Numenta 2014 Spring NuPIC Hackathon Demos, 7 May 2014

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: NuPIC in the Cloud: 2014 Spring NuPIC Hackathon Demo

Description

Austin Marshall. Source code: https://github.com/oxtopus/fee-fi-fo-fun. Wiki: https://github.com/oxtopus/fee-fi-fo-fun/wiki

A

Okay, can we all see that uh so in the interest of time uh I went ahead and did everything in advance and I'm just going to show you some screenshots. So hopefully this would go quickly, and so what I wanted to focus on um was finding an easy way to deploy a new pick in a readily available cloud like aws.

A

So I know at these hackathons a lot of people will run into some build issues and was there anyone here who tried and just couldn't get new pick up and running awesome, but for those of you who are watching the video, if you couldn't, this might be an option for you.

A

What it is effectively is a point-and-click solution, so amazon has a service called elastic beanstalk, it's sort of their answer to heroku. If you're familiar with that service, it's a it's an environment in which you can deploy an application. It wraps a lot of their other services, and so they've got this nice little wizard that you can walk through and this demo fun inspired by jack and the beanstalk.

A

And so what I've done is I've created just a simple demo application and run through the process of deploying it on elastic beanstalk using a product called docker, and so you just go through and you enter the application name. You select docker as the target environment and what it requires is a an application package in this case. It's looking for a zip file that has a docker file in it that describes your application and.

A

I'll show you what it looks like here: real quick.

A

If you're on github, you can go straight to the application hit, download, zip and wherever you download that too, you can just upload it directly to that interface. So you don't need git, you don't need new pic.

A

You just go straight from a github page to a running application.

A

There's some more, you know just basic boilerplate stuff. You have to go through your application and when you're ready to go, you just hit launch and it takes five to ten minutes and what it does. It launches an instance in the ec2 cloud that aws or amazon hosts and runs the docker runtime, and it reads the docker file. That's in it that sets up the entire new pic environment.

A

It does a complete, build from scratch and installs all the dependencies and runs a single command that I've uh and I'll get into some details in a second and in the end, you have just a running instance that you can log into and interact with.

A

So the docker file looks like this new pick itself. Has a base docker image that it extends from, so that that does all of the.

A

Both the environments set up.

A

So you has to go through the process of you know: installing ceiling, cmake, setting up the python environment, building new pic and you know building it and um in the end it gives you a working directory that that has a new pic installed. So the application just goes in and says: okay, add my current directory to it. Expose this port. I'm going to run this initial thing, because I'm loading in some data and then it runs my application, and so in this particular case it takes the jack and the beanstalk story from project gutenberg.

A

It tokenizes it and uses the category encoder to just do some simple prediction of words that come in in order. So it's not using like the step dp or anything. It's just doing simple categories.

A

So I'll show you what it looks like once.

A

I'll attempt the demo gods here.

A

So there's a simple command: you just do. Sudo docker, ps,.

A

A

Process and I'll show you the output from the the command that I had it run.

A

Docker attach there's the image that it's running, and so what we have here is the, for example, the words and then the probability distribution of the other words that are predicted in the next step, along with their probabilities for every word. In order, as it appears in the story.

A

And that's it yeah.

B

So if you have, if you want to host new pick on amazon.

A

Yeah really easy.

B

A

Pay for it by the minute, by the by the hour, yeah.

B

Because it's very.

A

Cool, um so you don't really need any information. Anything on your computer, except for a browser and an aws account.

B

I know some people who run nupik on amazon, yeah yeah, that.

C

Is really really cool awesome? I I just just trying to make sure I understand it correctly. So when it's running on amazon, do I interact it only through this file or do I or do I? Can I programmatically.

A

C

I would other my local installation so.

A

The way docker is meant to be run. um You know you set up your environment, you run some service and you expose those services in your docker file, and so that's normally going to be like a web server or you know some other service. So in this particular case, I just needed a simple application that just went through the motions of interacting with new pick and all it's doing is spitting some stuff on stuff.

C

Yes, I'm saying: is there any restrictions on you running it this way or is I mean, could we take uh uh cheyton and I'm sorry, the other gentleman's name I forgot maybe chatting to jeff? I think it was. Could you take the cerebral two and run out on this? Is.

A

C

A

I'm sure it's possible, but there's there's going to be some latency and some additional work that you have to go.

C

Through but in general, it's not like a restricted environment when you're running.

A

No, it's a sandboxed linux environment.

A

Yeah yeah I mean you, could you could set it up so that you take in some additional params? Then you get back some optimize. You know the results of the storm then use that to in your own environment,.

B

Very cool thanks, austin.