nielsborie / machine-learning-environments Goto Github PK

Current behavior

• At this point there is no convention to identify the changes that have been made to the project.

Expected behavior

• Create a first CHANGELOG.md file to identify the changes associated with the current version.
• Create a goal Makefile to automatically generate / update this CHANGELOG.md. This goal will be useful for the next releases.

 docker run --rm --name ML-env --entrypoint='' nielsborie/machine-learning-environments python3 -c 'import numpy'

gives:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'numpy'

The same for all the listed libraries. Am I doing something wrong, or misunderstanding this image? Is it not a base with libraries preinstalled?

Insights

https://caffeinedev.medium.com/how-to-install-tensorflow-on-m1-mac-8e9b91d93706
https://developer.apple.com/metal/tensorflow-plugin/

Hi Niels - my students and I are using your image again this year. Can I ask - whilst it is easy to open a notebook in the browser, on the other hand I cannot seem to connect to h2o-flow.

When I start h2o - it would be nice to demonstrate h2o-flow from the browser. This opens as stardard on port 54321. However even though I expose port 54321 on the container - I cannot connect in a browser. Did you even try this?

Current behavior

All the project documentation / information is staged in the README.md.
No information is available under the Wiki section.

Expected behavior

Initiate the Wiki section with items in the REAME.md.

(Update/ Adjust informations later)

Hi Niels,

I was actually trying to install kaggle docker (python) docker but to no avail due to some issue - raised on the kaggle docker but yet to get any response. Many thanks as I came across this useful docker.

I just trying around with the docker only past 3 days (pretty new to it I must say). My queries:

1. The docker image installed successfully if I used 'pull' command. However if I download from github and cd to the directory & run 'build -t' command, there's actually a bunch of error (didn't manage to snapshot it) as well as leaving intermediary image. Do you know how to resolve this issue?

2. Reason why I would like to pull it directly because I want to install new packages.
   Say if I want to install more packages on the currently installed image, is there an easy way out without duplicating the image or rebuilding from scratch? (My main goal is to edit the image).

3. By the way, why you used the jupyter/tensorflow-notebook build: 5811dcb711ba?

Thanks.

Dear Niels,

My name is Igor Marques, an MSc Business Analytics student from Queen Mary University of London. I am currently working on my dissertation project where I am building various machine learning models. My group and I are using your docker image, and I must say it is a great one, thank you! It has pretty much all packages we need at the moment. However, I have been having some issues considering that some libraries are outdated (eg.: scikit-learn, pandas, etc.). I am quite new to Docker and Containers, so I apologize in advance if my query is rather simple.

Is it possible to update some of the libraries in your docker image? If yes, is this something I'd be able to do in my VM (I am using an instance on Google Cloud), or would that have to be done by you (assuming I wish to continue to use your docker image)? I have looked for different docker images for machine learning, but so far I have not found one as comprehensive as yours. And since we are working in a group, it would be nice to have all members using the same environment.

Thank you a lot for your help.

Regards,
Igor

Hi Niels,

Many thanks for this image which is very useful.
pandas.read_hdf requires the tables module. Would you be able to add
pip install --upgrade tables
to your docker file?

(I have raised this with jupyter as sorting it in the earliest image that imports pandas probably makes most sense. It might be that they prefer not to add it and this extra module fits quite well with the long list of optional extras at the end of your docker file...

Also, your image does not seem to have a tag (other than latest). Are you able to give it a tag - so that if you change it we know which image we are pulling?

Current behavior

• The docker images construction was achived automatically with a commit push on the master branch.
• This feature does not seem to work anymore?
• For the moment, images are manually pushed on DockerHub!

Expected behavior

• Set up a minimal workflow allowing to build an image according to a defined condition?
• We can start by triggering the build and the push when there is a commit on the develop branch.
• The tag associated to this image will be the current release number suffixed with -SNAPSHOT.

Current behavior

• All previously built images are only compatible on the Linux os.

Expected behavior

• Explore the possibility of building the image using the multi-arch building mode?
• Investigate if it is easier to propose the same image but on different architectures.

Current behavior

This image being a little old it is necessary to make it evolve.
Everything is controlled from the Dockerfile at the root of the repository and includes all the packages to install.
All this is very heavy, both for the build and for the pull.

Expected behavior

First step to reach the target vision: create a base image providing the foundation to iterate on new images.
For now, my idea is to create a /base folder with the most minimal item possible in the Dockerfile.
We will then see how to do at the version levels, image types and packages to include later.

Current behavior

• Current name of the Github repository and the DockerHub repository is ml-docker.
• This naming seems not appropriate.

Expected behavior

The goal being to propose a stable development | production environment in Python using a Docker image.
So, facing the complexity and the immensity of the possible versions and uses, it seems more appropriate to rename this project to machine-learning-environments.

Current behavior

• Only the latest docker tag is available in DockerHub. No other versions exist.

Expected behavior

• Have the legacy image version (identified by the v1.0 version) available on DockerHub.