code for bayesian_optimization.gif #18

junpenglao · 2016-03-22T17:07:37Z

Hi there,
Could you please provide the code and the data for generating bayesian_optimization.gif? It would be nice to have more information of this.

fmfn · 2016-03-22T21:15:14Z

I didn't write the code thinking about sharing it, therefore it is messy and definitely not worth being a part of this repo.

I can tell that the idea is exactly the same as the one carried out in this example notebook, but for a two dimensional target.

If you really want it I can send you the code by email.

junpenglao · 2016-03-22T21:53:36Z

I think this is actually a really nice demo - if you send me the code I can try to tidy it up and create a notebook.
My email is Junpeng.lao@unifr.ch

SalemAmeen · 2016-03-23T00:10:34Z

Could you please send it to me as well, s.a.ameen@edu.salford.ac.uk

fmfn · 2016-03-23T14:12:40Z

No problem, I'll do it later today.

davidenitti · 2016-03-24T12:04:01Z

I'm also interested in the plotting. can you post the plotting code here? (if you don't want to put in the repo)

fmfn · 2016-03-24T14:10:10Z

There you go. If you copy and paste this into jupyter it should work, triple line skips are cell breaks.

ps: I said it wasn't pretty.

from bayes_opt import BayesianOptimization
import numpy as np

import matplotlib.pyplot as plt
from matplotlib import cm
from matplotlib import mlab
from matplotlib import gridspec
%matplotlib inline



def unique_rows(a):
    """
    A functions to trim repeated rows that may appear when optimizing.
    This is necessary to avoid the sklearn GP object from breaking

    :param a: array to trim repeated rows from

    :return: mask of unique rows
    """

    # Sort array and kep track of where things should go back to
    order = np.lexsort(a.T)
    reorder = np.argsort(order)

    a = a[order]
    diff = np.diff(a, axis=0)
    ui = np.ones(len(a), 'bool')
    ui[1:] = (diff != 0).any(axis=1)

    return ui[reorder]



def target(x, y):
    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) )
    return 2*a + b - c + 0.17 * d + 2*e



n = 1e5
x = y = np.linspace(0, 6, 300)
X, Y = np.meshgrid(x, y)
x = X.ravel()
y = Y.ravel()
X = np.vstack([x, y]).T[:, [1, 0]]
z = target(x, y)



print(max(z))
print(min(z))



fig, axis = plt.subplots(1, 1, figsize=(14, 10))
gridsize=150

im = axis.hexbin(x, y, C=z, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=-0.9, vmax=2.1)
axis.axis([x.min(), x.max(), y.min(), y.max()])

cb = fig.colorbar(im, )
cb.set_label('Value')



def posterior(bo, X):
    ur = unique_rows(bo.X)
    bo.gp.fit(bo.X[ur], bo.Y[ur])
    mu, sigma2 = bo.gp.predict(X, eval_MSE=True)
    return mu, np.sqrt(sigma2), bo.util.utility(X, bo.gp, bo.Y.max())

def plot_2d(name=None):

    mu, s, ut = posterior(bo, X)

    fig, ax = plt.subplots(2, 2, figsize=(14, 10))
    gridsize=150

    # fig.suptitle('Bayesian Optimization in Action', fontdict={'size':30})

    # GP regression output
    ax[0][0].set_title('Gausian Process Predicted Mean', fontdict={'size':15})
    im00 = ax[0][0].hexbin(x, y, C=mu, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=-0.9, vmax=2.1)
    ax[0][0].axis([x.min(), x.max(), y.min(), y.max()])
    ax[0][0].plot(bo.X[:, 1], bo.X[:, 0], 'D', markersize=4, color='k', label='Observations')

    ax[0][1].set_title('Target Function', fontdict={'size':15})
    im10 = ax[0][1].hexbin(x, y, C=z, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=-0.9, vmax=2.1)
    ax[0][1].axis([x.min(), x.max(), y.min(), y.max()])
    ax[0][1].plot(bo.X[:, 1], bo.X[:, 0], 'D', markersize=4, color='k')


    ax[1][0].set_title('Gausian Process Variance', fontdict={'size':15})
    im01 = ax[1][0].hexbin(x, y, C=s, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=0, vmax=1)
    ax[1][0].axis([x.min(), x.max(), y.min(), y.max()])

    ax[1][1].set_title('Acquisition Function', fontdict={'size':15})
    im11 = ax[1][1].hexbin(x, y, C=ut, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=0, vmax=8)

    np.where(ut.reshape((300, 300)) == ut.max())[0]
    np.where(ut.reshape((300, 300)) == ut.max())[1]

    ax[1][1].plot([np.where(ut.reshape((300, 300)) == ut.max())[1]/50., 
                   np.where(ut.reshape((300, 300)) == ut.max())[1]/50.], 
                  [0, 6], 
                  'k-', lw=2, color='k')

    ax[1][1].plot([0, 6],
                  [np.where(ut.reshape((300, 300)) == ut.max())[0]/50., 
                   np.where(ut.reshape((300, 300)) == ut.max())[0]/50.], 
                  'k-', lw=2, color='k')

    ax[1][1].axis([x.min(), x.max(), y.min(), y.max()])

    for im, axis in zip([im00, im10, im01, im11], ax.flatten()):
        cb = fig.colorbar(im, ax=axis)
        # cb.set_label('Value')

    if name is None:
        name = '_'

    plt.tight_layout()

    # Save or show figure?
    # fig.savefig('bo_eg_' + name + '.png')
    plt.show()
    plt.close(fig)



bo = BayesianOptimization(target, {'x': (0, 6), 'y': (0, 6)})



gp_params = {'corr': 'absolute_exponential', 'nugget': 1e-9}
bo.maximize(init_points=5, n_iter=0, acq='ucb', kappa=10, **gp_params)



plot_2d("{:03}".format(len(bo.X)))



# Turn interactive plotting off
plt.ioff()

for i in range(50):
    bo.maximize(init_points=0, n_iter=1, acq='ucb', kappa=10, **gp_params)
    plot_2d("{:03}".format(len(bo.X)))

junpenglao · 2016-03-24T16:24:36Z

Thanks a lot ( I dont think it's messy at all ;-) )

flavio-martinelli · 2020-05-17T19:09:36Z

This example code does not work anymore with the new version, attributes such as bo.X and bo.Y do not exist anymore in the BayesianOptimization class.

Is there any quick change to get this code working again? Thank you, that'd be great!

cornerfarmer · 2021-06-24T11:56:09Z

This code works for me with the newest version:

from bayes_opt import BayesianOptimization, UtilityFunction
import numpy as np

import matplotlib.pyplot as plt
from matplotlib import cm
from matplotlib import mlab
from matplotlib import gridspec


def unique_rows(a):
    """
    A functions to trim repeated rows that may appear when optimizing.
    This is necessary to avoid the sklearn GP object from breaking

    :param a: array to trim repeated rows from

    :return: mask of unique rows
    """

    # Sort array and kep track of where things should go back to
    order = np.lexsort(a.T)
    reorder = np.argsort(order)

    a = a[order]
    diff = np.diff(a, axis=0)
    ui = np.ones(len(a), 'bool')
    ui[1:] = (diff != 0).any(axis=1)

    return ui[reorder]



def target(x, y):
    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) )
    return 2*a + b - c + 0.17 * d + 2*e



n = 1e5
x = y = np.linspace(0, 6, 300)
X, Y = np.meshgrid(x, y)
x = X.ravel()
y = Y.ravel()
X = np.vstack([x, y]).T[:, [1, 0]]
z = target(y, x)



print(max(z))
print(min(z))



fig, axis = plt.subplots(1, 1, figsize=(14, 10))
gridsize=150

im = axis.hexbin(x, y, C=z, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=-0.9, vmax=2.1)
axis.axis([x.min(), x.max(), y.min(), y.max()])

cb = fig.colorbar(im, )
cb.set_label('Value')

util = UtilityFunction(kind='ucb',
                       kappa=10,
                       xi=0.0,
                       kappa_decay=1,
                       kappa_decay_delay=0)


def posterior(bo, X):
    ur = unique_rows(bo._space.params)
    bo._gp.fit(bo._space.params[ur], bo._space.target[ur])
    mu, sigma2 = bo._gp.predict(X, return_std=True)
    return mu, np.sqrt(sigma2), util.utility(X, bo._gp, bo._space.target.max())

def plot_2d(name=None):

    mu, s, ut = posterior(bo, X)
    #self._space.params, self._space.target
    fig, ax = plt.subplots(2, 2, figsize=(14, 10))
    gridsize=150

    # fig.suptitle('Bayesian Optimization in Action', fontdict={'size':30})

    # GP regression output
    ax[0][0].set_title('Gausian Process Predicted Mean', fontdict={'size':15})
    im00 = ax[0][0].hexbin(x, y, C=mu, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=-0.9, vmax=2.1)
    ax[0][0].axis([x.min(), x.max(), y.min(), y.max()])
    ax[0][0].plot(bo._space.params[:, 1], bo._space.params[:, 0], 'D', markersize=4, color='k', label='Observations')

    ax[0][1].set_title('Target Function', fontdict={'size':15})
    im10 = ax[0][1].hexbin(x, y, C=z, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=-0.9, vmax=2.1)
    ax[0][1].axis([x.min(), x.max(), y.min(), y.max()])
    ax[0][1].plot(bo._space.params[:, 1], bo._space.params[:, 0], 'D', markersize=4, color='k')


    ax[1][0].set_title('Gausian Process Variance', fontdict={'size':15})
    im01 = ax[1][0].hexbin(x, y, C=s, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=0, vmax=1)
    ax[1][0].axis([x.min(), x.max(), y.min(), y.max()])

    ax[1][1].set_title('Acquisition Function', fontdict={'size':15})
    im11 = ax[1][1].hexbin(x, y, C=ut, gridsize=gridsize, cmap=cm.jet, bins=None, vmin=0, vmax=8)

    np.where(ut.reshape((300, 300)) == ut.max())[0]
    np.where(ut.reshape((300, 300)) == ut.max())[1]

    ax[1][1].plot([np.where(ut.reshape((300, 300)) == ut.max())[1]/50.,
                   np.where(ut.reshape((300, 300)) == ut.max())[1]/50.],
                  [0, 6],
                  'k-', lw=2, color='k')

    ax[1][1].plot([0, 6],
                  [np.where(ut.reshape((300, 300)) == ut.max())[0]/50.,
                   np.where(ut.reshape((300, 300)) == ut.max())[0]/50.],
                  'k-', lw=2, color='k')

    ax[1][1].axis([x.min(), x.max(), y.min(), y.max()])

    for im, axis in zip([im00, im10, im01, im11], ax.flatten()):
        cb = fig.colorbar(im, ax=axis)
        # cb.set_label('Value')

    if name is None:
        name = '_'

    plt.tight_layout()

    # Save or show figure?
    # fig.savefig('bo_eg_' + name + '.png')
    plt.show()
    plt.close(fig)



bo = BayesianOptimization(target, {'x': (0, 6), 'y': (0, 6)})


bo.maximize(init_points=5, n_iter=0, acq='ucb', kappa=10)



plot_2d("{:03}".format(len(bo._space.params)))



# Turn interactive plotting off
plt.ioff()

for i in range(50):
    bo.maximize(init_points=0, n_iter=1, acq='ucb', kappa=10)
    plot_2d("{:03}".format(len(bo._space.params)))

Benjizhang · 2022-06-13T09:12:46Z

Hi, it should be noted that there is a bug in line 58(if paste the codes into any IDE), where it is not correct to give z = target(y, x). It should be z = target(x,y). (the order of x and y when calling "target" function)

GiacomoDG96 · 2024-04-08T15:46:46Z

Hi, the code does not work anymore and gives back the following error:

Exception:
Passing acquisition function parameters or gaussian process parameters to maximize
is no longer supported. Instead,please use the "set_gp_params" method to set
the gp params, and pass an instance of bayes_opt.util.UtilityFunction
using the acquisition_function argument

till-m · 2024-04-10T07:37:53Z

Hi @GiacomoDG96,

please note that we're not maintaining this code.
That being said, replacing

bo = BayesianOptimization(target, {'x': (0, 6), 'y': (0, 6)})


bo.maximize(init_points=5, n_iter=0, acq='ucb', kappa=10)


plot_2d("{:03}".format(len(bo._space.params)))



# Turn interactive plotting off
plt.ioff()

for i in range(50):
    bo.maximize(init_points=0, n_iter=1, acq='ucb', kappa=10)
    plot_2d("{:03}".format(len(bo._space.params)))

with

from bayes_opt import UtilityFunction

utility = UtilityFunction(kind="ucb", kappa=10, xi=0.0)
bo = BayesianOptimization(target, {'x': (0, 6), 'y': (0, 6)})

bo.maximize(init_points=5, n_iter=0, acquisition_function=utility)

plot_2d("{:03}".format(len(bo._space.params)))

# Turn interactive plotting off
plt.ioff()

for i in range(50):
    bo.maximize(init_points=0, n_iter=1, acquisition_function=utility)
    plot_2d("{:03}".format(len(bo._space.params)))

should work, or at least get you close to running code.

fmfn closed this as completed Mar 24, 2016

fmfn mentioned this issue Apr 13, 2017

2D plot #49

Closed

fmfn mentioned this issue Sep 17, 2017

Example gif visualization #62

Closed

fmfn mentioned this issue May 1, 2020

How to plot 2d map of bayesian opimization in action example? #210

Closed

bwheelz36 mentioned this issue Jun 9, 2022

About the BO in action #324

Closed

till-m mentioned this issue Apr 17, 2024

Visualization #474

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

code for bayesian_optimization.gif #18

code for bayesian_optimization.gif #18

junpenglao commented Mar 22, 2016

fmfn commented Mar 22, 2016

junpenglao commented Mar 22, 2016

SalemAmeen commented Mar 23, 2016

fmfn commented Mar 23, 2016

davidenitti commented Mar 24, 2016

fmfn commented Mar 24, 2016

junpenglao commented Mar 24, 2016

flavio-martinelli commented May 17, 2020 •

edited

cornerfarmer commented Jun 24, 2021

Benjizhang commented Jun 13, 2022 •

edited

GiacomoDG96 commented Apr 8, 2024

till-m commented Apr 10, 2024

code for bayesian_optimization.gif #18

code for bayesian_optimization.gif #18

Comments

junpenglao commented Mar 22, 2016

fmfn commented Mar 22, 2016

junpenglao commented Mar 22, 2016

SalemAmeen commented Mar 23, 2016

fmfn commented Mar 23, 2016

davidenitti commented Mar 24, 2016

fmfn commented Mar 24, 2016

junpenglao commented Mar 24, 2016

flavio-martinelli commented May 17, 2020 • edited

cornerfarmer commented Jun 24, 2021

Benjizhang commented Jun 13, 2022 • edited

GiacomoDG96 commented Apr 8, 2024

till-m commented Apr 10, 2024

flavio-martinelli commented May 17, 2020 •

edited

Benjizhang commented Jun 13, 2022 •

edited