This is a little hacked-together notebook demonstrating how to plot an arbitrary tensor in the "Projector" pane of TensorBoard, with images for the points.
It's super hacked-together because TensorFlow 2.0 has "simplified" the API for
plotting embeddings, and only plots Embedding layers by default.
See this issue for when this feature might be cleaned up.
-
Clone the repo
-
Create a conda environment,
conda create -n autoencoder-fashion-mnist python=3 -
Upgrade pip:
pip install pip --upgrade -
Install requirements:
pip install -r requirements.txt -
Run the notebook:
jupyter notebook... ! -
Run TensorBoard with
tensorboard --logdir logs -
Jump over to the "Projector" tab!
Install the requirements, and then:
FLASK_APP=app.py flask run