search

zalandoresearch / pytorch-ts

PyTorch based Probabilistic Time Series forecasting framework based on GluonTS backend
MIT License
1.24k stars 191 forks source link

how to extract the predicted cross-covariance matrix from estimators? #69

Closed sgdantas closed 3 years ago

sgdantas commented 3 years ago

Hi, first of all, great work.

I was wondering how to recover the information shown in Fig. 5 in the tempflow paper https://arxiv.org/abs/2002.06103

Thanks!

kashif commented 3 years ago

Right so that figure was made by first recalling that the ground truth is a matrix of dim: [T,D] and the predictions are S=100 or so sample predictions so a tensor of shape [T,S,D]. If I remember correctly we first calculate the mean of each time step of the prediction to obtain [T,D] matrix. Finally, we can calculate the cross-covariance of the ground truth and this prediction matrix and the last figure is the absolute difference of these cross covariances.

Hope that helps?

sgdantas commented 3 years ago

awesome, thanks! Closing the issue