Repository for the 'Usable ML' Software Project @ FU Berlin

General

This repository is for the 'Usable ML' software project course at FU Berlin, provided by Fraunhofer AISEC. Students will develop a graphical user interface that allows creating machine learning models and manipulating them. Possible features include:

• Training Interface
  • start training
  • interrupt training
  • continue training
  • adjust parameters (e.g., learning rate, loss function, momentum, dropout-rate) (0.5 P)
  • before the training
  • during the training
  • revert to an earlier epoch (1 P)
  • freeze parts of the model (1 P)
• Training Monitor
  • display accuracy and loss over time for training set
  • indicate point in training where a parameter was changed
  • display accuracy and loss over time for test set (0.5 P)
  • display layer-specific information (e.g., gradients) (1 - 2 P)
  • compare different runs (1 P)
  • fork graph when parameters of earlier epochs are changed (1.5 P)
• Model Creator
  • create models using a GUI (2 P)
  • change the composition of layers (1 P)
  • change aspects of the layers (e.g., sizes) (0.5 P)
• Model Evaluator
  • select stored model to be evaluated (0.5 P)
  • evaluate per-class accuracy on test set (or training set, or arbitrary dataset) (1 P)
  • display special examples which (2 P)
  • are falsely predicted
  • are predicted with a small loss

Items in bold are expected as a minimum feature set.

Installation

1. Clone the environment and go into the folder.

   git clone ...; cd UsableML

2. Install the requirements

   pip install -r ./requirements.txt

3. Install PyTorch into your environment.
   Refer to this page for specific instructions: https://pytorch.org/get-started/locally/

Usage

Activate the environment

conda activate UsableML

Run the backend

python backend.py

Run the frontend

export PYTHONPATH="${PYTHONPATH}:$(pwd)"
streamlit run app.py

License

This project is licensed under the GNU Affero General Public License v3.0.