail.gmail.com.