Avoid splitting trees by value missing

Hi, could anyone kindly let me know how to prevent the Decision Trees model from using value missing as a criterion to split trees?

Thanks,

hz

Decision trees only splits on missing if you have sparse data. Are you using nested tables possibly? If not, then you have a column in your table with enough NULLs that are correlated with your target that there is enough information gain to cause a split. You can use NOT NULL, but this will just prevent the model from processing altogether if nulls are present|||

Yes, unfortunately, there are a lot of null values. I am not sure what you mean by "use NOT NULL". I have many predictor columns. Their null values are not in sync. If "NOT NULL" is used as a filter for every predictor, there may be no case left.

|||Unfortunately there's no way to not split by a value that appears in the data in SQL 2005. It's something for us to think about for future versions (expecially the "NULL" case)

Avoid splitting trees by value missing

Hi, could anyone kindly let me know how to prevent the Decision Trees model from using value missing as a criterion to split trees?

Thanks,

hz

Decision trees only splits on missing if you have sparse data. Are you using nested tables possibly? If not, then you have a column in your table with enough NULLs that are correlated with your target that there is enough information gain to cause a split. You can use NOT NULL, but this will just prevent the model from processing altogether if nulls are present|||

Yes, unfortunately, there are a lot of null values. I am not sure what you mean by "use NOT NULL". I have many predictor columns. Their null values are not in sync. If "NOT NULL" is used as a filter for every predictor, there may be no case left.

|||Unfortunately there's no way to not split by a value that appears in the data in SQL 2005. It's something for us to think about for future versions (expecially the "NULL" case)

autoregression Trees

hi,

I am using Time series alogorrithm.I just wants to know about the autoregression tree.I am having data like

Studid Date Perf

001 01/01/2007 90

001 02/01/2007 95

001 03/01/2007 89

002 01/01/2007 79

002 02/01/2007 90

002 03/01/2007 95

Like that. when I use my Model Viewer --> Descision Tree --> It shows like

Perf = 90.0084 + 1.02 * Perf(-2) + 0.25 * Perf(-2).

What is this value and how its getting calculated?

Basically, the formula tells that the forcasted Perf value is determined by using a regression formula that uses, as regressors, the value Perf had two time slices ago.

If your time key were a date, then Today's Perf depends linearly on Perf two days ago.

More details on the autoregressive tree can be found in the paper on "Autoregressive tree models for Time Series Analysis" by C. Meek, D.M Chickering and David Heckerman, available here: http://research.microsoft.com/~dmax/publications/dmart-final.pdf