Preparing your database

Before you proceed, make sure you know the requirements for the format of the database.

The first step in building your model is to prepare the database you want to use to train it. When you choose to prepare a model, you should see the following screen:

To indicate which database you want to use, you can either enter its path or use the "Browse..." button to select it.
Once you have done that, just press "prepare".

What is being done

Before you can train the neural network, the following needs to be done:

The number of data lines is made even if necessary, and the labels are removed and kept apart.
The database is randomised, to ensure that both the half used for training and testing contain similar informations.
The references are removed and kept apart.
The minimum and maximum of each variable and the target are searched, the information is stored in the file MINMAX.
You can choose whether you want to have all the lines containing the minima and maxima in the training set.
All the information is normalised (ie each variable now has all its values between -0.5 and 0.5).
The database is split into two files (variables and target).

As is said in the message window, all the files are kept in the directory you have chosen initially for the model you are now using. However, note that the database you have chosen is not modified, but it is neither copied to the directory of the model. It can be a good idea to copy the database into the directory of the model to ensure you keep trace of the database as you have used it for the model.

When the preparation is finished, all the commands issued and their outputs can be found in "prepare.log"

When you press "Exit", the model manager will come back.

Back to main

Top