Machine Learning on Raspberry Pi, With Python and Intel Movidius Neural Compute Stick

Been digging deep into machine learning since cracking on with home assistant. Personally AI driven devops and manufacturing is the next wave of tech but that’s just me.

But overall I know a lot of you out there are wanting to learn the basics or just get one’s feet wet.

So here you go:

And for @Team_Civic_Hacking

Pothole dataset

https://www.researchgate.net/post/Is_there_any_dataset_of_potholes_on_roads_available

Below is what it would take to build one’s own setup that would fit in their closet:

Initial Hardware

Item Price
Raspberry Pi 3+ ( 3x ) $36.97 each
Intel Neural Compute Stick 2 ( 3x ) $68.99 each
MikroTIK Router (1x) $38.95 each
SD Card preloaded with os (3x) $19.99 each
Cat 6 (6pk 1 unit) $9.90
Sub total $387.75

So for less than the price of the latest iPhone or even two weeks of a single cloud vm one can have a three node machine learning cluster.

Of course that also gives one enough funds to do full immersion liquid cooling:

Plus if one wants to see the truly sexy stuff… visit hackaday or read the code:

https://hackaday.com/tag/movidius

1 Like

Could something like this be setup at DMS for group learning events/projects?

Already has. Look in the VCC area. Its been there for almost a year waiting to be used by members.

Plus if one jumps on communitygrid.dallasmakerspace.org they get full access via a browser.

image

hmm ok… that cannot keep happening, was working last Friday and nothing was changed by myself. Hell that same stack runs enterprise sites that need 110% uptime.

https://monitoring.dallasmakerspace.org/sources/0/hosts

That’s working!

http://traefik.dallasmakerspace.org/dashboard/

The load balance is up and guess what; some host names where f’d with and someone deleted the ui container.

Easy fix.

1 Like

So just on a whim I googled “GAN on Movidious” and found this:

So much to learn for a “relatively easy” project.
Long Sigh…

Yeah I know what you mean. Data science is way out of my headspace. I’m still struggling how that PEEK and POKE actually translates to an electrical signal send across this magical board and rats nest of copper cables. (might have something to do with smurfs and a jinni, maybe some gophers too)

My end goal with learning this is able to do text processing for analytical models (ie correlated events) and predictive behaviors based on past metrics.

Or in more simple terms; teaching a machine that’s currently about as smart as a Ford Pinto and has about the same maintenance overhead to be able to adjust its own carburetor while driving down the highway at 99 miles an hour.

4 posts were split to a new topic: Pair of car guys remembering about missed loves… cars

Instead of buying hardware that doesn’t get utilization all the time.
Why not just run Microsoft Azure Cognitive Services on the free tier.
GCloud ML (Machine learning)
You get like a million request for free a month

It’s stupid easy, Has a highly trained data set, way more polished plus free

It’s like 95% sure you’re in human form. Go Dwight go!

2 Likes

It’s like 95% sure you’re in human form.

Lol… you sure?!? I could just be a borg drone in disguise