5

I'm trying to use the Anaconda environment on a Windows 10 system to install tensorflow 1.9.0. Unfortunately, I'm behind a firewall that blocks all access to the Amazon cloud (please don't ask why).

I know that I can install version 1.9.0 from here but this repo heavily relies on conda-forge.

Thus, to circumvent this issue I'm using the following command to (at least) install tensorflow 1.8.0:

 conda install -c https://repo.continuum.io/pkgs/main/win-64/ tensorflow-gpu=1.8.0

This results in the following output:

Solving environment: done

Package Plan

environment location: d:\Programs\Anaconda\Anaconda

added / updated specs: - tensorflow-gpu=1.8.0

The following packages will be downloaded:

package                    |            build
---------------------------|-----------------
bleach-1.5.0               |           py36_0          24 KB  conda-forge
html5lib-0.9999999         |           py36_0         182 KB  conda-forge
------------------------------------------------------------
                                       Total:         206 KB

The following NEW packages will be INSTALLED:

absl-py:             0.2.2-py36_0          defaults
astor:               0.6.2-py36_1          defaults
gast:                0.2.0-py36_0          defaults
grpcio:              1.12.1-py36h1a1b453_0 defaults
markdown:            2.6.11-py36_0         defaults
tensorboard:         1.8.0-py36he025d50_0  defaults
tensorflow-base:     1.8.0-py36h1a1b453_0  defaults
tensorflow-gpu-base: 1.8.0-py36h376609f_0  defaults
termcolor:           1.1.0-py36_1          defaults

The following packages will be UPDATED:

cudatoolkit:         8.0-4                 defaults --> 9.0-1                 defaults
cudnn:               6.0-0                 defaults --> 7.1.4-cuda9.0_0       defaults
numpy:               1.12.1-py36hf30b8aa_1 defaults --> 1.14.3-py36h9fa60d3_2 defaults
tensorflow:          1.1.0-np112py36_0     defaults --> 1.8.0-0               defaults
tensorflow-gpu:      1.1.0-np112py36_0     defaults --> 1.8.0-h21ff451_0      defaults

The following packages will be DOWNGRADED:

bleach:              2.1.3-py36_0          defaults --> 1.5.0-py36_0          conda-forge
html5lib:            1.0.1-py36_0          defaults --> 0.9999999-py36_0      conda-forge

Proceed ([y]/n)? y

Unfortunately, the packages bleach and html5lib are not on the default repo but on conda-forge (which is blocked) instead, resulting in:

CondaHTTPError: HTTP 503 SERVICE UNAVAILABLE for url https://conda.anaconda.org/conda-forge/win-64/bleach-1.5.0-py36_0.tar.bz2 Elapsed: 00:00.287467

An HTTP error occurred when trying to retrieve this URL. HTTP errors are often intermittent, and a simple retry will get you on your way.

CondaHTTPError: HTTP 503 SERVICE UNAVAILABLE for url https://conda.anaconda.org/conda-forge/win-64/html5lib-0.9999999-py36_0.tar.bz2 Elapsed: 00:00.272954

An HTTP error occurred when trying to retrieve this URL. HTTP errors are often intermittent, and a simple retry will get you on your way.

Any help would be highly appreciated. Thank you very much!

Hagbard
  • 3,430
  • 5
  • 28
  • 64
  • I finally managed to install TensorFlow 1.8.0 by simply downloading bleach and html5lib from this repo (https://repo.continuum.io/pkgs/free/win-64/). However, the general issue that I cannot access conda-forge behind my firewall still remains. – Hagbard Jul 23 '18 at 08:37
  • I am encountering the same situation, I was using everything fine up until a new update. The reason my company has blocked anaconda forge is because it is considered a peer to peer network and no exceptions are made for peer to peer. Oddly enough installing on Anaconda prompt using `pip` seems to work – Moo10000 Dec 09 '19 at 19:30

1 Answers1

6

Eleven months later but...

Try to set http_proxy and https_proxy enviroment variables.

set http_proxy=http://user:password@yourcompanyproxy:port
set https_proxy=https://user:password@yourcompanyproxy:port

Other usefull tip is setup conda to don´t verify ssl.

conda config --set ssl_verify no

By the other hand, if the previous option failed you can use pip:

pip install --trusted-host pypi.org --trusted-host files.pythonhosted.org desired_package_name

But, the most common if you have a lot of developers is to install a package manager, like sonatype Nexus.

conda config --prepend channels http://nxrmserver:8081/repository/conda-public/

Hope that can help someone.

Josemar Pereira
  • 71
  • 1
  • 3
  • Unfortunately, I cannot test this any more but I still appreciate the effort. Thank you. – Hagbard Jun 24 '20 at 08:26
  • 2
    conda config --set ssl_verify no -> worked right away behind the corporate firewall, thank you for posting this solution! – Matt Aug 09 '21 at 16:19