If this is your first time here, please refer to the repository's associated web page here. (It contains my findings and presents my code)
This is the master branch, and it contains code that trains an autoencoder on our dataset. The idea is to find latent representations for the images.
The Plotting folder contains an ipynb that lets you plot a clustering from the data.
To get started using this repository, please install dependancies using pip:
$ pip install -r requirements.txt
The project file structure should be as follows:
.
├── LICENSE.md
├── Plotting
│ ├── clustering.ipynb
│ ├── umap_pixel.png
│ └── umap_pixel_2.png
├── README.md
├── main.py
├── model.py
├── Resources
│ └── stimuli
│ └── <data>
├── requirements.txt
├── run.sh
├── scratch
│ ├── Dataloading.ipynb
│ └── thoughts.md
├── settings.json
├── train.py
└── utils.py
Add the data at Resources/stimuli
If you plan to log results to your comet.ml repository, please add and populate a settings.json file.
Your file should look something like this:
{"username":"<username>", "apikey":"<key>", "restapikey":"<key>", "project":"neuromlnoodle"}
In order to launch the program, you can launch run.sh or run using python3
python main.py --args
The various arguments and their functionality are listed in main.py file.
The .ipynb notebooks can be moved to their parent directory and run interactively.
Logs of the experiments that I ran, are available here.
Don't forget to look at the other two branches in this repository:
- The classification branch: Where I shuffled the dataset and built the emotion classifier
- The gh-pages branch: for this repository's website and 'homework' presentation.
Enjoy!
(PS - if you're wondering why I named this noodling, I'm refering to this definition)