does this approach (training the model so many times) makes the model computationally expensive? Because you have to train the model if you make 100 predictions for example 100 times or 25 if you make 4 predictions at a time