PP: DeepAR: probabilistic forecasting with autoregressive recurrent networks

FROM

Amazon research Germany

PROBLEM

probabilistic forecasting: estimate the probability distribution of a time series in future.

INTRODUCTION

a global model, which learns from historical data of all time series.

METHOD

an autoregressive recurrent network architecture.

lstm.

也是根据likelihood，条件概率进行计算。

分为两部分，train and prediction part。但是还不知道这两部分的区别。