==================================== Bayesian Optimization Configuration ==================================== Given below are the attributes needed for performing Bayesian Optimization: -------------------------------- Bayesian Optimization Attributes -------------------------------- - :bglb:`n_trials` * Number of repeats of the Bayesian Optimization process to gain convergence information of the performance of different models. * Every trial is set to a different random seed which results in different sets of initial data used to initiate training of surrogate models before starting off the Bayesian Optimization iterations. * Values = Integer value greater than 0 * Default value = 50 - :bglb:`n_update` * Number of Bayesian Optimization iterations at which we want to update our surrogate models. * Values = Integer value greater than 1 * Default value = 10 - :bglb:`random_seed` * Choice for choosing the random seed to initialize the initial training dataset. * *iteration* for trial number or *time* for CPU clock time. * Values = iteration, time * Default value = iteration ------------------------------ GP Model Selection Attributes ------------------------------ - :bglb:`GP_0_BO` * Set to True if we want to run Bayesian Optimization for a Gaussian Process surrogate with 0-prior mean * Values = :lime:`True` or :red:`False` * Default value = :lime:`True` - :bglb:`GP_C_BO` * Set to True if we want to run Bayesian Optimization for a Gaussian Process surrogate with constant function prior mean * Values = :lime:`True` or :red:`False` * Default value = :lime:`True` - :bglb:`GP_L_BO` * Set to True if we want to run Bayesian Optimization for a Gaussian Process surrogate with linear function prior mean * Values = :lime:`True` or :red:`False` * Default value = :lime:`True` - :bglb:`GP_NN_BO` * Set to True if we want to run Bayesian Optimization for a Gaussian Process surrogate with Neural Network prior mean * Values = :lime:`True` or :red:`False` * Default value = :lime:`True` ------------------- GP Model Attributes ------------------- - :bglb:`kernel` * Choice of the kernel for the Gaussian Process models. * Values = Matern, RBF * Default value = Matern ^^^^^^^^^^^^^^^ Learning rates ^^^^^^^^^^^^^^^ - :bglb:`learning_rate_gp0` * Learning rate for training the Gaussian Process with 0-prior mean model * Values = Floating point in (0,1) * Default value = 0.1 - :bglb:`learning_rate_gpC` * Learning rate for training the Gaussian Process with constant function prior mean model * Values = Floating point in (0,1) * Default value = 0.1 - :bglb:`learning_rate_gpL` * Learning rate for training the Gaussian Process with linear function prior mean model * Values = Floating point in (0,1) * Default value = 0.1 - :bglb:`learning_rate_gpNN` * Learning rate for training the Gaussian Process with Neural Network prior mean model * Values = Floating point in (0,1) * Default value = 0.01 ^^^^^^^ Epochs ^^^^^^^ - :bglb:`epochs_gp0` * Number of training epochs for training the Gaussian Process with 0-prior mean model * Values = Integer value greater than 0 * Default value = 100 - :bglb:`epochs_gpC` * Number of training epochs for training the Gaussian Process with constant function prior mean model * Values = Integer value greater than 0 * Default value = 100 - :bglb:`epochs_gpL` * Number of training epochs for training the Gaussian Process with linear function prior mean model * Values = Integer value greater than 0 * Default value = 100 - :bglb:`epochs_gpNN` * Number of training epochs for training the Gaussian Process with Neural Network prior mean model * Values = Integer value greater than 0 * Default value = 500 -------------------------- Neural Network Attributes -------------------------- - :bglb:`learning_rate` * Learning rate for training the Neural Network model * Values = Floating point in (0,1) * Default value = 1e-6 - :bglb:`batch_size_nn` * Minibatch size to split up the data per epoch during the Neural Network training * Values = Integer value greater than 0 * Default value = 5 - :bglb:`epochs` * Number of training epochs for training the Neural Network model * Values = Integer value greater than 0 * Default value = 3000 - :bglb:`l1` * L1 regularization parameter for Neural Network weights * Values = Floating point in (0,1) * Default value = 0.1 - :bglb:`l2` * L2 regularization parameter for Neural Network weights * Values = Floating point in (0,1) * Default value = 0.4 - :bglb:`num_nodes` * Number of nodes in the first hidden layer of the Neural Network * Values = Integer value greater than 0 * Default value = 50 - :bglb:`saveModel_filename` * File name that is used to the save the Neural Network model to be used later when fitting the Gaussian Process Neural Network model. * Values = Any string literal * Default value = connor_90p_chanNNarch.pt - :bglb:`saveModel_folder` * Folder name where the abovementioned *saveModel_filename* is stored * Values = Any string literal * Default value = NN_savedmodels_BO/ ------------------------ Acquisition Attributes ------------------------ - :bglb:`standardize_data` * Set to True if we want to normalize/standardize the input dataset. * Values = :lime:`True` or :red:`False` * Default value = :lime:`True` - :bglb:`save_output` * Set to true if we want to save the output of the acquisition function * This is equivalent to saving the output from the Bayesian Optimization iterations. * Values = :lime:`True` or :red:`False` * Default value = :lime:`True` - :bglb:`n_batch` * Parameter for training of the acquisition function * Values = Integer value greater than 0 * Default value = 100 - :bglb:`num_restarts` * Parameter for training of the acquisition function * Values = Integer value greater than 0 * Default value = 10 - :bglb:`raw_samples` * Parameter for training of the acquisition function * Values = Integer value greater than 0 * Default value = 512