Sorting: What is it ? What is it ? Name: What is it ? LPE: What is it ? TE: What is it ? HU and seed:What is it ?

Testing individual models

When you train your model, the training is actually done on the first half of the database, which means the second half contains data that have not been used for training.

The window you obtain when you press "Test" in the Model Manager is:

Sorting your models  Top

The first submodel appearing is always the best one. However, you have the opportunity to select the criterion which defines the best model. By default, the best submodel is that with the highest Log Predictive Error (LPE). By selecting TE, you can sort the submodels by increasing Test Error.

Testing a submodel  Top

You can test any of the submodels you want by selecting it and pressing "Test Selected Model":

When you test a model, the Model Manager calculates predictions using the whole database, and compare them with the actual target values. It divides the results into training set (for which the agreement is expected to be good since these are data that have been used to build the model) and testing set (for which the agreement is always worse, because these are unseen data).

The calculation time for the predictions can be relatively long (a few minutes), but the results are all saved. This means that, once you have tested, say, the first model, the time you have to wait if you later repeat this will be much shorter.
This will also shorten the time required in the first step of the committe preparation.

Model Manager Tutorial
Copyright Neuromat Ltd, All rights reserved
Author: Thomas Sourmail, Neuromat Ltd
Back to main Top