What is one reason not to use the same data for both your training set and your testing set?
You might not have enough data for both.
You will almost certainly underfit the model.
You will pick the wrong algorithm.
You will almost certainly overfit the model.