Best way to install python libraries on Mark 1 device


#1

Currently, I’m installing the libraries with sudo mycroft-pip install xxx. But for certain libraries such as tensorflow I’m facing some issues, it cant find any version for tensorflow. Is there another way to install libraries on mark 1?


#2

To my knowledge there is no current tensorflow version available as binary distribution for the Raspberry Pi. You will find some older versions here: https://github.com/samjabrahams/tensorflow-on-raspberry-pi/releases

For a current version (Tensorflow 1.14) you will have to build it yourself. I tried this myself and gave up after i have encountered too many dependency problems starting with the bazel-build tool…


#3

I actually did manage to install the 1.14 version, haven’t actually tested it though. I sent it the link to a whl file. Other than that, scipy installation is getting really tricky.

Is there any other alternative, if I’m not able to install all the libraries?


#4

when scipy fails with „lapack/blas resource not found“:
sudo apt-get install libblas-dev liblapack-dev


#5

I was able to get scipy whl file to install it, but librosa (one of the libraries I’m trying to install), requires llvm which is a bit hard to install on rasberry pi


#6

Just for anyone in future, who’s gonna get lost like me but hopefully save a LOT of time.

If you’re not using llvm at all, install librosa 0.42 (or tweak versions), it doesn’t need llvm. For other libraries, normally downloading whl files directly works as well.

keywords - LLVM, Rasberry pi, LLVMlite


#7

Mods, I accidentally deleted my post. How can I revert it? Contain potentially useful information.


#8

possibly check the edit history?


#9

Okay, I’m stupid. The button was right there lol. I really don’t know how I missed it.