Tools

Thumbnails: Stable Diffusion

Some of the thumbnails are created by hand, some are created via Stable Diffusion and some are taken from other sources (in that case, they are credited in the ressources below the post).

Web search with Large Language Models (Metaphor)

Link.

GIF-Editor

Link.

change the speed
crop, resize and rotate

MKV to GIF converter

Small file but a lot of quality loss: Link. (supports many other formats as well.)

Also bad quality: Link

Docker

I’ve been wanting to try Docker for quite a while now, but today i’m finally looking into if it makes sense to containerize your ML models.

-> it makes sense when your institution is using Kubernetes or other workflows involving containers

Pros

Docker is platform-agnostic, meaning you can run your code on any machine and OS without any problems

Challenges

Linux Mint: could not find a good application to manage containers (Using Whaler for now.)
had problems with the docker socket path with the Whaler app (Fixed. The path is /var/run/docker.sock, see Stackoverflow for troubleshooting)

Dockerfile

tells container what packages to install
gives container command to execute after packages are installed

A Dockerfile might look like this:

FROM python:3.10

# Launch new folder
WORKDIR /app

# Put all files into the app directory
ADD . /app

# Install all packages
RUN pip install -r requirements.txt

# Run the application
CMD ["python", "app.py"]

Container

instance of an image
isolated environment

requirements.txt

put all required packages into this text file as follows:

tensorflow
matplotlib
# ...

CMD

Build the Docker image:

docker build -t myapp:1.0 .

Args: -t: title

Start a container

docker container run -d -p 5000:5000 myapp:1.0

Args: -d: detach from terminal -p: port

Docker Images

environment where the specified packages are installed

References

Stackoverflow: Docker path fix