3

I am trying to use tensorflow on my college server, as some of the things that I have written are too heavy for my laptop. So I don't have sudo privileges. Below is what I unsuccessfully tried.

I am able to install tensorflow by pip install --user <url> . But when I import tensorflow I get the error glibc 2.17 not found.

I found this link which solved the exact same problem but when I run

virtualenv --system-site-packages ~/tensorflow

I get this error:

ImportError: No module named pkg_resources

For resolving this I tried:

wget https://bitbucket.org/pypa/setuptools/raw/bootstrap/ez_setup.py -O - | python

And then I get following error:

[Errno 13] Permission denied: '/opt/anaconda/lib/python2.7/site-packages/test-easy-install-10463.write-test'

And I don't have sudo access so I can't use sudo with above command.

For installing virtualenv I used:

curl -sL https://raw.githubusercontent.com/brainsik/virtualenv-burrito/master/virtualenv-burrito.sh | $SHELL

I think that the problem is with virtualenv setup but I tried removing it and installing virtualenv by:

pip install --user virtualenv

But this fails and I get:

InsecurePlatformWarning Could not find a version that satisfies the requirement vitualenv (from versions: ) No matching distribution found for vitualenv

I have looked at many (1,2) SO answers dealing with this problem but none work for me. I get stuck at some other error.

I am now stuck in a loop of errors without sudo privilege.

So, my question is that do I necessarily have to install virtualenv or can my problem of installing a working tensorflow be solved much simply.

It is a linux server and default python version is 2.6. So, I had to install 2.7 separately for my use.

Community
  • 1
  • 1
Pukki
  • 586
  • 1
  • 7
  • 18

3 Answers3

2

Try building from source instead of using the precompiled binary version. That way you won't have to worry about the glibc incompatibility with the installed system one.

dga
  • 21,757
  • 3
  • 44
  • 51
  • 1
    As tensorflow (google) uses bazel, I couldn't install this way too. When trying to install bazel it asks for some java packages that I don't have. Or can I install it without using bazel. I am a little new to this so I am not sure, but there documentation doesn't mention anything about installing without bazel. – Pukki Nov 21 '15 at 01:11
0

@Pukki you may be able to have a professor request the system administrator to install it. Then everyone on the system could use it.

technologiclee
  • 198
  • 1
  • 1
  • 10
  • Yes this is the conclusion I came to, too. – Pukki Nov 21 '15 at 08:05
  • But this is not a technical solution. It is an administrative workaround. – Pukki Nov 21 '15 at 08:06
  • @Pukki I assure that in the real world we also have to have administrative workarounds to get certain programming issues resolved. – Guy Coder Nov 21 '15 at 13:14
  • I think installing Docker or the virtual environment would need the same permissions. How about making it a portable app. I was able to put these on my student shared directory at school. http://portableapps.com/ – technologiclee Nov 21 '15 at 23:59
  • If pip is installed maybe you could use virtualenv. https://github.com/tensorflow/tensorflow/blob/master/tensorflow/g3doc/get_started/os_setup.md – technologiclee Nov 22 '15 at 00:54
  • @technologiclee If you go through my question, you will find that I had already tried virtualenv. BTW I talked to a professor and had it installed. So, thanks. – Pukki Nov 28 '15 at 08:26
  • Can you see TensorBoard when you run it from the server? http://stackoverflow.com/questions/33836728/view-tensorboard-on-docker-on-google-cloud – technologiclee Nov 29 '15 at 13:41
0

I followed the instructions from this page and it resolved my issue. You need the --upgrade URL parameter

Reza S
  • 9,480
  • 3
  • 54
  • 84