Experiments with regard to explanation of embedded spaces and multi modal models.
First make sure you have a Python (<3.11) environment with dianna. Currently, we use a development branch that you can install with:
git clone https://github.com/dianna-ai/dianna.git
cd dianna
git checkout 279-embeddings
pip install -e .
Then clone the explainable_embedding
repository, cd
into it and install the additional dependencies for these experiments:
pip install -r requirements.txt
Run all experiments with
python benchmarking/distance_benchmark.py
A directory called output
will be created with the results.