I have a Pc with Ubuntu 16.04 with fastai 0.7 and lessons 2018. Can I install course-3.fast.ai incl lessons 2019 and still get access to lessons 2018? If yes how do I install and configure 2019?
In the menu at “https://course.fast.ai/” there are under “Server setup” many different plattforms but no “Your own server”. Could you add a link to a page with how to install your own server 2019?
I’ve extracted some parts from SageMaker setup. You own server should be no different than “my pc”. Keep in mind that you will need a proper GPU (I don’t have one on my MBP, for example) to run these things fast enough (I believe you get 3 to 4 orders of magnitude faster training times on GPUs, please somebody correct me if I am wrong).
The following setup will create a separate conda environment for fastai. It is possible it’s unnecessary, but will be safer so it won’t interact with packages from the rest of the system.
# install the fastai library and dependencies
echo “Installing the fastai library and dependencies in new conda enviornment”
mkdir -p ~/fastai-course
conda create -mqyp envs/fastai -c pytorch -c fastai fastai ipykernel
echo “Finished installing environment”
# clone the course notebooks
echo “Clone the course repo”
git clone https://github.com/fastai/course-v3.git ~/fastai-course/v3
# install jupyter extensions
source activate envs/fastai
echo “Install jupyter nbextension”
pip install jupyter_contrib_nbextensions
jupyter contrib nbextensions install --user
source activate envs/fastai # to enter created environment
jupyter notebook v3
Edit: Edited the setup to have less commands.
THANK YOU!!! (I have 2 GTX 980)
I believe installing fastai kernel will fail because it’s a little bit SageMaker specific. I will post complete instructions in few hours.
search the forums for “Local Linux setup” …
There are a number of already excellent posts describing everything you need to do to build your on machine for DL and set it up with all the pytorch/fastai bits required.
The 2018 repository had an
environment.yml file that set up the conda environment perfectly. Why no such file in the current repo. I really don’t want days of configuration hell, or to pay for a commercial service when I have everything at home