Yearly seasonality "hallucinating" on missing data.

[luap2703]

We are running prophet on datasets with weekly input data + a lot of missing data (in this case, the first 2 month of the last year are missing). However, the results don't make any sense, as we expected Prophet to work well with such gaps.

How can we fix this problem?