search

rsnk96 / Ubuntu-Setup-Scripts

Scripts to help you set up your Ubuntu quickly, especially if you're in any subfield of Data Science or AI!
Apache License 2.0
126 stars 76 forks source link

Add Dockerfile for building containers #80

Closed rajat2004 closed 4 years ago

rajat2004 commented 4 years ago

Closes https://github.com/rsnk96/Ubuntu-Setup-Scripts/issues/72 Docker Images - https://hub.docker.com/repository/docker/rajat2004/ubuntu-setup-scripts

TODO:

Maybe, possibly not required or useful

rajat2004 commented 4 years ago

ubuntu18.04-cuda10.1-cv2-tf was built after using this image and then running 3-ML-Basic.sh by hand inside container Will update the info on the docker images page on the details of the setup, running, etc.

rajat2004 commented 4 years ago

I think such a minimal Dockerfile should work, and the image with Cuda especially is already bloated The items left in the list don't seem very useful, or are too time-consuming

Any specific things which should be added?

rsnk96 commented 4 years ago

I think such a minimal Dockerfile should work, and the image with Cuda especially is already bloated

Agreed. We could follow some of the steps listed in this link to minimize the docker image size

The items left in the list don't seem very useful, or are too time-consuming

I think we should add the other components (zsh/conda/....). If people want a docker container for just using AI libraries, they would go with one of the other 100 options out there. I think where this would become useful is in also setting up the development environment (zsh+zim+conda+aliases+...). What are your thoughts on this?

rajat2004 commented 4 years ago

I think we should add the other components (zsh/conda/....). If people want a docker container for just using AI libraries, they would go with one of the other 100 options out there. I think where this would become useful is in also setting up the development environment (zsh+zim+conda+aliases+...). What are your thoughts on this?

I'm still not sure about Anaconda, since that always felt a bit too much to me, but there's no downside as such in creating the images, will look into building them. Zsh is okay, will add that. Had a thought, after installing OpenCV, we don't really need the entire downloaded folder and the build one as well, removing those should reduce by 3-4GB at least Will try to build the images in the next few days, I specifically kept less information in the Readme so that every time a new image is added, some changes are made, etc, no need to open PR for keeping the info upto date

If the Dockerfile and the Readme updates seem okay, I think that it can be merged, the updated images will be directly added to the Docker Hub

rsnk96 commented 4 years ago

I'm still not sure about Anaconda, since that always felt a bit too much to me, but there's no downside as such in creating the images, will look into building them. Zsh is okay, will add that. Had a thought, after installing OpenCV, we don't really need the entire downloaded folder and the build one as well, removing those should reduce by 3-4GB at least Will try to build the images in the next few days, I specifically kept less information in the Readme so that every time a new image is added, some changes are made, etc, no need to open PR for keeping the info upto date If the Dockerfile and the Readme updates seem okay, I think that it can be merged, the updated images will be directly added to the Docker Hub

Agreed, it might be better to keep conda out of it. If someone wants to create another Python environment, they could simply create another container off the docker image

rajat2004 commented 4 years ago

Tested removing the opencv folders after building, reduces the size by 3 GB Will be updating the images and info on Docker Hub as I create them This seems good to go to me

Docker image updated to include zsh, reduced image size

rajat2004 commented 4 years ago

Merging, don't think any more updates are required in this PR, if so, can be done in follow-up PRs