Streaming TCM Node - General Build options
- Confidence interval width (%)
- This setting controls the confidence intervals for both forecasts and model parameters. You can specify any positive value less than 100. By default, a 95% confidence interval is used.
- Maximum number of inputs for each target
- This setting specifies the maximum number of inputs that are allowed in the model for each target. You can specify an integer in the range 1 - 20. The model for each target always includes lagged values of itself, so setting this value to 1 specifies that the only input is the target itself.
- Model tolerance
- This setting controls the iterative process that is used for determining the best set of inputs for each target. You can specify any value that is greater than zero. The default is 0.001. Model tolerance is a stop criterion for predictor selection. It can affect the number of predictors that are included in the final model. But if a target can predict itself very well, other predictors may not be included in the final model. Some trial and error may be required (for example, if you have this value set to high, you can try setting it to a smaller value to see if other predictors can be included or not).
- Outlier threshold (%)
- An observation is flagged as an outlier if the probability, as calculated from the model, that it is an outlier exceeds this threshold. You can specify a value in the range 50 - 100.
- Number of Lags for Each Input
- This setting specifies the number of lag terms for each input in the model for each target. By default, the number of lag terms is automatically determined from the time interval that is used for the analysis. For example, if the time interval is months (with an increment of one month) then the number of lags is 12. Optionally, you can explicitly specify the number of lags. The specified value must be an integer in the range 1 - 20.
- Continue estimation using existing models
- If you already generated a temporal causal model, select this option to reuse the criteria settings that are specified for that model, rather than building a new model. In this way, you can save time by reestimating and producing a new forecast that is based on the same model settings as before but using more recent data.