deeplearning4j / deeplearning4j-examples

Deeplearning4j Examples (DL4J, DL4J Spark, DataVec)

Home Page:http://deeplearning4j.konduit.ai

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Hyperparameter tuning example missing

thhart opened this issue · comments

Issue Description

Only a small suggestion:
I checked out latest examples but could not find a Hyperparameter tuning example like in:
https://deeplearning4j.konduit.ai/v/en-1.0.0-beta7/getting-started/tutorials/hyperparameter-optimization

Side note: The latest documentation https://deeplearning4j.konduit.ai/ (1.0.0-M2) is broken for searching in my Chrome. 1.0.0-beta7 I can search around.

Version Information

Please indicate relevant versions, including, if relevant:

1.0.0-SNAPSHOT

@thhart arbiter has been removed as of M1. You may use the beta7 examples branch if you want. See: https://github.com/eclipse/deeplearning4j-examples/tree/ab_beta7 They were removed due to lack of investment and not enough users to justify continued investment in the module. If you would like to step up and contribute we'd be happy to guide.

I can understand the lack of audience, however from a professional point of view it is a pity this topic fell off the table. I am wondering if it is really overtaken by other practices or better solutions. Maybe you can consider to simply keep this as an approach if it is not as complicated to maintain.

@thhart see: https://community.konduit.ai/t/dl4j-1-0-0-m1-1-released/1515

The main problem is we are migrating a lot of our work to samediff. Implementing whole new code for it isn't really worth it. I understand the appeal but most of dl4j's usage comes from importing models from python and finetuning/deploying them so we're ensuring that becomes easy to do by focusing more on shrinking the framework down.

Tons of people still use the dl4j interface in deeplearning4j-nn and it's still updated for keras import reasons. No reason to get rid of it but leaving the code there just confuses people. It's also more code we have to document/test.