First, put only 900 samples into your predictor table (not all 1000). Maybe use randsample(). So you'll have a tPredictor table with 900 rows, and a tTest table with 100 rows. Now you have 900 observations for training and you want 750 of those for training the model and, 150 of them for validation.On the Start New Session window, Select the "Holdout Validation" radio button and pick 150/900 which is 16 or 17%. I presume you know what to do after that.