How Amazon Forecast leverages automated machine learning
During the training of your first predictor, you disabled AutoPredictor and then allowed the algorithm to choose its default value, which is AutoML. In Automated Machine Learning (AutoML) mode, you don't have to understand which algorithm and configuration to choose from, as Amazon Forecast will run your data against all of the available algorithms (at the time of writing, there are six) and choose the best one.
So, how does Amazon Forecast rank the different algorithms? It computes the weighted quantile loss (wQL; to read more about wQL, please check the Algorithm metrics section) for the forecast types you selected. By default, the selected forecast types are the median (p10) and the boundaries of the 80% confidence interval (p10 and p90). Amazon Forecast computes the wQL for these three quantiles. To determine the winning algorithm, it then takes the average over all of the wQL values that have been calculated: the...