ISO Dates + BigQuery ML ARIMA models = less errors

ISO Weeks save the day when it comes to 31st Dec/1st Jan

Goal = Forecast weekly growth using ARIMA model using 2 years of data

I was extracting the month from the timestamp using

EXTRACT(WEEK FROM TIMESTAMP) AS week

However, that would result in 53 (0–52) weeks in the year, which the BigQuery ML ARIMA model did not like because week_num = 0 would not be a full week (less than 7 days)