| CPC G06N 5/04 (2013.01) [G06F 18/2113 (2023.01); G06F 18/213 (2023.01); G06N 3/084 (2013.01); G06N 20/20 (2019.01)] | 19 Claims |
|
1. A device which ensembles data received from a plurality of prediction devices through a wireless or wired network, the device comprising:
a data manager configured to generate output data based on time-series data, to provide the output data to a first prediction device and a second prediction device through the network, to receive a first device prediction result corresponding to the output data from the first prediction device through the network, to receive a second device prediction result corresponding to the output data from the second prediction device through the network, to calculate a first device error based on a difference between the first device prediction result and the time-series data, and to calculate a second device error based on a difference between the second device prediction result and the time-series data;
a learner configured to adjust a parameter group of a prediction model for generating a first device weight corresponding to the first prediction device and a second device weight corresponding to the second prediction device, based on the first device prediction result, the second device prediction result, the first device error, and the second device error; and
a predictor configured to apply the first device weight to the first device prediction result to generate a first result, apply the second device weight to the second device prediction result to generate a second result, and to generate an ensemble result based on the first result and the second result,
wherein the ensemble result is provided to a terminal,
wherein the time-series data includes a plurality of features of an item at a plurality of times, respectively, the plurality of times being from a first time to a N-th time,
wherein the data manger generates the output data by sequentially generating (N−1) cumulative data, each subsequent pair of the (N−1) cumulative data including first cumulative data and second cumulative data, the first cumulative data including a first plurality of features of the item in the time-series data at a first plurality of times from the first time to an i-th time to generate prediction features of the first and second device prediction results corresponding to an (i+1)-th time, the second cumulative data including a second plurality of features of the item in the time-series data at a second plurality of times from the first time to a (i+1)-th time to generate prediction features of the first and second device prediction results corresponding to an (i+2)-th time.
|