Converting Logistic Regression Runs to Classification

GeneXproTools also allows you to convert Logistic Regression runs to Classification. This means that, among other things, you can easily access all the Classification fitness functions to drive the evolutionary fitting (there are a total of 15 different fitness functions in the Classification Framework, which is a nice addition to the basic R-square and Correlation Coefficient fitness functions of the Logistic Regression Framework). By going back and forth between both platforms, you can explore different modeling tools to fine-tune your models.

But there is another advantage to being able to convert Logistic Regression runs to Classification, especially if your interests lay in crisp classifications alone.

In the Logistic Regression Framework, the 0/1 Rounding Threshold (which is called the Optimal Model Threshold in the Logistic Regression Framework) is entirely evolved and finely crafted by the evolutionary system, rather than being ad hoc imposed by the modeler. This capability of being able to evolve the Optimal Rounding Threshold might prove invaluable for modeling certain datasets, but, most importantly, we are dealing here with a much more flexible and adaptable system that is totally free to co-evolve a unique, finely adjusted, rounding threshold for each and every model.

When a Logistic Regression model is converted to Classification, the Optimal Model Threshold that is inferred from the ROC Curve and Optimal Cutoff Point Analysis, is automatically set up as the 0/1 Rounding Threshold of the new Classification run. Note, however, that this rounding threshold is finely adjusted to the training data that was used to create a particular model, more specifically, the active model of this run. This means that you can only expect the confusion matrix it now generates for the training dataset to match the ROC Confusion Matrix inferred in the Logistic Regression Framework. Note, however, that for inverted models, the confusion matrix you’ll get in the Classification Framework will be inverted relatively to the ROC Confusion Matrix, which was adjusted to match the configuration of the Logistic Confusion Matrix.

It is also worth pointing out that, when you convert a Logistic Regression run to Classification, you can also use the Logistic Cutoff Point as your 0/1 Rounding Threshold. Note, however, that in this case you'll have to set up the 0/1 Rounding Threshold manually in the Fitness Function Tab of the Settings Panel. The confusion matrix you'll get in this case on the Classification Framework will match obviously the Logistic Confusion Matrix.

To convert a Logistic Regression run to Classification you need to:

1. Within the Function Finding Framework, choose Convert To Classification in the Logistic Regression menu.
   This opens the Save As dialog box and also asks if you want to save the current run before converting it to Classification. This way you will be able to come back to it if you need to.
2. Type the run name for the new Classification run and then click Save.
   When you click Save, GeneXproTools takes you immediately to the Classification Framework. Note that, in the Run History, the fitness values that are shown there correspond to the ones evaluated in the Function Finding Framework. By choosing Refresh All you can rapidly update these values to their true values in this new context. The default fitness function for the new run is Squared Accuracy, but you can easily choose another one in the Fitness Function Tab of the Settings Panel.