Automatic Miscalibration Diagnosis

$ # installation (originally with Python 3.10.4 and CUDA 11.7)
$ virtualenv venv
$ source venv/bin/activate
$ pip install -r requirements.txt

$ # seeds for model training
$ python
>>> import random
>>> sorted(random.sample(range(100), k=5))
[4, 7, 8, 9, 15]

$ # generate synthetic multimodal data set
$ python generate.py

$ # train interpreter
$ python train.py interpreter
$ # train density network or mixture density network
$ # on synthetic / year / protein / power data set
$ python train.py --seed 4 mdn --components 1 --neurons=50 power
$ # see train.py Python script for more options

$ # explore experiment.ipynb Jupyter notebook
$ jupyter lab