Hi, I read the code in n_hits_multivariate.py, but get confused of the way the datasets are loaded. In the tsdataset.py, the DataFrame Y_df is defined as 'Target time series with columns ['unique_id', 'ds', 'y']', well, take ETT dataset as an example, there are one column as 'date' and other 7 columns as 'variables of different nodes', currently I view the 'unique_id' part as the default index of panda.DataFrame, so what is the 'ds' and 'y'? What's more, it seems that N-Hits works in an univariate way in the following line: https://github.com/cchallu/n-hits/blob/d882ee60b34c0ab6b67b31001f735b181a9efb93/src/models/nhits/nhits.py#L330
That makes me confused, how does N-Hits make multivariate predictions? By individually yielding prediction results of each univariate variable?
Hi, I read the code in
n_hits_multivariate.py
, but get confused of the way the datasets are loaded. In the tsdataset.py, the DataFrameY_df
is defined as 'Target time series with columns ['unique_id', 'ds', 'y']', well, take ETT dataset as an example, there are one column as 'date' and other 7 columns as 'variables of different nodes', currently I view the 'unique_id' part as the default index of panda.DataFrame, so what is the 'ds' and 'y'? What's more, it seems that N-Hits works in an univariate way in the following line: https://github.com/cchallu/n-hits/blob/d882ee60b34c0ab6b67b31001f735b181a9efb93/src/models/nhits/nhits.py#L330 That makes me confused, how does N-Hits make multivariate predictions? By individually yielding prediction results of each univariate variable?