Hi all, There was a issue with the time-series extension where all events keep getting added to the extension and all subsequent regressions consider ALL previous events for the regression calculation. This obviously leads to 'out of memory' issues and the solution was to implement an upper limit to the number of events that will be held in memory for regression calculation.
This has been implemented and the user can specify the batchsize as part of the siddhi query for the time series regression. Furthermore, I felt it useful to allow the user the ability to calculate regression every 'n' events, instead of at every event (especially in the case of extremely low latency data streams such as stock market data). This too has been implemented, in a way that user can specify the calculation interval as part of the siddhi query for timeseries regression. So an example of a time series regression query would be as follows... *from DataStream#transform.timeseries:regress( 1000, 1000000, 0.95, y, x1, x2, x3, x4 )* *select ** *insert into RegressionResult* where, 1000 is the calculation interval 1000000 is the batchsize 0.95 is the confidence interval y is the dependent variabe (which we are forecasting) x1, x2, x3, x4 are the independent variablese that are used for forecasting y. Let me know if you have any thoughts/suggestions on this. Cheers, Seshika
_______________________________________________ Architecture mailing list Architecture@wso2.org https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
