Setup using Paperspace Docker

Hi all!

I’m so excited to be working through the course again. I found fastai a while ago, but decided to focus on learning data science fundamentals that would land me a job (which worked out!) and have now setup my desktop (16gb ram, ubuntu 16.04, 6gb gtx1060) to start working through the course.

I wanted to share a few gotchas I found after running the paperspace docker locally on ubuntu 16.04 to get it to work.

  1. I needed to install cuda 9 to run the newest course-v3 notebooks by going here: [](http://cuda 9)

  2. I needed to run the docker cmd with --shm-size to allow docker to share the appropriate memory from the host. The docker image is from: [](http://paperspace docker course-v3)
    sudo docker run --runtime=nvidia -d -p 8888:8888 -v /home/xbno/data:/data -v /home/xbno/ml/course-v3:/course-v3 --shm-size 26G paperspace/fastai:1.0-CUDA9.2-base-3.0-v1.0.6

  3. I also needed to figure out the password for the jupyter notebook which I couldn’t find via the typical running process and ctrl+c’ing. Entering the running docker container via docker exec -it <container_id> bash then source activate fastai then running jupyter notebook list shows the token you need.

  4. I found you can point the notebook to data you’ve saved locally instead of downloading it within the notebook by pointing the untar_data function to it. path = untar_data(URLs.PETS,'/data/oxford-iiit-pet.tgz'); path I have a feeling the other modifications I make to the notebooks etc will vanish if I stop the container, so this along with knowing how to persist the models outside of the notebook should prove useful.

Hopefully these tidbits are useful to somebody else out there. Encouraged by Jeremy’s inspiration, my goal this time around is to be an active participant in the forums. Can’t wait to download my own dataset and start classifying. Cheers!

Corrected the forum, hopefully it will let me post