Find the right hyperparametters
In the solution of the last exercise it is said that to find a combination of parameters that achieves 98.5% acuracy took +3h50min to train the model.
Is there any tip to select good hyperparameters in advance instead of just doing trial and error?
There are a few different ways to improve the selection of your hyperparameters instead of relying on a trial and error approach:
1. One option is to use Grid search - specifying possible values for every hyperparameter and then training the model for different combinations of these possible values;
2. There's also a technique called Random search - instead of specifying possible values you use a random set of parameter values.
Other possible approaches include Bayesian optimization and using Hyperparameter tuning libraries.
Hope this helps!