Error while installing Anaconda

Question

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.