Skip to content

lmoddemann/Discret2Di

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Discret2Di - Discretization for automated consistency based diagnosis

First steps

At the start, take a look at the instructions in the notebook file named "prepare_datasets.ipynb". Once you're done, you can access the preprocessed datasets for training the model and evaluating our approach. Also, make sure to run the "tank_generation.ipynb" file to create the simulated Three-Tank dataset.

The resources for finding the datasets are as follows:

Model Training of the CatVAE

Please find the model hyperparameters for our evaluation in this Repo under "Discret2Di_Appendix.pdf".
The parameters can also be found in the folder /CatVAE_training_hparams/ for every single evaluation dataset.
As well, you can find the tuned parameters for the Gaussian Mixture Models.

Evaluation

Within the following folders, the evaluation of the specific sections of the paper can be reviewed.

Evaluation of CatVAE (folder path: notebooks_evaluation_CatVAE): Discretisations can be carried out in the notebooks with the pre-trained models and compared on the basis of the plots.

Evaluation of GMMs compared to CatVAE (folder path: notebooks_evaluation_gmm): As a baseline to the CatVAE, a GMM is trained to discretize the different datasets and compute their according likelihood.

Evaluation of Discret2Di (folder path: notebooks_evaluation_discret2di): Based on the simulated data set of the three-tank model, various anomalies were simulated as described in the preprocessing step.
Within the folder, the discretisations of the CatVAE can first be checked and then the notebook "discret2di_tank.ipynb" can be executed, in which the various diagnoses are carried out.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors