Machine Learning Basics with ml5js Video Lecture | Machine Learning with Java - AI & ML

hello explorers and welcome to another
video today I'm gonna go through some
machine learning basics and we are going
to talk about a site called ml5js.org
and in these machine learning basics we
need to first off install node and that
can we be found at nodejs.org
and at nodejs.org
you have two versions you have the
long-time support version LTS and you
have the latest version if you download
either of those it's fine we don't use
any extreme features only new features
of the node.js servers so you can use
either or and just download installs you
are ready
you also need git so you can actually
fetch the material you can get the
source from a zip at github but if you
don't want to do that you will clone the
repository down to disk you will enter
that directory and run npm install so
you will install all the
dependencies so to get this thing
running
we need a node server and the node
server is needed because we need to have
cors and cors is a security measure
that makes it impossible for you to
fetch sources from other servers that
don't that require that they come from
the same source so it's a security
measure that you can't fetch things from
different servers if they are not game
so we need to have a server that can
handle cors for us we can fetch the
images and also fetch the other
resources and do that in a good manner
so the online resource at ml5js
actually works with our examples
otherwise they can't actually fetch our
images and predict them so to set up
this this is the code it's pretty simple
it's an express server with cors I'm
not gonna go into it fully but we will
share this
the resources or the source on line so
we have some set up so we can fetch
images and sources so let's continue
with starting this server up and then we
will do some prediction with the API
implementation from ml5js and this
API is very much evolving from
the time that I actually made this
slideshow until today the actual example
code changed and I actually report an
error for to them because their
documentation was out of sync and that
from that time they had changed the API
as well see I had to update these slides
so you get something more current so if
you want be will be really sure that you
get something working you copy paste
code from their site to be sure that it
actually works because they are heavily
working on it it's an API I believe it's
version 0.0.8
by the time I'm filming this but that
might change pretty rapidly because they
are still in active development and so
to get the source you set up a script
tag in your HTML so you get the
unpackaged and unpackage is actually a
service that takes things that is hosted
on github and makes them available as
packages so you get ml 5 from unpackaged
and then we will set up an image tag and
i set one image as the starter image so I
have a directory of images that is
believe 12,500 images of cats so I will
load one of those and then on load I
will run my onload image ready and
that will
do my action prediction and then I will
get the result and I will put the result
back into these spans here so in the
results span I will tell what we actually
think it is in the image and my
confidence probability is how likely it
is that I actually predict it right and
we need to set up a classifier in on our
js this is the image classifier using
the mobile net and mobile net is a very
common network that is trained on a
large set of images many people use
mobile net and then take off the last
layer to make their own classifier and
so on but it's a very versatile network
with a lot of images and can predict a
lot of things and they also have a
squeeze net that is also quite good at
predicting things but to start up we
will try this mobile net and on image
ready we will fetch the image we will
push it into our classifier and the
predict function it will get a result
for us and the results is actually a set
of images so it's not just one image I
will fetch the first result there and
take the class name and put that into
the result inner text and the
probability with up to four numbers I
will put into the probability inner text
and this will take the best prediction
because they are sorted and then I have
this little function that I run on
stored up to actually request new image
every five seconds
so we see the last time out there five
seconds it will load this function again
it will fetch the image tag get a random
number between zero and
12500 and then it will
load that image in so as I have
mentioned multiple time ml5js.org
is the place if
you want to go into this
see their resources and code and try
this out yourself or you can try my code
out as well
so let's look at my example here and so
this is my classifier and this is
actually taken from their specific code
but I have added this reloading function
so I get a new cat every 5 seconds and
here we see that it figure out it was an
Egyptian cat Egyptian cat it's a lot of
Egyptian cats but I are quite confident
that it is an Egyptian Cat and that's
a Siamese and I couldn't really figure out
that that was a Siamese so it did a pretty
good job there more Siameses guess
it's just saying that because they are
white and that's a dog I believe a
Persian cat yeah I guess it could be a
Persian cat and a tabby tabby cat so
it's pretty interesting that you can
have as you see here as the image is
loaded you get the prediction so it's
very fast at actually predicting things
which is quite interesting and we can
also look at this resource here you see
that you have getting started guide on
this ml5 you have experiments and here
you can look into how to do different
things and what kind of things you can
do with machine learning and this
resource is updating quite rapidly as I
said my example code was actually broken
because they had updated the API so they
are working very hard to make this a
place where you can get up and running
with machine learning very easily and
you can get some foothold in this
technique and actually know a lot about
machine learning so I think it's a
really good resource if you want to try
this out and it's a again some
University here that has put this up so
it's a project maintained by
New York University I believe and
today we'll work more on this if you
like this resource or like this video or
have any questions or any comments leave
them down below if you like this video
give it a like share it with your
friends and colleagues if you haven't
subscribed yet please do that and I
really hope to see you in the next video