To successfully train a model, you need to come up with the right set of property values for an algorithm. Most of the time, doing this is not an easy task. First, you need to have a clear understanding of the algorithm and the mathematics behind it. Second, you have to run an experiment many times, trying out many combinations of parameters for an algorithm. At times, this can be very time consuming and daunting.

For example, in the same preceding example, what should be the right value for L2 regularization weight? It is used to reduce overfitting of the model. A model overfits when it performs well on a training dataset, but performs badly on any new dataset. By reducing overfitting, you generalize the model. However, the problem here is that you have to manually adjust this L2 regularization weight, which can be done by trying different values, running the experiment many times, and evaluating its performance in each run...