LatentSpaceVisualizer

The Latent Space Visualizer is an interactive Python notebook that visualizes the latent space of an image dataset. The dataset contains a collection of latent variables that describes a set of labeled images.

Two variables of the latent space are displayed in a 2D scatter plot. When the mouse is positioned near a point in the scatter plot, the image corresponding to that point is displayed to the side.

Getting started

Clone this repository.

git clone https://github.com/compSPI/LatentSpaceVisualizer.git

Change the current working directory to the root of this repository.

cd LatentSpaceVisualizer

Download from the Anaconda Cloud and install the Python environment that has the dependencies required to run the code.

conda env create compSPI/compSPI

Activate the environment.

conda activate compSPI

Install the kernel.

python -m ipykernel install --user --name compSPI --display-name "Python (compSPI)"

Exit the environment.

conda deactivate

Running the notebook

Run jupyter notebook.

jupyter notebook 

Open the tutorial notebook latent_space_visualizer.ipynb.

Change the Python kernel to compSPI.

Set dataset_file to an HDF5 file containing the dataset.

dataset_file = '../data/cspi_synthetic_dataset_diffraction_patterns_1024x1040.hdf5'

Run the notebook.

Installation

The project package can be installed by running the following command.

python setup.py install

Code structure

The relevant files and folders in this repository are described below:

• README.md: Highlights the usefulness of the Latent Space Visualizer.

• latent_space_visualizer.ipynb: Provides a tutorial notebook for using the Latent Space Visualizer.

• latent_space_visualizer/: Contains the Python file required to run the notebook.