Time Series Anomaly Dectection using a Variational Inference Approach
In this project we want to achieve anomaly detection in time series using a variational inference approach.
The project is divided in different files to achieve a more efficient modularization.
The files used to define the models, the losses, the loading of the data, the training
and other useful functions are in this root folder and have the ts_ (timeseries) prefix.
Everything was put together in the notebook main.ipynb, that was used to train all the
models with all the datasets and display the anomalies detected.
The figures and trained networks are available in the folder runs
Different notebooks are available in the forlder notebooks. They display the different
architectures built, parameters used and results achieved in training (this means that some
pieces of code look redudant because they have been placed there for reference).