In [ ]:

    

%load_ext autoreload
%autoreload 2


    

In [ ]:

    

%matplotlib inline
#%config InlineBackend.figure_format = 'svg'
import matplotlib.pyplot as plt
import seaborn as sns; sns.set() # prettify matplotlib

import numpy as np
import sklearn.gaussian_process as gp


    

In [ ]:

    

# local modules
import turbo as tb
import turbo.modules as tm
import turbo.plotting as tp
import turbo.gui.jupyter as tg


    

In [ ]:

    

# Make deterministic
np.random.seed(0)


    

In [ ]:

    

xmin, xmax = 0, 6
ymin, ymax = 0, 6
noisy = True
x = np.linspace(xmin, xmax, 100)
y = np.linspace(ymin, ymax, 100)

X, Y = np.meshgrid(x, y)

# vectorize needed for accessing noise
#@np.vectorize
def f(x, y):
    ''' from https://github.com/fmfn/BayesianOptimization/issues/18 '''
    a = np.exp(-( (x - 2)**2/0.7 + (y - 4)**2/1.2) + (x - 2)*(y - 4)/1.6 )
    b = np.exp(-( (x - 4)**2/3 + (y - 2)**2/2.) )
    c = np.exp(-( (x - 4)**2/0.5 + (y - 4)**2/0.5) + (x - 4)*(y - 4)/0.5 )
    d = np.sin(3.1415 * x)
    e = np.exp(-( (x - 5.5)**2/0.5 + (y - 5.5)**2/.5) )
    val = 2*a + b - c + 0.17 * d + 2*e
    return val + np.random.normal(0, 0.2, size=None if isinstance(x, float) else x.shape) if noisy else val

Z = f(X, Y)

tp.surface_3D(X, Y, Z)

def calc_best():
    # 2.06434770773
    x = np.linspace(2.0, 2.2, 1000)
    y = np.linspace(3.9, 4.1, 1000)
    return np.max(f(*np.meshgrid(x, y)))
best_z = calc_best()


    

In [ ]:

    

bounds = [
    ('x', xmin, xmax),
    ('y', ymin, ymax)
]

op = tb.Optimiser(f, 'max', bounds, pre_phase_trials=10, settings_preset='default')
#op.pre_phase_select = tm.LHS_selector(num_total=5)
op.surrogate = tm.SciKitGPSurrogate(model_params=dict(
    alpha = 1e-3, # larger => more noise. Default = 1e-10
    kernel = 1.0 * gp.kernels.RBF(length_scale_bounds=(1e-2, 5)),
    n_restarts_optimizer = 10,
    normalize_y = True
))
op.acquisition = tm.UCB(beta=lambda trial_num: np.log(trial_num+1))
tp.plot_acquisition_parameter_function(op.acquisition.beta, 0, 20)
rec = tb.Recorder(op)


    

In [ ]:

    

tg.OptimiserProgressBar(op)
op.run(max_trials=30)


    

In [ ]:

    

op.get_incumbent()


    

In [ ]:

    

tp.plot_error(rec, true_best=best_z);


    

In [ ]:

    

tp.plot_surrogate_hyper_params_2D(rec);


    

In [ ]:

    

tp.interactive_plot_trial_1D(rec)


    

In [ ]:

    

tp.interactive_plot_trial_2D(rec, true_objective=f);


    

In [ ]:

    

ro = tb.Optimiser(f, 'max', bounds=bounds, pre_phase_trials=float('inf'), settings_preset='random_search')
rrec = tb.Recorder(ro)


    

In [ ]:

    

tg.OptimiserProgressBar(ro)
ro.run(max_trials=100)


    

In [ ]:

    

ro.get_incumbent()


    

In [ ]:

    

tp.plot_error(rrec, true_best=best_z, log_scale=False, fig_ax=plt.subplots(figsize=(8, 3)));


    

In [ ]:

    

tp.interactive_plot_trial_1D(rrec, param='x');


    

In [ ]:

    

tp.interactive_plot_trial_2D(rrec, x_param='x', y_param='y', true_objective=f);


    

In [ ]: