Is this the idiomatic way to take a two column Matrix and create an average for each column?

@User uploaded a file: Untitled

@thomasnield yeah that should work. you can also use

mapCols

if you want a 1x2 matrix output instead of a list

Interesting, I can feel the R influence here.

@User uploaded a file: Untitled

This might need some extension functions on native Kotlin iterable types, like

toMatrix()

or

toEye()

. Overloads can specify dimensions.

yeah for sure

I'll keep digging and playing with this. I've always found scientific libraries somewhat esoteric. While this does have that to some degree, it feels pretty good so far. I'll have to dig through all the functionality and depending on how much work there is to be done, I can help document stuff.

I helped Edvin document TornadoFX (https://github.com/edvin/tornadofx) and broke ground on the eBook (https://edvin.gitbooks.io/tornadofx-guide/content/). That helped the project have immense success in just one year. I think documentation will be crucial to give your work momentum and adoption as well, and for any Kotlin DS projects we do.

sounds great for sure. definitely open issues if you find anything-- and help is always welcome. documentation unfortunately is the part that often slips for my "fun" projects

@kyonifer absolutely. I can help on the documentation front as things start to get steam. I enjoy technical writing in a warped way, but at least it got me doing stuff for O'Reilly and Packt! It's fun to use it to help drive adoption for projects too.

@kyonifer one thing that is confusing in usage is the Matrix implements Iterable and picks up all those Kotlin stdlib extension operators. It's hard to separate Matrix-specific operators from stdlib ones, such as mean() and average(). How much refactoring effort would it be to have a Matrix method return an Iterable instead of implementing it?

should be a quick fix, one that i'd considered doing. I agree that

mapMat

vs

map

is non-obvious

along with the other clashes

the downside is that there will be more

toIterable()

and

toMatrix()

conversions floating around, but that may be a reasonable compromise

@thomasnield went ahead and patched it at https://github.com/kyonifer/golem/tree/iterable_change

probably should change

mapMat

->

map

now that the name is free

People enjoying technical writing are pure gold in OSS world, @thomasnield :D

I like Golem!! Once I have my data well structured I'll take a look at it

@kyonifer wow that was quick. Awesome. Ill keep playing with it and keep massaging out an API that feels natural. Great to see all the hard work under the hood seems to be done.

@holgerbrandl I started to file some issues with Krangl... they might just be questions. Once i get a good feel for it ill help write some documentation.

@holgerbrandl i see you responded to my iissues will follow up soon

You know what would be awesome? A plugin that builds data classes from tabular data sources...

@holgerbrandl I followed up finally. I'm going to start using this at work and I'll probably have plenty of more input and feedback if there is anything needing to be ironed out.

Just a random PSA: A data wrangling and analysis project I'm doing at work with Kotlin is flying together using stdlib, Rx, and TornadoFX. If anybody builds models or prototypes that would benefit from getting a user interface (so others can interact with it) #tornadofx is phenomenal for this purpose. It's a rapid-turnaround, comprehensive Kotlin UI library.

I’m working on #ktor (web framework) and in background thinking about balancing requests queueing in a worker thread pool. The main issue here is when there are several long-running requests, they could fill the queues and other requests that otherwise would be instant are spending their time in queues. Of course, there are plenty of balancing algorithms with this or that idea behind them, like “keep a fast lane for specific preconfigured requests”, etc . I wonder if it is possible, for a given application, train a machine to predict processing time from the URL alone, and then schedule it based on the estimation. It shouldn’t be very precise, but it should train fast. Any ideas?

@orangy The main problem is you need a time to gather sufficient data to train the model. Moreover, an execution time of specific request should not vary so much to provide stable behaviour of the model. But I believe that for given application the problem can be solved. So I need some time to think about it. I hear that Netflix used ML to tune JVM parameters well.

Good thing, I suppose, that it could be possible to have a nice starting model trained for some synthetic “normal” application. I don’t know if there are models that are “adaptable”, i.e. that continue training during normal work