The ProphetForecast writes a components table to BigQuery. The components table has 2 columns called y and yhat. When I query this table in BigQuery, I see the following issues:
y and yhat are very far apart, differing by up to 4 orders of magnitude.- based on common sense about what these metrics mean,
y is wrong. this is concerning because y is supposedly the target data used to train the model.
- based on how
yhat is set, yhat should be list of 1,000 numbers, but in the BigQuery table, there is only 1 number.
Other suspicious things you could try investigating as a way to resolve the 3 issues above:
- Maybe the wrong aggregation is getting used on
y. It's not obvious to me why we'd apply the same aggregations to both y and yhat.
- This comment sounds like it's doing the wrong thing. What does this sentence even mean?
In overlapping periods, the forecasted value will always be larger because it is the sum of the observed and forecasted values.
The ProphetForecast writes a components table to BigQuery. The components table has 2 columns called
yandyhat. When I query this table in BigQuery, I see the following issues:yandyhatare very far apart, differing by up to 4 orders of magnitude.yis wrong. this is concerning becauseyis supposedly the target data used to train the model.yhatis set,yhatshould be list of 1,000 numbers, but in the BigQuery table, there is only 1 number.Other suspicious things you could try investigating as a way to resolve the 3 issues above:
y. It's not obvious to me why we'd apply the same aggregations to bothyandyhat.In overlapping periods, the forecasted value will always be larger because it is the sum of the observed and forecasted values.