I played around with DL4J a bit but found the syntax for declaring neural networks a bit lacking. To...

# datascience

[Tilman Krokotsch]

I played around with DL4J a bit but found the syntax for declaring neural networks a bit lacking. Too Java for my taste. As I am currently trying to pick up Kotlin, I coded a small "library" to help with that using the type-safe builder patterns from the tutorials. You can find it on GitHub or read about it on my blog if you are interested.

[mccorby]

I think DL4J has some things in Kotlin already. It’d be good to check with them.

[Tilman Krokotsch]

I had a look at the examples and some had variants in Kotlin. But they looked like stuff generated by the Java2Kotlin tool from ItelliJ. Do you have a link, by chance?

[Michal Harakal]

@Tilman Krokotsch did you try it in kotlin-jupyter ?

[Tilman Krokotsch]

No, I didn't. I don't like notebooks that much for my day to day work. Why are you asking?

[Michal Harakal]

I am just curious, I guest, since its a jvm library, it can be added to a kotlin kernel the same way as any other, I will check that later…

[altavir]

There are modules for dl4j in kotlin jupyter already

[Michal Harakal]

I can confirm this is indeed working even with klay from jitpack

%use deeplearning4j
@file:DependsOn("com.github.tilman151:klay:-SNAPSHOT")

import org.klay.nn.* 

val conf = sequential {
                    layers {
                        dense {
                            nIn(10)
                            nOut(100)
                            activation(Activation.RELU)
                        }
                    }
                }
print(conf)

[Tilman Krokotsch]

@Michal Harakal now I am getting what you meant. Yeah, as klay is only plain old Kotlin it should all work out of the box^^

[altavir]

You can use plain java libraries in kotlin-jupyter. The only thing you can not are non-jvm librararies like native

[Tilman Krokotsch]

Sure, but my project was not getting dl4j to work in kotlin, but making it look better^^

[Michal Harakal]

@Tilman Krokotsch yeah DSL helps a lot, I liked your article explaining the step from builders to dsl.. Even in jupyter it works quite well with context sensitive help etc.

[Tilman Krokotsch]

@Michal Harakal thanks a lot

[Michal Harakal]

Last time I was hearing data science podcast from 2015 where ipython and python was hot thing, I hope have the same discussion about kotlin, hopefully sooner 🙂

[altavir]

@Michal Harakal I give a talk about that in a meetup tomorrow 🙂

[Michal Harakal]

do you have a link?

[Tilman Krokotsch]

I would like that. Just today I had to work with multiprocessing in Python again...the pain

[altavir]

@Tilman Krokotsch I do not actually do ML, but I think thigs like the one you did are quite important for the ecosystem

[altavir]

@Michal Harakal. It is in Russian. But it mostly follows my KotlinConf-19 talk.

[Michal Harakal]

I could refresh my grammar school russian 🙂

[Tilman Krokotsch]

@altavir Thanks, maybe someone will use my stuff one day^^
I would like to see your talk, too, but I do not speak russian. Do you have a link to your KotlinConf talk?

[altavir]

Here you are:

https://youtu.be/LI_5TZ7tnOE▾

. The problem with visualization seems to be mostly fixed since that time.

[Tilman Krokotsch]

Thx