InfoDecipher | Innovative Predictive Modeling

Frequently Asked Questions

Q: Do I need to pay for upgrades?

A: No! All upgrades are free to licensed users.

Q: How do I upgrade to newer versions?

A: You can either uninstall the older version (using "Start|All Programs|InfoDecipher|Gains# verX.XX.XX|uninstall"), then install the new version (available here). Alternatively, you may install the newer version alongside the older one.

Q: I have Gains# installed on a PC that doesn't have SAS, what should I do to read SAS datasets?

A: Don't worry. You don't need SAS on the PC. Go to "Start|All Programs|InfoDecipher|Gains#|Launch sasoledb.exe" to install a SAS driver (make sure you check the box "Local, single-user, read-only raw data access"). With the installation of this driver, you can directly read SAS datasets in Gains#.

Q: I accidentally excluded a variable while doing data exploration. How can I get it back?

A: Go to the Data tab. In the Exclusion column you will find the variable being checked. Uncheck the box, and save the setting. You will see the variable re-appear.

Q: Why do I sometimes only see the development curve in the gains chart, but not the holdout curve?

A: This is likely to happen when there are too few variables in the model and the holdout/validation data doesn't have enough values to form the same number of deciles as in the development data. The problem will usually go away when you increase number of variables in the model.

Q: How do I use the Z-statistic in variable robustness analysis?

A: For a variable, the Z-statistic measures the predictive value difference between the development and the holdout/validation datasets. For robust variables, you should expect the value to be below 2. You will need to pay attention to variables with Z>2.

Q: In the robustness table, some variables have Predictive Strength<0. Why is that?

A: The Strength measure used by Gains# is the Pearson correlation between the ATS (Adaptive Transformation Score) function and the target variable. A negative correlation for the holdout sample indicates a reverse correlation between the predictive variable and target variable, relative to the development data. Such variables are most likely not NOT robust predictive variables and it is recommended that they are not included in models, if possible.