Hello,
Can anyone post steps required to install Fastai_v2 on a Linux system ?
It’s in the readme here: https://github.com/fastai/fastai
1 Like
But those instructions do not seem complete. They are missing the installation of the jupyter notebook, and probably some supporting utilities. Please we should not assume that the os has various random stuff installed.
If using conda we should be using environments. It’s pretty standard Python practice. Makes support and setup a lot easier. ref: conda - Managing environments
An official environment.yml file for students would be really usefull. There is a fastai/environment.yml file in the reppsitory, which is nice, but we need one that installs both jupyter and any end-user tools, ass well as fastai lessons/framework
If I create a new conda environment and activate it:
% conda create --name fastai
% conda activate fastai
Then I do the documented install:
% conda install -c fastai -c pytorch fastai
Clone the repository, change into the directory, and try to run jupyter notebook:
% git clone git@github.com:fastai/fastbook.git
% cd ./fastai
% jupyter notebook
% jupyter notebook
> command not found: jupyter
Straw-man student my_environment.yml
<repository>::<library> is a supported syntax, that forces a library to be configured from a particular repository. I’m trying to avoid confusing the important plain-vanilla fastai configuration. I don’t say that this will be perfect, but it does allow more knowledgeable people to incrementally improve it.
% cat my_environment.yml
name: fastai
channels:
- fastai
- pytorch
- defaults
dependencies:
- fastai
# My additions
- conda-forge::notebook
Example usage
Not too tricky… and precise.
https://github.com/mamba-org/mamba is a lot quicker than conda - optional…
% conda install -c conda-forge mamba
% mamba env create -f ./my_environment.yml
% conda activate fastai
% jupyter notebook
1 Like
I agree. I really hope they had said I should install jupyter.
Looks like most of what I asked for has arrived.