diff --git a/scikits/learn/bayes/bayes.py b/scikits/learn/bayes/bayes.py
index 01e21fe86b590af5c523c891ff6f075317b223c8..d75492b413584f4add3c27d6cbdc06679c5b4ff8 100755
--- a/scikits/learn/bayes/bayes.py
+++ b/scikits/learn/bayes/bayes.py
@@ -25,7 +25,7 @@ def fast_logdet(A):
-def bayesian_ridge( X , Y, step_th=300,th_w = 1.e-6,ll_bool=True) :
+def bayesian_ridge( X , Y, step_th=300,th_w = 1.e-12,ll_bool=False) :
"""
Bayesian ridge regression. Optimize the regularization parameter alpha
within a simple bayesian framework (MAP).
@@ -39,15 +39,17 @@ def bayesian_ridge( X , Y, step_th=300,th_w = 1.e-6,ll_bool=True) :
target
step_th : int (defaut is 300)
Stop the algorithm after a given number of steps.
- th_w : float (defaut is 1.e-6)
+ th_w : float (defaut is 1.e-12)
Stop the algorithm if w has converged.
- ll_bool : boolean (default is True).
+ ll_bool : boolean (default is False).
If True, compute the log-likelihood at each step of the model.
Returns
-------
w : numpy array of shape (dim)
mean of the weights distribution.
+ log_likelihood : list of float of size steps.
+ Compute (if asked) the log-likelihood of the model.
Examples
--------
@@ -64,7 +66,9 @@ def bayesian_ridge( X , Y, step_th=300,th_w = 1.e-6,ll_bool=True) :
beta = 1./np.var(Y)
alpha = 1.0
- log_likelihood = []
+ log_likelihood = None
+ if ll_bool :
+ log_likelihood = []
has_converged = False
gram = np.dot(X.T, X)
ones = np.eye(gram.shape[1])
@@ -89,7 +93,6 @@ def bayesian_ridge( X , Y, step_th=300,th_w = 1.e-6,ll_bool=True) :
step_th -= 1
-
# convergence : compare w
has_converged = (np.sum(np.abs(w-old_w))<th_w)
old_w = w
@@ -104,11 +107,11 @@ def bayesian_ridge( X , Y, step_th=300,th_w = 1.e-6,ll_bool=True) :
ll -= X.shape[0]*np.log(2*np.pi)
log_likelihood.append(ll)
- return w,log_likelihood[1:]
+ return w,log_likelihood
-class BayessianRegression(object):
+class BayesianRegression(object):
"""
Encapsulate various bayesian regression algorithms
"""
@@ -119,9 +122,9 @@ class BayessianRegression(object):
def fit(self, X, Y):
X = np.asanyarray(X, dtype=np.float)
Y = np.asanyarray(Y, dtype=np.float)
- self.w = bayesian_ridge(X, Y)
+ self.w,self.log_likelihood = bayesian_ridge(X, Y)
def predict(self, T):
- T = np.asanyarray(T)
- # I think this is wrong
return np.dot(T, self.w)
+
+
diff --git a/scikits/learn/bayes/tests/test_bayes.py b/scikits/learn/bayes/tests/test_bayes.py
index 64879acde3f175b6656d3c99fd24979511e37671..5c1eaa7511e95839ad2301c19d65ac629af69339 100644
--- a/scikits/learn/bayes/tests/test_bayes.py
+++ b/scikits/learn/bayes/tests/test_bayes.py
@@ -1,19 +1,33 @@
import numpy as np
-from scikits.learn.bayes.bayes import bayesian_ridge, BayessianRegression
+from scikits.learn.bayes.bayes import bayesian_ridge, BayesianRegression
from numpy.testing import assert_array_almost_equal
-
-X = np.array([[1], [2]])
-Y = np.array([1, 2])
+from scikits.learn.datasets.samples_generator import linear
def test_toy():
- w = bayesian_ridge(X, Y)
+ X = np.array([[1], [2]])
+ Y = np.array([1, 2])
+ w ,log_likelihood = bayesian_ridge(X, Y)
assert_array_almost_equal(w, [1])
def test_toy_object():
"""
- Test BayessianRegression classifier
+ Test BayesianRegression classifier
"""
- clf = BayessianRegression()
+ X = np.array([[1], [2]])
+ Y = np.array([1, 2])
+ clf = BayesianRegression()
clf.fit(X, Y)
Test = [[1], [2], [3], [4]]
assert_array_almost_equal(clf.predict(Test), [1, 2, 3, 4]) # identity
+
+def test_simu_object():
+ """
+ Test BayesianRegression classifier with simulated linear data
+ """
+ X,Y = linear.sparse_uncorrelated(nb_samples=100,nb_features=10)
+ clf = BayesianRegression()
+ clf.fit(X, Y)
+ Xtest,Ytest = linear.sparse_uncorrelated(nb_samples=100,nb_features=10)
+ mse = np.mean((clf.predict(Xtest)-Ytest)**2)
+ assert(mse<2.)
+
diff --git a/scikits/learn/datasets/samples_generator/linear.py b/scikits/learn/datasets/samples_generator/linear.py
index 72fc93c5369c97451965947bd405c4ead6dfc5e6..cf649ea5a06a19fd4d5c5ab0865eff8214a4ddda 100755
--- a/scikits/learn/datasets/samples_generator/linear.py
+++ b/scikits/learn/datasets/samples_generator/linear.py
@@ -1,12 +1,12 @@
import numpy as np
-
+import numpy.random as nr
def sparse_uncorrelated(nb_samples=100,nb_features=10):
"""
Function creating simulated data with sparse uncorrelated design.
(cf.Celeux et al. 2009, Bayesian regularization in regression)
X = NR.normal(0,1)
- Y = NR.normal(2+X[:,2]+2*X[:,3]-2*X[:,6]-1.5*X[:,7])
+ Y = NR.normal(X[:,2]+2*X[:,3]-2*X[:,6]-1.5*X[:,7])
The number of features is at least 10.
Parameters
@@ -23,6 +23,6 @@ def sparse_uncorrelated(nb_samples=100,nb_features=10):
Y : numpy array of shape (nb_samples)
"""
X = nr.normal(loc=0,scale=1,size=(nb_samples,nb_features))
- Y = nr.normal(loc=2+x[:,2]+2*x[:,3]-2*x[:,6]-1.5*x[:,7],
+ Y = nr.normal(loc=X[:,2]+2*X[:,3]-2*X[:,6]-1.5*X[:,7],
scale=np.ones(nb_samples))
return X,Y
\ No newline at end of file