Setting the Number of Training Samples

By default, all measuring points are used as training sample. You can reduce the number of training samples, e.g. to have test data available for a quality assessment of the model or – in case of a very high number of measuring points – to reduce the duration of the model training.

Proceed as follows.

1. In the ISP view, select Data > Set Number Training Samples.

   The "Training Samples" window opens. The current number of training samples is displayed.

2. Do one of the following:

   • In the input field, enter the desired number of training samples.

     You have to enter an integer number in [2 .. <n_measuringPoints>]. A sufficiently large number of measuring points should naturally remain for a successful model training.

   • Click Select All to use all measuring points as training data.

3. Activate the option for your desired method to select the subsample.

   Available methods: Random Selection and Farthest First.

   Farthest First selects a space-filling subset of measurement points. The algorithm works as follows:

   • The first point is selected at random.

   • The second point is the one farthest from the first point.

   • The third point is the one farthest from the first and second point.

   • The fourth point is the one farthest from the first, second and third point.

   • ... (continued until the specified number of training data is selected)

4. Click OK.

   The "Training Samples" window closes. The number of measurement points is updated in the bottom-right corner of the ISP view.

   

See also

ASCMO-STATIC Main Window