In [42]:

    

import numpy as np
import scipy

import matplotlib.pyplot as pl
from sklearn.kernel_ridge import KernelRidge
import sklearn as skl

from tools.helpers import Progress


    

In [6]:

    

def mix_gauss(means, sigmas, samples):
    return [sigma * np.random.randn(samples, 2) + np.array(mean)[None, :]
            for mean, sigma in zip(means, sigmas)]


    

In [7]:

    

data = mix_gauss([(0, 0), (1, 1)], [.5, .25], 1000)


    

In [8]:

    

for x, c in zip(data, ['b', 'r']):
    pl.scatter(x[:, 0], x[:, 1], color=c)


    

In [9]:

    

X = np.reshape(data, (2000, 2))
y = np.concatenate((np.zeros(1000), np.ones(1000))).astype(int)


    

In [10]:

    

x_min, x_max = -2, 2
y_min, y_max = -2, 2
h = .1
xx, yy = np.meshgrid(np.arange(x_min, x_max + h, h),
                     np.arange(y_min, y_max + h, h))


    

In [16]:

    

pl.figure(0, figsize=(8, 8))
model = KernelRidge(alpha=.01, kernel='rbf')
model.fit(X, y)

Z = model.predict(np.c_[xx.ravel(), yy.ravel()])

# Put the result into a color plot
Z = Z.reshape(xx.shape)
pl.contourf(xx, yy, Z, alpha=0.5)

for x, c in zip(data, ['b', 'r']):
    pl.scatter(x[:, 0], x[:, 1], color=c)

pl.axis((x_min, x_max, y_min, y_max))


    

    
Out[16]:





(-2, 2, -2, 2)






    

In [22]:

    

from scipy import io

data = io.loadmat('/Users/dsuess/Downloads/Lab/moons_dataset.mat')


    

In [23]:

    

list(data)


    

    
Out[23]:





['__globals__', 'Yts', 'Ytr', '__header__', 'Xtr', '__version__', 'Xts']

In [30]:

    

X = np.concatenate((data['Xtr'], data['Xts']))
y = np.concatenate((data['Ytr'], data['Yts']))


    

In [51]:

    

pl.figure(0, figsize=(8, 8))

model = KernelRidge(alpha=.00001, kernel='rbf')
model.fit(X, y)


x_min, x_max = np.min(X[:, 0]), np.max(X[:, 0])
y_min, y_max = np.min(X[:, 1]), np.max(X[:, 1])
h = .1
xx, yy = np.meshgrid(np.arange(x_min, x_max + h, h),
                     np.arange(y_min, y_max + h, h))

Z = model.predict(np.c_[xx.ravel(), yy.ravel()])

# Put the result into a color plot
Z = Z.reshape(xx.shape)
pl.contourf(xx, yy, Z, alpha=0.5)

pl.scatter(X[:, 0], X[:, 1], color=['b' if this_y == -1 else 'r' 
                                    for this_y in y])

pl.axis((x_min, x_max, y_min, y_max))


    

    
Out[51]:





(-1.3141865820206633,
 0.80085626465716686,
 -1.6774104385590278,
 1.1606079540390035)






    

In [38]:

    

X_train, X_test, y_train, y_test = skl.cross_validation.train_test_split(
    X, y, test_size=.4, random_state=0
)


    

In [41]:

    

np.logspace(1, 2, 10)


    

    
Out[41]:





array([  10.        ,   12.91549665,   16.68100537,   21.5443469 ,
         27.82559402,   35.93813664,   46.41588834,   59.94842503,
         77.42636827,  100.        ])

In [50]:

    

alphas = np.logspace(-5, 2, 100)
models = [KernelRidge(alpha=alpha, gamma=None, kernel='rbf').fit(X_train, y_train)
          for alpha in Progress(alphas)]

pl.plot(alphas, [model.score(X_test, y_test) for model in models])
pl.plot(alphas, [model.score(X_train, y_train) for model in models])


    

    




100% (100 of 100) |#######################| Elapsed Time: 0:00:00 Time: 0:00:00






    
Out[50]:





[<matplotlib.lines.Line2D at 0x10a0a65f8>]






    

In [ ]: