What is the language of choice when delving into the topic of machine learning?

Python seemed most common IIRC.

I thought so but just wanted to confirm

scala!

On a serious note, could you tell me why you might use Scala as opposed to [insert most similar language here]?

Ok. As an aside I believe that QTCreator can output qmake files.

Clojure is also gaining steam in the field. Check out Prismatic, Walmart Labs, Target, etc. for some large examples.

Lisp in general has been widely used and it still used to this day for this kind of stuff. See Orbitz, ITA, Factual, [company I used to work for] and of course, the Norvig AI book.

The reason why is because the Lisp syntax is the nearest 1-to-1 mapping you can find for neural nets and other items. It also natively supports immutable data streaming and concurrency. Nth derivatives are very easy to represent, plus 3/2 stays 3/2.

Overall or in the ML world? ML python is obviously the King, but Scala is pretty focused on ML, especially with the rise in Spark, while maintaining type safety.

Overall, functional, good concurrency models, static typing without as much verbosity.


I do need to try out Closure. Not really sure what the point of Kotlin is since its basically just a scala clone, but anyone tried it?

I thought kotlin was a Java clone for android

Yeah I read that kotlin announcement a week or so back and they really did not make a compelling case for choosing kotlin. In fact, I couldn't discern any tangible benefits for kotlin over java.

Something about the version of Java android uses and giving updates that would be included with newer versions of Java or something. Seems like it's a stopgap language that'll be made irrelevant by someone else's update

Its essentially a middle ground for java/scala, and seems like theyre trying to take the best from each and combine them. I read something about it being able to compile to javascript too so you can write your full stack in Kotlin, which sounds interesting but I havent found much more on it. I do like JetBrains a lot, but I still just dont get it

https://medium.com/@octskyward/why-k...6e3#.rb5npbs17


meh

A few months ago I took a Mongo class and ended up talking about my node framework with a guy who works for Microsoft and is something of an evangelist. He runs lots of hackathons where kids tend to use the MEAN stack (I forget how MSFT fits into that).

Anyway he was intrigued by the idea that my framework would make node setup easier. He said a ton of time is taken up at these hackathons just getting a basic web app with mongo up and running. Things he wanted to see was a Yo generator and mongo support out of the box. I'm working on that right now.

My question for the group is - have you had any kind of similar experience at hackathons? And if so what building-block-type features would you like to see available in Yo and or some kind of example MEAN-stack app?

I am not that familiar with mongo. So other than read document, write document - I am not sure what else to add that won't just be FPS clutter. Would hackathon participants be likely to use stuff like indexing, map-reduce or other aggregation? Obviously I don't care about sharding. Anything else?

Is Mongo in particular difficult to set up? When I started messing around with Node I was shocked by how easy it was to get a basic web server (albeit without any third party modules involved) up and running.

No especially with node it's very easy. But I'm guessing some of these hackathon kids aren't super familiar with node or something.

Look into mongoose and passport - mongoose.

Wouldn't the thing to do at a hackathon or a class or something be to have some kind of VM setup with pre-configured VM images? Either vmware or something like docker? I mean, you can get docker installed and set up in probably 10 or 20 minutes I think?

I was just going to use MongoClient. Can you tell me why I need mongoose? What examples should I create for it (over and above find/read/write document)?

Well for whatever reason they don't do that. Maybe because some people are on Mac, some on PC? Does that complicate things? Also node might be the default choice to create a web app but they might be allowed to use other technologies they feel more comfortable with.

Maybe I should go to a hackathon. I'm sure the kids would love to work with my 47-year-old ass.

I think either docker or vmware images would work fine on various OSes available.

Also, if they don't, you can actually build really basic docker images in just a few minutes. Seems like a good way to do it.

Seems like pretty much every job I've had for at least a decade has used virtual machines for dev, even if the machines the devs use could theoretically support the dev infrastructure natively. It's just a really easy way to make sure that every dev has an environment that matches production.

I've never had a job that used VMs for dev.

Go to a hackathon. Getting a street level view of anything can only help increase understanding, and it is especially good at making learning curves go way faster.

Yeah I will.

Python + scikit-learn (and thus numpy/scipy). The big advantage of using Python is that you don't just get good ML libraries but also good "glue code".

If you just want to run some quick statistic stuff on a dataset R and RStudio are great.

My experience with Mongo (only a loal instance for prototyping nothing fancy with multiple DBs spread over multiple machines):
- Very easy to set up and get going
- Easy to interact with programatically (from Python in my case)
- I had to look up all the query stuff repeatedly
- Pretty easy to use overall if your data naturally fits into JSON
- I saw performance issues already (didn't optimize much but indexing didn't help a lot) when dealing with a DB of ~15 million tweets with metadata and running simple searches on that