Installing CPU and GPU TensorFlow on Windows Video Lecture | Machine Learning with Python - AI & ML

what is going on everybody welcome to a
quick update I have heard the good word
that tensorflow is now supported on
Windows so if you go to test flow org
you go to get started you come down to
pip installation there is now a special
section for the Windows users because we
are pretty special so um so here's the
information it should just be a simple
pip install and this will get us just
that straight I'm curious if we could
just pip install tensorflow
I'm pretty sure we might even get away
with that but let's let's just see let's
see what happen so pull up here pip
install tensorflow
well see if that worked
hopefully we get the 64-bit version I
can't see anything that's happening
right now but we'll go back and look
it's replacing numpy for me and all that
I'm just hoping that we can get by with
just simply running pip install
tensorflow rather than having to do like
this would be the RO a real specific
version but we'll see if pip works first
oh I can't scroll up to see all of the
things ok fantastic so I'm going to go
ahead and pull up here and in fact
actually we could just do it straight up
in here
so actually python normally will lead to
for me we'll go to python 2 so i'm going
to type the full path to python now and
first let's import tensorflow as TF ah i
know we're going to get screwed on so
this looks like it must be the GPU
version of tensorflow
I mean I'm sorry the CPU version of
tensorflow now looks like it's probably
actually going to work anyway so that
would be for the CPU version which I'm
sure a lot of people might only have now
I want the GPU version because I have a
GPU that will
we'll work nicely so pip install - -
upgrade that's probably going to throw I
might have to uninstall it first but
let's go for the GPU version so I'm
going to copy that come back over here I
close the command window reopen that
paste GPU cool hopefully that I'll just
override it the reason why I knew is
when you do the GPU version it'll like
it'll have to import a bunch of GPU
stuff although the other thing I'm
curious about is I don't have like the
CUDA toolkit and all that other stuff on
here that I know of yet you can now test
your installation surely there's more to
the story for installing the GPU version
I don't know if anybody followed the GPU
installation optional Linux because you
got a you know install the cool toolkit
and all that stuff I thought I could
just hit down and it would not have this
here so C colon slash PI 35 yeah so we
still have two if you want the GPU
version you're still going to have to
get the CUDA toolkit and all that stuff
yeah but that's so you can at least
install the GPU version or the CPU
version now with the GPU version we have
to get all that other stuff now I'm
curious if all you wanted was the CPU
version then you're really good to go
you could probably leave but if you want
the GPU version I'm gonna go back to a
my own tutorial for these steps for the
GPU version I know you need the KU DNN
and the tool kit and that was kind of a
pain in the butt to do at least on Linux
but let's see so we got that to scroll
on down or done this partitioning stuff
basically all I want let's see
any of that here we go so really we need
this and this is just a folder so let's
see if we can do and this is koutou
eight could eight is actually quite a
bit faster
and I forget if like Huda eight is
supposedly like 50% or even a hundred
faster so that'll be kind of interesting
if that's good if this will work for us
I want to say the current version of
tensorflow really wants a 7.5 so we
might have to make a change trying to
see if I can find that in here again if
you don't need up nope nope they're
supporting 8.0 now cool so yeah we're
all caught up if you're watching this
too and you're currently using the cuda
7.5 eye hi Lisa highly recommend and
suggest that you you you upgrade to CUDA
8.0 really a huge improvement so anyway
yeah so Windows I'm on 10 obviously you
would pick some other version if you're
on a different version to the network or
a local I tend to prefer the local
installer just in case I need to do this
again sometime
[Music]
okay so I'll download that it's going to
take a few minutes while we wait there I
think I should be able to grab this -
coup d NN and this is upgraded to great
I forget what I even used for this
first try I don't know if you're someone
that uses like multiple different
passwords for everything that's a that's
a big win right there anyway I've got at
least like 15 password let's see oh man
I forget which from let's see if they
specify to this one which which version
we want to DNN they want v5 so in
previous in the previous GPU version we
wanted for in CUDA toolkit 7.5 now it's
8 and version 5 so we're going to
upgrade we're going to see some huge
speed improvements here I'm excited to
try out these new speed improvements I
haven't upgraded my other versions so
kudi then
boot specifically does say 5 I'm going
to go ahead and not risk it we're going
to go truly with B 5 there is five one I
would imagine that would probably still
work I think we're going to Yolo and go
for five actually and yeah alright that
CUDA um
the other downloads slow now because of
that probably hopefully that'll speed up
so I think I'll probably pause this
while we wait and I don't know what the
CUDA toolkits going to do to me when I
install I don't know if it's going to
force a restart on me but we'll see we
will see and I forget what we're
supposed to do with coot I'm pretty sure
with Kuti and then all we really do is
move it to the the directory of our
kunuk toolkit but we'll see we'll see
so I'm going to go ahead and pause this
uh for now and I will pick back up when
the download has completed so I will see
you in about two minutes or something
all right we are back I've got my
downloads I'm going to assume this needs
to be run as administrator' but I'm
gonna go ahead and run it without
administrator just see it might ask me
in a moment when this opens in ten years
sorry it's in this I've just moved both
downloads into this directory here so
right now I'm just installing the
toolkit and then we'll deal with kuden
ends in a moment so yeah see we really
want it to be in oh this is a temporary
like what would we want this to be in
temporary okay so it's just extracting
it I guess to temporary so first time
I've ever install anything on Windows
wait for it I would pause but I'm pretty
sure everybody's going through this wait
different people have different download
speeds though so that's why I didn't I
pause for that so at least on Linux what
happens is once you download and install
this you'll go to qu DNN you extract it
it's literally just like a copy and
paste operation you're just like moving
these files to the CUDA toolkit location
so that's really easy but I will have to
say unless they've changed their
download and install instructions that
was actually not clear when I was going
through it I could not figure out what
the heck I was supposed to do with the
kudiye and stuff so that was kind of
annoying at least when I was going
through it maybe I was reading too fast
and not paying enough attention or
something so it is installing I'll bring
it over here and before this fails this
stuff is just crazy I remember like
years ago trying so hard to get
computing on the GPU to do anything with
CUDA it's just such a little sensitive
system
luckily tensorflow has made it really
easy lately it
but we'll see there's always just
hang-ups in my experience with this
stuff no apparently checking
compatibility is taking four years there
we go green continued I will also say
this is probably so much easier than
Duke on Linux the whole setup process on
Linux was a panda but snip snap please
don't tell me you just restarted on me
haha I don't know what's happening
everything's coming in and out I'm gonna
lose this recording I'm calling it
everything I've done so far is going to
be lost probably fact I'm gonna save
this recording up to this point because
I'm pretty like whenever I install this
stuff sometimes a restart happens to you
so alright and we're back I was
definitely seeing the future because it
well it didn't restart my computer but
instead it crashed my recording it also
for whatever reason changed my audio
device on me that's cool I'm not sure
why it's but alright it actually made my
monitor the audio device um that's a
first
anyway I'm sure everything works oh cool
the installation is done we'll hit next
I don't need to see any of that
documentation let's be serious
okay so we have kudia and installed I
have absolutely no idea where it went I
guess I'll go hunt for it let's see
probably Program Files oh there it is ok
let's see so for me I wouldn't yet this
see Program Files I was looking in
Program Files x86 so just straight up
program files and video GPU computing -
okay it's probably we're live for then
you got CUDA V 8.0 and so on so now what
I'm going to do is find where our of
course I got rid of it and oh my
goodness where did it all go
my original windows TF testing
cool so so this is is where CUDA was
installed at least on my machine that
was by default see Program Files NVIDIA
GPU and all that now what we do is we're
going to extract this right to our same
directory and then this is when it's you
okay so this is that cuda director you
click on that you got been including lib
that should match here
there's your been included and lib and
so what we're hoping to do is to just
merge these two together let's just drag
it in here just move to 8.0 we need
permission continue continue continue
Wow okay do this for all the concurrent
items please thank you I was really
hoping to see you were going to tell me
that it was a merge but yeah that's
Maura was in there I thought that would
be like a warning that like this already
existed or something but no okay I guess
we made the copy we're gonna find out I
suppose
alright so let's uh minimize this this
in fact may open up now a show command
window
I forget okay thank you any Python still
yeah Python is four to seven on my
machine so I'm gonna see : python 35
Python import a tensorflow
I see mostly successful so you get it
all successful successful oh and we win
that's really exciting I'm really happy
to see that really happy to see that now
let's go ahead and test it I'm pumped
I'm pumped
all right let's um I can't remember I
think this uses I forget whose data this
uses you guys want hoodies let me know
oh this explains them let's see we'll go
to this one hopefully about some sample
code here Harrison this was the previous
one and I hear we just changed it I
guess we could just go ahead and run
this one I really just want to see it
run that's all ah this is going to throw
us for a loop - okay paste there what in
the heck did I have some sort of uh-oh I
did I just want just that code um we'll
just read it in wherever we are I guess
I wonder if I just leave this blank if
it'll just be local let's see what
happens
we'll have to extract the data for it up
okay fine maybe this access is denied we
really ought to there we go
okay I didn't see anything happen here
this just is a deprecation warning not
quite sure why why we got kicked out
there let's go to the command window and
stay let's do a seat : / those testing
that PI weird so I just doesn't want to
work in oppression I've done this it it
doesn't want to work in very fast very
good for some reason it doesn't want to
work in idle okay but it works in the
command line still gives us that
deprecation warning but at least it
actually still runs that that's how a
deprecation learning ought to be that's
always like why did it stop cool so yeah
this appears to be running pretty at the
pace that it ought to be I don't know if
there's an SMI equivalent to let's see
if I have I should have GP I think I
have gpu-z installed all right I quickly
just went to grab gpu-z just to see what
we're doing
so what I'll do is I'll put this down
here and then here you see the GPU load
currently 5% will run this hopefully
we'll see a jump up it shouldn't have to
jump up - oh it's actually that I use
quite a bit there to initialize that's
weird so currently we're sitting about
50% GPU load which is probably to be
expected this isn't heute the biggest
network cool so it's definitely using
the GPU that's exciting
cool so if anybody has any questions
comments concerns whatever if you've if
you're unable to install the GPU version
or really the CPU version of tensorflow
let me know one thing I will just say is
in the Windows version they do make it
clear that if you get this like missing
thing here if you see this error no
module named PI wrap tensorflow blah
blah blah blah or DLL failed
uh you might be missing this it might
not be in your path but also you're
probably also is if this is happening to
you you probably just need to download
C++ 2015 redistributable click on that
come down here English download install
that otherwise if that's not your
problem let me know and I'll do my best
to help you out and I will see you guys
in another tutorial