Many real-world functions are defined over both categorical and continuous variables. To optimize such functions, we propose a new method that encodes the categorical variables as a continuous variables, where in each category corresponds to function value.
- optimize mse of MLP regression model
- categorical variables:
- activation function: {0: 'logistic', 1: 'tanh', 2: 'relu'}
- learning rate: {0: 'constant', 1: 'invscaling', 2: 'adaptive'}
- optimization solver: {0: 'sgd', 1: 'adam'}
- early_stopping: {0: True, 1: False}
- continuous variables:
- hidden_layer_sizes: (1, 200)
- alpha: (0.0001, 1)
- tol: (0.00001, 1)
- categorical variables:
- converts a continuous variables to 17 choices
- continuous variables: