Probabilistic forecasting with nonparametric quantile functions

A computer-implemented method comprising:

• receiving a request to generate a time series forecast for a time window comprising one or more points of time;
• generating the time series forecast using at least a recurrent neural network (RNN) trained based on time series data, wherein the RNN was trained by minimizing a continuous ranked probability score (CRPS) loss function derived from a nonparametric representation of an output distribution, the nonparametric representation of the output distribution defined by parameters generated by the RNN based on the time series data;
• and
• sending data comprising the time series forecast.

A system comprising:

• a front end implemented by a first one or more electronic devices of a provider network, the front end including instructions that upon execution cause the front end to:
• receive a request to generate a time series forecast for a time window comprising one or more points of time;
• and
• send the request to a forecasting service;
• and
• the forecasting service, implemented by a second one or more electronic devices of the provider network, the forecasting service including instructions that upon execution cause the forecasting service to:
• generate a time series forecast using at least a recurrent neural network (RNN) trained based on time series data, wherein the RNN was trained by minimizing a continuous ranked probability score (CRPS) loss function derived from a nonparametric representation of an output distribution, the nonparametric representation of the output distribution defined by parameters generated by the RNN based on the time series data;
• and
• send data comprising the time series forecast.