DeepUQ is a package for injecting and measuring different types of uncertainty in ML models.
(MIT or Apache 2.0 or another requires link changed)
To be updated, not yet released.
First, cd to where you'd like to put this repo and type:
git clone https://github.com/deepskies/DeepUQ.git
Then, cd into the repo:
cd DeepUQ
Poetry is our recommended method of handling a package environment as publishing and building is handled by a toml file that handles all possibly conflicting dependencies. Full docs can be found here.
Install instructions:
Add poetry to your python install
pip install poetry
Then, from within the DeepUQ repo, run the following:
Install the pyproject file
poetry install
Begin the environment
poetry shell
After following the installation instructions, verify installation is functional is all tests are passing by running the following in the root directory:
pytest
The scripts can be accessed via the ipython example notebooks or via the model modules (ie DeepEnsemble.py
). For example, to ingest data and train a Deep Ensemble from the DeepUQ/ directory:
python src/scripts/DeepEnsemble.py
With no config file specified, this command will pull settings from the default.py
file within utils
. For the DeepEnsemble.py
script, it will automatically select the DefaultsDE
dictionary.
Another option is to specify your own config file:
python src/scripts/DeepEnsemble.py --config "path/to/config/myconfig.yaml"
Where you would modify the "path/to/config/myconfig.yaml" to specify where your own yaml lives.
The third option is to input settings on the command line. These choices are then combined with the default settings and output in a temporary yaml.
python src/scripts/DeepEnsemble.py --noise_level "low" --n_models 10 --out_dir ./DeepUQResources/results/ --save_final_checkpoint True --savefig True --n_epochs 10
This command will train a 10 network, 10 epoch ensemble on the low noise data and will save figures and final checkpoints to the specified directory. Required arguments are the noise setting (low/medium/high), the number of ensembles, and the working directory.
For more information on the arguments:
python src/scripts/DeepEnsemble.py --help
The other available script is the DeepEvidentialRegression.py
script:
python src/scripts/DeepEvidentialRegression.py --help
The config settings for the models used in the paper can be found in src/utils/defaults.py
.
The user should run the following commands from the cli:
python src/scripts/DeepEnsemble.py --save_all_checkpoints --noise_level "low"
The noise level argument should be modified to run the medium and high settings as well.
Repeat for the DER:
python src/scripts/DeepEvidentialRegression.py --save_all_checkpoints --noise_level "low"
Next run the analysis scripts:
python src/scripts/AleatoricandEpistemic.py
python src/scripts/LossFunctions.py
python src/scripts/ParitySigma.py
To reproduce the random initialization runs for the DER (these already exist for the DE):
python src/scripts/DeepEvidentialRegression.py --save_all_checkpoints --noise_level "low" --save_chk_random_seed_init --rs 10
Change the value of the random seed to match those given in the src/scripts/Aleatoric_and_inits.py
script.
Finally:
python src/scripts/Aleatoric_and_inits.py