[SOLVED] Error while installing Anaconda

Error while installing Anaconda

I have Ubuntu 16.04 LTS 64-bit Python 2.7.12

I am installing Anaconda 5.2 for Python 2.7 version,and I am always getting this error:

[/home/nvidia/anaconda2] >>> nvidia/anaconda3 PREFIX=/home/nvidia/nvidia/anaconda3 installing: python-2.7.15-h1571d57_0 ... Anaconda2-5.2.2.0-Linux-x86_64.sh: line 370: /home/nvidia/nvidia/anaconda3/pkgs/python-2.7.15-h1571d57_0/bin/python: cannot execute binary file: Exec format error

I tried to change directory, it didn't work ? I tried different versions of the Anaconda, it didn't work ?

Solution

Your core problem is that you're trying to install software compiled for the x86_64 processor, but you have ARM8 processors. (According to the tech specs for the Jetson TX2, it has HMP Dual Denver 2/2 MB L2 and Quad ARM® A57/2 MB L2.)

The fact that they're both 64-bit doesn't make any difference; they're still radically different process architectures, with different instruction sets, so they can't run each other's code.

And that's why you get cannot execute binary file: Exec format error.

From what I understand, Anaconda only provides supported installers for a handful of linux platforms—x86_64, x86, and POWER.

And I'm not aware of anyone providing third-party ported binary installers.

However, it should be possible to build Anaconda, or at least Miniconda, from source.

For example, this site explains how to build it for IBM Z-Series computers running Ubuntu, which probably isn't too different from what you'd need to do.

Keep in mind that this is apparently just some guy's project, not official Anaconda, so there's no guarantees it's all correct and up-to-date.
You need to use whatever your platform says for uname instead of s390x.
There probably won't be prebuilt packages in Conda and Conda-Forge for your platform, so installing things may take a lot longer and fail a bit more often.
You may need to explicitly disable Intel MKL for source builds.

I don't think anything else is likely to be a problem, but I haven't tried anything remotely similar.

Alternatively, you could just use Ubuntu's Python, and manage your Python packages with the traditional combination of apt-get and pip.