Please talk me out of this $99 machine learning board.

Thread Starter

Raymond Genovese

Joined Mar 5, 2016
1,658
I have been reading (and watching) a lot about this new jetson nano board today https://developer.nvidia.com/buy-jetson?product=jetson_nano&location=US I have to admit that it does seem to have some capabilities that (along with supporting devices like cameras) that you really could use to develop some kind of machine learning project...I am reading too much "cloud" though. I'm conflicted - On the one hand, I need this like I need a
Paronychia Correction Tool but on the other, it does seem like some fun.

Does anyone have one?
 
Last edited:

nsaspook

Joined Aug 27, 2009
7,351
I have been reading (and watching) a lot about this new jetson nano board today https://developer.nvidia.com/buy-jetson?product=jetson_nano&location=US I have to admit that it does seem to have some capabilities that (along with supporting devices like cameras) that you really could use to develop some kind of machine learning project...I am reading too much "cloud" though. I'm conflicted - On the one hand, I need this like I need a
Paronychia Correction Tool but on the other, it does seem like some fun.

Does anyone have one?
That looks wonderful if you're already into Linux based systems and machine learning. Getting to the point of doing something useful past the canned demos is real work.
https://github.com/dusty-nv/jetson-inference
 

Thread Starter

Raymond Genovese

Joined Mar 5, 2016
1,658
Well, does using Linux (actually more Free-BSD than Linux) 153 years ago count? Actually the Galileo had some kind of naked version of Linux. In any event, Linux does not bother me but I might have to learn python and I just read this article https://www.allaboutcircuits.com/technical-articles/Python-vs-C-C-which-language-electrical-engineers-why-learn-Python/ and it said that I would then be able to "impress the tech hipsters you talk to at Starbucks" and I don't think that I could handle that.

On the other hand, the video below is pretty impressive (although it would have been more impressive if eclipse would have worked).


Plus it is only $99 (less a bunch of stuff you would want) - what is that? Five lattes?
 

Aleph(0)

Joined Mar 14, 2015
597
@Raymond Genovese Seriously, I say go for it:)! It's worth way more than $99 just to satisfy ur curiosity! Also Python is vry intuitive, easy to learn language! BTW I say if u don't like _yuppies_ (who does:rolleyes:?) just stay clear of their haunts (like Starbucks) it works for me;)!
 

nsaspook

Joined Aug 27, 2009
7,351
Well, does using Linux (actually more Free-BSD than Linux) 153 years ago count? Actually the Galileo had some kind of naked version of Linux. In any event, Linux does not bother me but I might have to learn python and I just read this article https://www.allaboutcircuits.com/technical-articles/Python-vs-C-C-which-language-electrical-engineers-why-learn-Python/ and it said that I would then be able to "impress the tech hipsters you talk to at Starbucks" and I don't think that I could handle that.

On the other hand, the video below is pretty impressive (although it would have been more impressive if eclipse would have worked).


Plus it is only $99 (less a bunch of stuff you would want) - what is that? Five lattes?
I might just get one for work to see if some of the custom kernel drivers I use for RPi 32-bit based projects can get ported easily to the arm 64-bit source tree.

https://elinux.org/Jetson/Nano/Upstream

It looks like the patches should be in the mainline upstream Linux kernel soon but it's not there yet.
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/arch/arm64/boot/dts/nvidia?h=next-20190416
 

-live wire-

Joined Dec 22, 2017
911
It doesn't even have active cooling! What kind of processor that's serious doesn't? Just do this for slightly more and get more AI capabilities:
 

Mark Hughes

Joined Jun 14, 2016
409
Does anyone have one?
Ooh buddy -- you and I do have some similar tastes. Fortunately for us, they seem to be sold out just about everywhere...and things have gone to backorder.

Here's what I'd recommend while you wait -- as I recall, you've got access to a Raspberry Pi -- and although Mathematica isn't on the default Raspian distribution anymore, I'm pretty sure it can still be downloaded.

(I'm using an older distribution of raspian + "sudo apt-get update && sudo apt-get dist-upgrade")

Mathematica supports the machine learning you want to play with -- but it's free.
https://www.wolfram.com/language/12/machine-learning-for-images/built-in-object-recognition.html?product=mathematica

I can't tell you how well it works yet -- I've been updating my distributions and playing with the NBody simulation first.....it ain't fast, I'll tell you that.
 

Mark Hughes

Joined Jun 14, 2016
409
http://wac.36f4.edgecastcdn.net/0036F4/raspbian/install-wolfram-engine.sh

#!/usr/bin/env bash

TEMPDIR=/tmp/wolfram-engine-install
WOLFRAMENGINEURL=http://wac.36f4.edgecastcdn.net/0036F4/raspbian/wolfram-engine_11.3.0+2018052403_armhf.deb
WOLFRAMSCRIPTURL=http://wac.36f4.edgecastcdn.net/0036F4/raspbian/wolframscript_1.2.0-18_armhf.deb
WOLFRAMENGINEDEB=wolfram-engine_11.3.0+2018052403_armhf.deb
WOLFRAMSCRIPTDEB=wolframscript_1.2.0-18_armhf.deb

DEPENDS="coinor-libipopt1v5 libblas3 libcurl3 libexiv2-14 libexpat1 libfontconfig1 \
libfreetype6 libglib2.0-0 libglu1-mesa libgmime-2.6-0 libgmp10 libgomp1 \
libharfbuzz0b libhunspell-1.4-0 liblzma5 libmad0 libmp3lame0 liboauth0 \
libpcre3 libpixman-1-0 libportaudio2 libraw15 libsndfile1 libsqlite3-0 \
libssl1.0.2 libuuid1 libxmu6 oracle-java8-jdk zlib1g"

SUDO=''
if [ ${EUID} != 0 ]; then
SUDO='sudo'
fi

pushd ${PWD} > /dev/null
rm -fr ${TEMPDIR}
mkdir -p ${TEMPDIR}
cd ${TEMPDIR}

echo -n "Downloading Wolfram Engine... "
wget -q ${WOLFRAMENGINEURL} -O ${WOLFRAMENGINEDEB}
echo "done."

echo -n "Downloading wolframscript... "
wget -q ${WOLFRAMSCRIPTURL} -O ${WOLFRAMSCRIPTDEB}
echo "done."

echo -n "Installing prerequisites... "
${SUDO} apt-get -qq install ${DEPENDS}
echo "done."

echo -n "Installing wolframscript ..."
${SUDO} dpkg -i ${WOLFRAMSCRIPTDEB}
echo "done."

echo -n "Installing Wolfram Engine... "
${SUDO} dpkg -i ${WOLFRAMENGINEDEB}
echo "done."

popd > /dev/null
echo -n "Cleaning up... "
rm -fr ${TEMPDIR}
echo "done."
 

Thread Starter

Raymond Genovese

Joined Mar 5, 2016
1,658
Ooh buddy -- you and I do have some similar tastes. Fortunately for us, they seem to be sold out just about everywhere...and things have gone to backorder.

Here's what I'd recommend while you wait -- as I recall, you've got access to a Raspberry Pi -- and although Mathematica isn't on the default Raspian distribution anymore, I'm pretty sure it can still be downloaded.

(I'm using an older distribution of raspian + "sudo apt-get update && sudo apt-get dist-upgrade")

Mathematica supports the machine learning you want to play with -- but it's free.
https://www.wolfram.com/language/12/machine-learning-for-images/built-in-object-recognition.html?product=mathematica

I can't tell you how well it works yet -- I've been updating my distributions and playing with the NBody simulation first.....it ain't fast, I'll tell you that.
Hi Mark. Yes, they are back-ordered, but only by two weeks (so it is said). Did you see how much faster the board is according to that video I linked? I was hearing 3-10 times faster or more. Think about it...10 times faster! I might be able to control the world before the weekend!!:)
 

Thread Starter

Raymond Genovese

Joined Mar 5, 2016
1,658
Not really, but it is a tarpit. Edge computing is really interesting and you will probably enjoy it, and be frustrated by it, and build cool stuff, and... well... it’s a trap.
That is likely the most accurate information on the “phenomenon” that I will receive today. Thank you.

Learning is a need. In many ways for many people, learning satisfies a need to achieve and a need to express creativity.

Last week I made a little fan controller for a close friend who had repurposed a rabbit hutch as a greenhouse/seed starter kind of thing. I worked on it for about a week (well, you know, there is work and then there is work). It turned out very well. I always learn something from those kinds of projects, and when they turn out well, there is a sense of satisfaction. Those simple projects, however, lack the level of satisfaction that comes from the persistent effort that accompanies a higher level of task difficulty.

I have spent my adult working life studying the output of the nervous system. I can, with a good deal of certainty, state that I know something about neurons – not everything, but something. What is discussed as neural networks has little to do with neurons. I know this, and have said as much, to many people, many times, for many years. Machine learning, deep learning and the like are nothing but elaborations as far as I am concerned.

Back in the early 90s, I was overjoyed to receive this book as a gift (I had specifically asked for it as an alternative to receiving some clothing that I would never wear but that fit the idea of what someone thought I should wear).

20190417_105858_s.jpg

I read all of it and it was very difficult reading for me. I spent a lot of time thinking about it and wondering if this was really possible. The new American Dream, sit back as your computer makes “cha-ching” sounds. I never really stopped thinking about it and continued to have a good deal of interest in the topic (not so much in finance or the “cha-ching” part). I am particularly keen on reinforcement theory (e.g., https://daiwk.github.io/assets/dqn.pdf) in regard to deep learning (including the stark weaknesses of the approaches).

IOW, I can see myself heading in that direction….but, I can say that about a lot of directions. Will I order that board? I don’t know if I will, but I might.

Choosing your tar pit is a freedom. Less so, living with the tar pit you have chosen.
 

Yaakov

Joined Jan 27, 2019
1,612
So you are going to build something that makes spending cycles on bitcoin mining seem stupid.

I want one when you get it done. But please, let me get one before GA, for at least a few months.
 

nsaspook

Joined Aug 27, 2009
7,351
Not really, but it is a tarpit. Edge computing is really interesting and you will probably enjoy it, and be frustrated by it, and build cool stuff, and... well... it’s a trap.
I've been in and out of the tar-pit a few times starting with 'Expert Systems'.


A old 80's era book from my tech library.

It never fails to amaze and disappoint at the same time.
 

Yaakov

Joined Jan 27, 2019
1,612
Fuzzy logic got a bad rap, largely because of its name. The originators regretted that decision.

The Japanese embraced it for many things, but apparently it just seemed too risky to "decision makers" to be applied in industrial setting where it could have done a lot of good.

Fuzzy logic isn't really fuzzy at all, it gives definite and functional answers to the questions posed by the inputs to the system.
 

Thread Starter

Raymond Genovese

Joined Mar 5, 2016
1,658
So you are going to build something that makes spending cycles on bitcoin mining seem stupid.

I want one when you get it done. But please, let me get one before GA, for at least a few months.
:) N/P you are on the list and let's hope that for both of us, that means the 144K or, at the least, the great crowd. :)
 

nsaspook

Joined Aug 27, 2009
7,351
Top