02-Embeddings

Intro

• See Image Embeddings, Clustering, and Similarity Search

Embeddings

We will use DinoV2, a vision transformer model, to extract image embeddings from the Fashion MNIST dataset. These embeddings will be used to visualize image similarity via clustering and to perform similarity search using k-nearest neighbors (KNN).

Environment Setup

Temp Directory

• The virtual environment will be created in the fast, temporary file system located at $TMPDIR

echo '# Set a temporary directory' >> ~/.bashrc
echo 'export TMPDIR=${TMPDIR:-/tmp}' >> ~/.bashrc
source ~/.bashrc
echo $TMPDIR

• Create the virtual environment in $TMPDIR

uv venv $TMPDIR/.venv

• Activate the environment

source $TMPDIR/.venv/bin/activate

• Create an alias for activation

echo "alias activate='source \$TMPDIR/.venv/bin/activate'" >> ~/.bashrc
source ~/.bashrc

• Navigate to the user/ directory and link dependencies

# cd user folder
 
# Create a symbolic link named .venv in your user folder, 
# pointing to the actual environment in $TMPDIR
ln -s $TMPDIR/.venv $(pwd)/.venv
 
# Install project dependencies defined in pyproject.toml
cd ~/dsgt-arc/fall-2025-interest-group-projects/project/02-embeddings
uv pip install -e .

Run the Jupyter Notebook

• Activate the .venv

activate

• Install ipykernel

uv pip install ipykernel

• Register the kernel

python -m ipykernel install --user \
    --name dsgt-tmp \
    --display-name "DSGT test ($TMPDIR)"

• Open a copy of embedding.ipynb in Jupyter and select the linked .venv kernel

jupyter notebook

Set up the virtual environment in scratch

• Create a .venv in scratch

# create a directory to store the environment
mkdir -p ~/scratch/dsgt-arc/embeddings
 
# create a virtual environment using uv
uv venv ~/scratch/dsgt-arc/embeddings/.venv

• Create alias in ~/.bashrc

alias activate='source ~/scratch/dsgt-arc/embeddings/.venv/bin/activate'

• Activate it

activate
# verify
which python

~/scratch/dsgt-arc/embeddings/.venv/bin/python

• Link the scratch .venv with the lora project .venv and Install dependencies using pyproject.toml

Link .venv and install dependencies

# navigate to project directory
cd ~/dsgt-arc/fall-2025-interest-group-projects/user/ctio3/
 
# Delete existing symlink
rm .venv
 
# create a symbolic link to your environment in this project folder
ln -s ~/scratch/dsgt-arc/embeddings/.venv $(pwd)/.venv
 
# verify
ls -l .venv
 
# Install project dependencies defined in pyproject.toml
cd ~/dsgt-arc/fall-2025-interest-group-projects/project/02-embeddings
uv pip install -e .

• Register Jupyter kernel

python -m ipykernel install --user \
  --name embeddings \
  --display-name "DSGT Embeddings (scratch)"

• Connect to remote Jupyter server

ssh pace-interactive
jupyter lab --no-browser --ip=0.0.0.0 --port=8888

Create a start Jupyter script

• Create a start-pace-jupyter.sh script

start-pace-jupyter.sh

#!/usr/bin/env bash
set -e
 
PORT=8889
REMOTE=pace-interactive
PROJECT_DIR=~/dsgt-arc/fall-2025-interest-group-projects/user/ctio3/02-embeddings
VENV=~/scratch/dsgt-arc/embeddings/.venv
 
ssh -L ${PORT}:localhost:${PORT} ${REMOTE} << EOF
cd ${PROJECT_DIR}
source ${VENV}/bin/activate
jupyter lab --no-browser --port=${PORT}
EOF

• Make it executable

chmod +x start-pace-jupyter.sh

• Run the script

./start-pace-jupyter.sh

Next

03-information-retrieval