Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

Hey @BWBarger,

We added a numerical protection for the case when series dataset are completely zeros, some reconciliation methods make divisions using standard deviations.

Here is an example where we add some minor noise to make the algorithms work:
```# MinT along other methods require a positive definite covariance matrix
# for the residuals, when dealing with 0s as residuals the methods break
# data is augmented with minimal normal noise to avoid this error.
Y_df['y'] = Y_df['y'] + np.random.normal(loc=0.0, scale=0.01, size=len(Y_df))```

I still see this error with the small noise added to y. As I mentioned before, I only see this error when I try to include futr_exog_list

Wanted to check if this has been looked into further. I'm experiencing this issue in the google colab environment (https://colab.research.google.com/github/Nixtla/neuralforecast/blob/main/nbs/examples/HierarchicalNetworks.ipynb) as well, so I don't think it has anything to do with my personal setup. Are there any full samples of NHITS working with future exogenous features? Do we need to create an issue in github?