Windows support

[aebk2015]

I just learned about TensorFlow and I want to install on Windows 7, 64. I have both Python 2.7 and 3.5 on my PC.

Thanks,

[filR]

Did you see the Download and Setup section on the TensorFlow website?

It should be a normal pip setup for Windows. Though of course there is always Docker for a super fast setup.

[urish]

You can run it inside docker (that is what worked for me). Here are the steps:

1. Download and install docker toolbox for windows
2. Run the folowing command to create a docker virtual machine:
   • docker-machine create vdocker -d virtualbox
3. In a cmd window type the following:
   • FOR /f "tokens=*" %i IN ('docker-machine env --shell cmd vdocker') DO %i
   • docker run -it b.gcr.io/tensorflow/tensorflow

After a short while, you will get a linux shell. Type python and have fun :)

[aebk2015]

I rather use PIP but when I try :

pip install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.5.0-cp27-none-linux_x86_64.whl

it gives me an error:

tensorflow-0.5.0-cp27-none-linux_x86_64.whl is not a supported wheel on this pla
tform.

On Mon, Nov 9, 2015 at 2:24 PM, Phil Reither notifications@github.com
wrote:

Did you see the Download and Setup
http://www.tensorflow.org/get_started/os_setup.md section on the
TensorFlow website?

It should be a normal pip setup for Windows
https://stackoverflow.com/questions/4750806/how-to-install-pip-on-windows.
Though of course there is always Docker for a super fast setup
http://www.tensorflow.org/get_started/os_setup.md#docker-based_installation
.

—
Reply to this email directly or view it on GitHub
#42 (comment)
.

[aebk2015]

It gives me an error:

An error occurred trying to connect: Post
http://localhost:2375/v1.21/containers
/create: dial tcp 127.0.0.1:2375: ConnectEx tcp: No connection could be
made bec
ause the target machine actively refused it.

On Mon, Nov 9, 2015 at 2:24 PM, Uri Shaked notifications@github.com wrote:

You can run it inside docker (that is what worked for me). Here are the
steps:

1. Download and install docker toolbox for windows
   https://www.docker.com/docker-toolbox
2. Open a cmd window, and type: docker run -it
   b.gcr.io/tensorflow/tensorflow
3. After a short while, you will get a linux shell. Type python and
   have fun :)

—
Reply to this email directly or view it on GitHub
#42 (comment)
.

[urish]

@aebk2015 is that with the docker? I forgot another step, you also need to create a docker vm after installing the docker toolbox, the commands are:

• docker-machine create vdocker -d virtualbox
• FOR /f "tokens=*" %i IN ('docker-machine env --shell cmd vdocker') DO %i

[aebk2015]

Uri, I tried installing docker toolbox following your previous commands
but still it gives me the same error:

'docker-machine' is not recognized as an internal or external command,
operable program or batch file.

On Mon, Nov 9, 2015 at 2:48 PM, Uri Shaked notifications@github.com wrote:

@aebk2015 https://github.com/aebk2015 is that with the docker? I forgot
another step, you also need to create a docker vm after installing the
docker toolbox, the commands are:

• docker-machine create vdocker -d virtualbox
• FOR /f "tokens=*" %i IN ('docker-machine env --shell cmd vdocker')
  DO %i

—
Reply to this email directly or view it on GitHub
#42 (comment)
.

[den-run-ai]

conda binary package would be nice (in addition to wheels) on windows, and build instructions with msvc or mingw compilers

[vrv]

Merging this discussion with #17 to keep all the good info in one place.

[jgharris7]

I installed docker and it fails with an error message when I try a quick start

[Wizardofoddz]

@jgharris7 run as admin

[netinstructions]

I put up some instructions / screenshots of running TensorFlow in a Docker Machine using the commands above if it's helpful for anyone. I was able to get it running on my Windows machine.

[Robot-HaoL]

where is the version tensorflow that supporting windows ? thank you

[arose13]

TensorFlow needs a windows compatible version please!

[David-OConnor]

Looking forward to a Win version or Conda package; getting Docker working's turning out to be a PIA; errors about bios settings, hosts already existing, vdocker not being running, Missing opening '(' after keyword 'for' etc.

[lrn2fly]

I followed all official docs on my win machine and managed to painfully install docker and then to download docker... uhm container(?) inside oracle VM VirtualBox

[IAMJackal]

@urish Thanks for your docker install! Total newb here :(

Ran the following command to create a docker virtual machine:
docker-machine create vdocker -d virtualbox

In a cmd window typed the following:

FOR /f "tokens=*" %i IN ('docker-machine env --shell cmd vdocker') DO %i
docker run -it b.gcr.io/tensorflow/tensorflow

The "vdocker" VM loaded and started, but never got the linux shell. Maybe didn't wait long enough or can it be launched from VM?

Also, any tips for setting up the files from https://github.com/anishathalye/neural-style Tensorflow/Python implementation so as to run on vdocker via Python commands or better yet ipython notebook commands? Thanks again for the docker install!

[nogi-ogi]

I got far as getting to the python in cmd but for some weird reason if I try to do the same thing in powershell, I get a TLS connection error saying host is not running or the connection times out.

Is this due to error in my powershell configuration? I'm not too familiar with powershell but I prefer to work from there as opposed to the default cmd. Any ideas on how to circumvent this issue?

[IAMJackal]

Got TF to work in an elementary OS (ubuntu) virtual machine in Windows 10, gave up on running TF in docker.

[altaetran]

Does anyone know if it is possible to use Conda installations simultaneously with tensorflow running through Docker? I have packages I need to use that can only be installed through Conda. Thanks!

[iaroslav-ai]

For prototyping I would suggest stick to vritualbox solution.
I installed tensorflow and theano in virtualbox ubuntu with windows host, and made some comparison with purely windows intallation of theano. I found little to no difference in speed between virtual box and pure winodws theano versions; Furthermore, I found that tensorflow in virtualbox is even slightly better (~25% faster) for cnns compared to windows theano in single threaded mode. Furthermore, both windows and vm theano has some threading problems (runs in 1 thread) which I did not figure out yet, but which makes tensorflow 2.5 times faster with cnns overall on my system compared to theano for cpu computations.

[IAMJackal]

@iaroslav-ai , thanks for the update!

[HWiese1980]

As long as you run it on the CPU that's correct. They should be pretty much equally fast. However, the major intention to get Tensorflow natively running on Windows is to be able to use the GPU. That's not possible from within a virtual machine. At least not when using Windows as the host.

[iaroslav-ai]

@SeveQ, yes, you are correct. For the case of host other than Windows, do you mean PCI paththrough? Would be really cool to have it on Windows. For those who would like to dig further into virtualbox direction, there is rCUDA, which does virtualization of GPUs, and is available for Windows. But then it might be easier as of now if GPU usage is absolutely necessary e.g. use ubuntu on usb drive.

[HWiese1980]

@iaroslav-ai Yepp, exactly. More precisely PCI-E passthrough which afaik isn't possible in Windows. And I agree, best way to go for the time being is to use Ubuntu. Be it on a USB stick or a second partition... However, as I mentioned in another issue here, dual booting is a pita. Especially if you're a hobbyist and want to do hobby-hopping between gaming and Tensowflowing every now and then. For this case - and I want to believe for others too - it'd be really awesome to have Tensorflow working natively and with GPU support on Windows.

[zkailinzhang]

C:\Users\Desktop>docker run -it b.gcr.io/tensorflow/tensorflow

Unable to find image 'b.gcr.io/tensorflow/tensorflow:latest' locally

docker: Error response from daemon: unable to ping registry endpoint https://b.g
cr.io/v0/
v2 ping attempt failed with error: Get https://b.gcr.io/v2/: dial tcp 64.233.188
.82:443: i/o timeout

[sarahharun]

The Linux shell to write python code or jupyter notebook in browser, nothing pops out. I tried to run the the jupyter manually on browser, didn't work. After the commands I'm attaching below, it doesn't do anything. I am using Windows 10.

[mrry]

Today we announced native Windows support, with packages for Python 3.5. If you have installed the 64-bit version of Python 3.5 (either from Python.org or Anaconda), you can install TensorFlow with a single command:

C:\> pip install tensorflow

For GPU support, if you have CUDA 8.0 installed, you can install the following package instead:

C:\> pip install tensorflow-gpu

Please open a new issue if you have problems with these packages, and thanks for your interest in TensorFlow on Windows!

[offchan42]

Finally! The time has come. So I don't have to use Docker without utilizing GPUs anymore.

[JerryKurata]

Thank you!!! great news.

[Techlucy]

Wooh,Great update.. and Last Saturday lost my windows partition files trying to install Ubuntu on Windows machine to use Tensorflow... Thanks to Testdisk software helped me to retrieve some... Now let us try this Tensorflow on windows......!!!!

[HWiese1980]

Holy mother of god! I didn't sleep very well last night but this is great news to untimely wake up to. Installing CUDA 8 right now (had to reinstall my Windows a couple of weeks ago and didn't bother yet to also install CUDA)!

[HWiese1980]

Oh, and just to inform you: with the corresponding wheels Keras now works with Tensorflow backend on Windows as well. I've just tried it: http://i.imgur.com/uO0Z1zz.png

You need the wheels of numpy-1.11.2+mkl+cp35 and scipy‑0.18.1‑cp35 for your system (32 or 64 bits) from there: http://www.lfd.uci.edu/~gohlke/pythonlibs/ and you need to configure your Keras appropriately to use Tensorflow as backend. That should do the job.

[Techlucy]

@HWiese1980 May I write you privately? Thanks!!!

[HWiese1980]

@Kamatio Sure!

[HWiese1980]

I have to mention that some things regarding Keras still don't seem to work. I can run many of the examples shipping with the Keras git repo but some just respond with several different error messages...

But this repo here is about Tensorflow, so... just wanted to mention it.

[Techlucy]

Can someone advise me how to run these function imports on recent released tensorflow on windows?
import audioTrainTest as aT
import audioBasicIO
import utilities
Have tried all ways to correct the errors but it failed. I am using Anaconda, Python 3.5 ....
Thanks!! @ry

[Pdash]

Hi,

I am trying to install Tensorflow in windows machine through Docker. I have also installed iPython. (See below image for details).

If i am running the command "docker-machine ls", it is giving me below result in docker quick start terminal. I had already created a virtual machine "vdocker".

When i am trying to run below code in cmd.exe, i am getting this.
"FOR /f "tokens=*" %i IN ('docker-machine env --shell cmd vdocker') DO %i
docker run -it b.gcr.io/tensorflow/tensorflow"

Also python window is not opening even if i am trying to open through browser.
http://192.168.99.101:8888/

Could you please tell me what is happening?

Thanks in advance !

[Techlucy]

But Tensorflow is now compatible with windows, May be no need to install it via Docker.. Just consult here and install on windows..

[JerryKurata]

I am not familiar with docker, but I agree with Kamatio that running natively is the best option, unless you need the Docker environment.

I run TensorFlow natively on my Windows 10 system with Juypter (was iPython) and it works great. The only glitch is that installing Tensorflow into a conda environment, instead of system-wide, is not yet supported.

[brando90]

it seems that this issue also happens with Docker images. For example, if you try to do

docker run -it --rm gcr.io/tensorflow/tensorflow

it freaks out with the notebook security warnings and doesn't open a bash shell.

[chitro]

Hello I am running Tensorflow examples on my Win 10 machine with Python 3.5.2 and latest TF 0.12.0 (those out of the box) and seeing errors on several of them. Seeing several errors. For example, the wide-N-deep tutorial is throwing the following (along with lots of deprecation messages)

Traceback (most recent call last):
File "C:\projects\tf\tensorflow\tensorflow\examples\learn\wide_n_deep_tutorial.py", line 208, in
tf.app.run()
File "C:\Users\chitro\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\platform\app.py", line 43, in run
sys.exit(main(sys.argv[:1] + flags_passthrough))
File "C:\projects\tf\tensorflow\tensorflow\examples\learn\wide_n_deep_tutorial.py", line 204, in main
train_and_eval()
File "C:\projects\tf\tensorflow\tensorflow\examples\learn\wide_n_deep_tutorial.py", line 197, in train_and_eval
m.fit(input_fn=lambda: input_fn(df_train), steps=FLAGS.train_steps)
File "C:\Users\chitro\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\contrib\learn\python\learn\estimators\dnn_linear_combined.py", line 711, in fit
max_steps=max_steps)
File "C:\Users\chitro\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\util\deprecation.py", line 191, in new_func
return func(*args, **kwargs)
File "C:\Users\chitro\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\contrib\learn\python\learn\estimators\estimator.py", line 355, in fit
max_steps=max_steps)
File "C:\Users\chitro\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\contrib\learn\python\learn\estimators\estimator.py", line 690, in _train_model
features, labels = input_fn()
File "C:\projects\tf\tensorflow\tensorflow\examples\learn\wide_n_deep_tutorial.py", line 197, in
m.fit(input_fn=lambda: input_fn(df_train), steps=FLAGS.train_steps)
File "C:\projects\tf\tensorflow\tensorflow\examples\learn\wide_n_deep_tutorial.py", line 159, in input_fn
for k in CATEGORICAL_COLUMNS}
File "C:\projects\tf\tensorflow\tensorflow\examples\learn\wide_n_deep_tutorial.py", line 159, in
for k in CATEGORICAL_COLUMNS}
TypeError: init() got an unexpected keyword argument 'dense_shape'

Any idea where this could be from? I did a git pull for TF again to ensure I have latest examples.

[sunsided]

Sadly one nasty part is the integration with Google's Cloud Dataflow like in this example. You now have the choice of having TensorFlow (supporting only Python 3.5) or Apache Beam SDK (supporting only Python 2.7), but not both. 🙈

[dsblr]

This below link explain step by step with screenshot to install tensorflow on windows, docker not required.
http://start2deeplearning.blogspot.com

[raphaccgil]

Hello everybody....

I was trying to install tensorflow with pip commands and it was not supported...I am using windows system and python 2.7. The tflearn was possible to install. Do you have any suggestions?

[anujdutt9]

@raphaccgil Hi. Tensorflow is not yet supported on Python 2.7 on Windows System. Instead you can do the following:

1. Install Python 3/3.5 [64-bit]
2. Go to python installation folder using CMD
3. cd to Scripts folder.
4. Use command:
   pip3.5.exe install tensorflow

It will install perfectly.
Check your installation using:
import tensorflow as tf
in IDLE.
Hope this helps :)

[NitinMahajan1]

Tried installing GPU and CPU versions.
Cannot install either and I am getting error as per screenshot above.

Info -
I have Anaconda 3 installed on my system
Python 3.6
Windows 10
i7 16 gigs ram
Intel HD Graphics 520

[anujdutt9]

@NitinMahajan1 Hi. Try to install it as follows:

1. Install Python 3/3.6 [64-bit]
2. Go to python installation folder using CMD
3. cd to Scripts folder.
4. Use command:
   pip3.6.exe install tensorflow

It will install perfectly.
Check your installation using:
import tensorflow as tf
in IDLE.
Hope this helps :)

[ehfo0]

iget this error:
tensorflow_gpu-1.0.0-cp35-cp35m-win_x86_64.whl is not a supported wheel on this platform.
I have a geforce gtx 970 and windows 10 x64 home edition and anaconde python 3.6 is already installed!
what is the problem?

[sunsided]

@ehfo0 It requires Python 3.5

[Coderx7]

@ehfo0 : when you want to download the wheel package, pay attention to the cp3x section.
tensorflow_gpu-1.0.0-cp35-cp35m-win_x86_64.whl
cp35 means it needs python 3.5!

[mrry]

@sunsided and @Coderx7: Thanks for helping out @ehfo0!

This thread is getting a bit long, so I'm going to lock it for new conversation. If you have new questions about Windows support, please try Stack Overflow first, or open a new issue here if there appears to be a bug or you have a new feature request.