Project for Models of sequential data course at skoltech 2021
The main inspiration will be this repository with the implementation of End-to-End Learning of Coherent Probabilistic Forecasts for Hierarchical Time Series Also, an example of a high-quality project for us will be Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting
First, the time series data goes into the RNN layer to generate the distribution parameters.
At the second stage, samples are generated from the distribution and they are projected onto the space of hierarchical. This projected data is considered to be the predictions of the model.
Instructions for starting and working with a software implementation are given in the readme of the corresponding repository.