def test_np_matrix():
"""
Confirm that input validation code does not return np.matrix
"""
X = np.arange(12).reshape(3, 4)
assert_false(isinstance(as_float_array(X), np.matrix))
assert_false(isinstance(as_float_array(np.matrix(X)), np.matrix))
assert_false(isinstance(as_float_array(sp.csc_matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csr(X), np.matrix))
assert_false(isinstance(atleast2d_or_csr(np.matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csr(sp.csc_matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csc(X), np.matrix))
assert_false(isinstance(atleast2d_or_csc(np.matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csc(sp.csr_matrix(X)), np.matrix))
assert_false(isinstance(safe_asarray(X), np.matrix))
assert_false(isinstance(safe_asarray(np.matrix(X)), np.matrix))
assert_false(isinstance(safe_asarray(sp.lil_matrix(X)), np.matrix))
assert_true(atleast2d_or_csr(X, copy=False) is X)
assert_false(atleast2d_or_csr(X, copy=True) is X)
assert_true(atleast2d_or_csc(X, copy=False) is X)
assert_false(atleast2d_or_csc(X, copy=True) is X)
def test_np_matrix():
"""
Confirm that input validation code does not return np.matrix
"""
X = np.arange(12).reshape(3, 4)
assert_false(isinstance(as_float_array(X), np.matrix))
assert_false(isinstance(as_float_array(np.matrix(X)), np.matrix))
assert_false(isinstance(as_float_array(sp.csc_matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csr(X), np.matrix))
assert_false(isinstance(atleast2d_or_csr(np.matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csr(sp.csc_matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csc(X), np.matrix))
assert_false(isinstance(atleast2d_or_csc(np.matrix(X)), np.matrix))
assert_false(isinstance(atleast2d_or_csc(sp.csr_matrix(X)), np.matrix))
assert_false(isinstance(safe_asarray(X), np.matrix))
assert_false(isinstance(safe_asarray(np.matrix(X)), np.matrix))
assert_false(isinstance(safe_asarray(sp.lil_matrix(X)), np.matrix))
assert_true(atleast2d_or_csr(X, copy=False) is X)
assert_false(atleast2d_or_csr(X, copy=True) is X)
assert_true(atleast2d_or_csc(X, copy=False) is X)
assert_false(atleast2d_or_csc(X, copy=True) is X)
def entropy(*args): xy = zip(*args) # probs proba = [ float(xy.count(c)) / len(xy) for c in dict.fromkeys(list(xy)) ] safe_asarray(xy) #very pythonic list comprehension # the follwoing line is just a list comprehnsion with x = # x[numpy.isfinite(x)] having ability to filter crap out # x = x[numpy.logical_not(numpy.isnan(x))] entropy = -np.sum([ ((p * np.log2(p)) , 0 ) [ math.isnan(p * np.log2(p)) or math.isinf(p * np.log2(p)) ] for p in proba ]) assert_all_finite(entropy) return entropy
def test_safe_asarray():
"""Test that array dtype conversion works."""
# Test with sparse arrays
X = sp.csc_matrix(np.arange(4, dtype=np.float))
Y = safe_asarray(X)
assert_true(Y.dtype == np.float)
# Check that no copy has been performed
Y.data[0] = 7 # value not in original array
assert_equal(X.data[0], Y.data[0])
Y = safe_asarray(X, dtype=np.int)
assert_equal(Y.data.dtype, np.int)
# Test with dense arrays
X = np.arange(4, dtype=np.float)
Y = safe_asarray(X)
assert_true(Y.dtype == np.float)
# Check that no copy has been performed
Y[0] = 7
assert_equal(X[0], Y[0])
Y = safe_asarray(X, dtype=np.int)
assert_equal(Y.dtype, np.int)
# Non-regression: LIL and DOK used to fail for lack of a .data attribute
X = np.ones([2, 3])
safe_asarray(sp.dok_matrix(X))
safe_asarray(sp.lil_matrix(X), dtype=X.dtype)
def test_safe_asarray():
"""Test that array dtype conversion works."""
# Test with sparse arrays
X = sp.csc_matrix(np.arange(4, dtype=np.float))
Y = safe_asarray(X)
assert_true(Y.dtype == np.float)
# Check that no copy has been performed
Y.data[0] = 7 # value not in original array
assert_equal(X.data[0], Y.data[0])
Y = safe_asarray(X, dtype=np.int)
assert_equal(Y.data.dtype, np.int)
# Test with dense arrays
X = np.arange(4, dtype=np.float)
Y = safe_asarray(X)
assert_true(Y.dtype == np.float)
# Check that no copy has been performed
Y[0] = 7
assert_equal(X[0], Y[0])
Y = safe_asarray(X, dtype=np.int)
assert_equal(Y.dtype, np.int)
# Non-regression: LIL and DOK used to fail for lack of a .data attribute
X = np.ones([2, 3])
safe_asarray(sp.dok_matrix(X))
safe_asarray(sp.lil_matrix(X), dtype=X.dtype)
def fit(self, X, y):
"""
Fit LogisticRegressor model.
Parameters
----------
X : numpy array or sparse matrix of shape [n_samples,n_features]
Training data
y : numpy array of shape [n_samples, n_targets]
Target values
n_jobs : The number of jobs to use for the computation.
If -1 all CPUs are used. This will only provide speedup for
n_targets > 1 and sufficient large problems
Returns
-------
self : returns an instance of self.
"""
self.X = safe_asarray(X)
self.y = np.asarray(y)
initial_theta = zeros(self.X.shape[1])
cost = lambda theta: compute_cost(
initial_theta, self.X, self.y, C=self.C_, normf=self.norm_)
grad = lambda theta: compute_grad(
self.coef_, self.X, self.y, C=self.C_)
self.coef_ = fmin_bfgs(self.decorated_cost,
initial_theta,
fprime=self.decorated_grad)
print 'check1:', check_grad(cost, grad, initial_theta)
print 'check2:', check_grad(cost, grad, self.coef_)
# self.coef_ = fmin_bfgs(self.decorated_cost, initial_theta)
return self
def _transform_selected(X, transform, selected="all", copy=True):
"""Apply a transform function to portion of selected features
Parameters
----------
X : array-like or sparse matrix, shape=(n_samples, n_features)
Dense array or sparse matrix.
transform : callable
A callable transform(X) -> X_transformed
copy : boolean, optional
Copy X even if it could be avoided.
selected: "all" or array of indices or mask
Specify which features to apply the transform to.
Returns
-------
X : array or sparse matrix, shape=(n_samples, n_features_new)
"""
if selected == "all":
X = safe_asarray(X, copy=copy, force_all_finite=False)
return transform(X)
X = atleast2d_or_csc(X, copy=copy, force_all_finite=False)
if len(selected) == 0:
return X
n_features = X.shape[1]
ind = np.arange(n_features)
sel = np.zeros(n_features, dtype=bool)
sel[np.asarray(selected)] = True
not_sel = np.logical_not(sel)
n_selected = np.sum(sel)
if n_selected == 0:
# No features selected.
return X
elif n_selected == n_features:
# All features selected.
return transform(X)
else:
X_sel = transform(X[:, ind[sel]])
X_not_sel = X[:, ind[not_sel]]
if sparse.issparse(X_sel) or sparse.issparse(X_not_sel):
return sparse.hstack((X_sel, X_not_sel)).tocsr()
else:
return np.hstack((X_sel, X_not_sel))
def test_safe_asarray():
"""Test that array dtype conversion works."""
# Test with sparse arrays
X = sp.csc_matrix(np.arange(4, dtype=np.float))
Y = safe_asarray(X)
assert_true(Y.dtype == np.float)
# Check that no copy has been performed
Y.data[0] = 7 # value not in original array
assert_equal(X.data[0], Y.data[0])
Y = safe_asarray(X, dtype=np.int)
assert_equal(Y.data.dtype, np.int)
# Test with dense arrays
X = np.arange(4, dtype=np.float)
Y = safe_asarray(X)
assert_true(Y.dtype == np.float)
# Check that no copy has been performed
Y[0] = 7
assert_equal(X[0], Y[0])
Y = safe_asarray(X, dtype=np.int)
assert_equal(Y.dtype, np.int)
def fit(self, X, y, sample_weight=1.0, solver=None):
X = safe_asarray(X, dtype=np.float)
y = np.asarray(y, dtype=np.float)
X, y, X_mean, y_mean, X_std = \
self._center_data(X, y, self.fit_intercept,
self.normalize, self.copy_X)
self.coef_ = ridge_regression(X, y,
alpha=self.alpha,
sample_weight=sample_weight,
solver=solver,
max_iter=self.max_iter,
tol=self.tol)
self._set_intercept(X_mean, y_mean, X_std)
return self
def fit_transform(self, X, y=None):
"""Fit estimator and transform dataset.
Parameters
----------
X : array-like, shape=(n_samples, n_features)
Input data used to build forests.
Returns
-------
X_transformed: sparse matrix, shape=(n_samples, n_out)
Transformed dataset.
"""
X = safe_asarray(X)
rnd = check_random_state(self.random_state)
y = rnd.uniform(size=X.shape[0])
super(RandomTreesEmbedding, self).fit(X, y)
self.one_hot_encoder_ = OneHotEncoder()
return self.one_hot_encoder_.fit_transform(self.apply(X))
def balance_weights(y):
"""Compute sample weights such that the class distribution of y becomes
balanced.
Parameters
----------
y : array-like
Labels for the samples.
Returns
-------
weights : array-like
The sample weights.
"""
y = safe_asarray(y)
y = np.searchsorted(np.unique(y), y)
bins = np.bincount(y)
weights = 1. / bins.take(y)
weights *= bins.min()
return weights
def balance_weights(y):
"""Compute sample weights such that the class distribution of y becomes
balanced.
Parameters
----------
y : array-like
Labels for the samples.
Returns
-------
weights : array-like
The sample weights.
"""
y = safe_asarray(y)
y = np.searchsorted(np.unique(y), y)
bins = np.bincount(y)
weights = 1. / bins.take(y)
weights *= bins.min()
return weights
def fit(self, X, y, sample_weight=1.0, solver='auto'):
"""Fit Ridge regression model
Parameters
----------
X : {array-like, sparse matrix}, shape = [n_samples, n_features]
Training data
y : array-like, shape = [n_samples] or [n_samples, n_responses]
Target values
sample_weight : float or numpy array of shape [n_samples]
Individual weights for each sample
solver : {'auto', 'dense_cholesky', 'sparse_cg'}
Solver to use in the computational
routines. 'dense_cholesky' will use the standard
scipy.linalg.solve function, 'sparse_cg' will use the
conjugate gradient solver as found in
scipy.sparse.linalg.cg while 'auto' will chose the most
appropriate depending on the matrix X.
Returns
-------
self : returns an instance of self.
"""
X = safe_asarray(X, dtype=np.float)
y = np.asarray(y, dtype=np.float)
X, y, X_mean, y_mean, X_std = \
self._center_data(X, y, self.fit_intercept,
self.normalize, self.copy_X)
self.coef_ = ridge_regression(X, y, self.alpha, sample_weight,
solver, self.tol)
self._set_intercept(X_mean, y_mean, X_std)
return self
def fit(self, X, y, sample_weight=1.0, solver='auto'):
"""Fit Ridge regression model
Parameters
----------
X : {array-like, sparse matrix}, shape = [n_samples, n_features]
Training data
y : array-like, shape = [n_samples] or [n_samples, n_responses]
Target values
sample_weight : float or numpy array of shape [n_samples]
Individual weights for each sample
solver : {'auto', 'dense_cholesky', 'sparse_cg'}
Solver to use in the computational
routines. 'dense_cholesky' will use the standard
scipy.linalg.solve function, 'sparse_cg' will use the
conjugate gradient solver as found in
scipy.sparse.linalg.cg while 'auto' will chose the most
appropriate depending on the matrix X.
Returns
-------
self : returns an instance of self.
"""
X = safe_asarray(X, dtype=np.float)
y = np.asarray(y, dtype=np.float)
X, y, X_mean, y_mean, X_std = \
self._center_data(X, y, self.fit_intercept,
self.normalize, self.copy_X)
self.coef_ = ridge_regression(X, y, self.alpha, sample_weight, solver,
self.tol)
self._set_intercept(X_mean, y_mean, X_std)
return self
def ordinal_logistic_fit(X, y, max_iter=10000, verbose=False, solver='TNC'):
"""
Ordinal logistic regression or proportional odds model.
Uses scipy's optimize.fmin_slsqp solver.
Parameters
----------
X : {array, sparse matrix}, shape (n_samples, n_feaures)
Input data
y : array-like
Target values
max_iter : int
Maximum number of iterations
verbose: bool
Print convergence information
Returns
-------
w : array, shape (n_features,)
coefficients of the linear model
theta : array, shape (k,), where k is the different values of y
vector of thresholds
"""
X = utils.safe_asarray(X)
y = np.asarray(y)
# .. order input ..
idx = np.argsort(y)
idx_inv = np.zeros_like(idx)
idx_inv[idx] = np.arange(idx.size)
X = X[idx]
y = y[idx].astype(np.int)
# make them continuous and start at zero
unique_y = np.unique(y)
for i, u in enumerate(unique_y):
y[y == u] = i
unique_y = np.unique(y)
# .. utility arrays used in f_grad ..
alpha = 0.
k1 = np.sum(y == unique_y[0])
E0 = (y[:, np.newaxis] == np.unique(y)).astype(np.int)
E1 = np.roll(E0, -1, axis=-1)
E1[:, -1] = 0.
E0, E1 = map(sparse.csr_matrix, (E0.T, E1.T))
def f_obj(x0, X, y):
"""
Objective function
"""
w, theta_0 = np.split(x0, [X.shape[1]])
theta_1 = np.roll(theta_0, 1)
t0 = theta_0[y]
z = np.diff(theta_0)
Xw = X.dot(w)
a = t0 - Xw
b = t0[k1:] - X[k1:].dot(w)
c = (theta_1 - theta_0)[y][k1:]
if np.any(c > 0):
return BIG
#loss = -(c[idx] + np.log(np.exp(-c[idx]) - 1)).sum()
loss = -np.log(1 - np.exp(c)).sum()
loss += b.sum() + log_logistic(b).sum() \
+ log_logistic(a).sum() \
+ .5 * alpha * w.dot(w) - np.log(z).sum() # penalty
if np.isnan(loss):
pass
#import ipdb; ipdb.set_trace()
return loss
def f_grad(x0, X, y):
"""
Gradient of the objective function
"""
w, theta_0 = np.split(x0, [X.shape[1]])
theta_1 = np.roll(theta_0, 1)
t0 = theta_0[y]
t1 = theta_1[y]
z = np.diff(theta_0)
Xw = X.dot(w)
a = t0 - Xw
b = t0[k1:] - X[k1:].dot(w)
c = (theta_1 - theta_0)[y][k1:]
# gradient for w
phi_a = phi(a)
phi_b = phi(b)
grad_w = -X[k1:].T.dot(phi_b) + X.T.dot(1 - phi_a) + alpha * w
# gradient for theta
idx = c > 0
tmp = np.empty_like(c)
tmp[idx] = 1. / (np.exp(-c[idx]) - 1)
tmp[~idx] = np.exp(c[~idx]) / (1 - np.exp(c[~idx])) # should not need
grad_theta = (E1 - E0)[:, k1:].dot(tmp) \
+ E0[:, k1:].dot(phi_b) - E0.dot(1 - phi_a)
grad_theta[:-1] += 1. / np.diff(theta_0)
grad_theta[1:] -= 1. / np.diff(theta_0)
out = np.concatenate((grad_w, grad_theta))
return out
def f_hess(x0, s, X, y):
x0 = np.asarray(x0)
w, theta_0 = np.split(x0, [X.shape[1]])
theta_1 = np.roll(theta_0, 1)
t0 = theta_0[y]
t1 = theta_1[y]
z = np.diff(theta_0)
Xw = X.dot(w)
a = t0 - Xw
b = t0[k1:] - X[k1:].dot(w)
c = (theta_1 - theta_0)[y][k1:]
D = np.diag(phi(a) * (1 - phi(a)))
D_= np.diag(phi(b) * (1 - phi(b)))
D1 = np.diag(np.exp(-c) / (np.exp(-c) - 1) ** 2)
Ex = (E1 - E0)[:, k1:].toarray()
Ex0 = E0.toarray()
H_A = X[k1:].T.dot(D_).dot(X[k1:]) + X.T.dot(D).dot(X)
H_C = - X[k1:].T.dot(D_).dot(E0[:, k1:].T.toarray()) \
- X.T.dot(D).dot(E0.T.toarray())
H_B = Ex.dot(D1).dot(Ex.T) + Ex0[:, k1:].dot(D_).dot(Ex0[:, k1:].T) \
- Ex0.dot(D).dot(Ex0.T)
p_w = H_A.shape[0]
tmp0 = H_A.dot(s[:p_w]) + H_C.dot(s[p_w:])
tmp1 = H_C.T.dot(s[:p_w]) + H_B.dot(s[p_w:])
return np.concatenate((tmp0, tmp1))
import ipdb; ipdb.set_trace()
import pylab as pl
pl.matshow(H_B)
pl.colorbar()
pl.title('True')
import numdifftools as nd
Hess = nd.Hessian(lambda x: f_obj(x, X, y))
H = Hess(x0)
pl.matshow(H[H_A.shape[0]:, H_A.shape[0]:])
#pl.matshow()
pl.title('estimated')
pl.colorbar()
pl.show()
def grad_hess(x0, X, y):
grad = f_grad(x0, X, y)
hess = lambda x: f_hess(x0, x, X, y)
return grad, hess
x0 = np.random.randn(X.shape[1] + unique_y.size) / X.shape[1]
x0[X.shape[1]:] = np.sort(unique_y.size * np.random.rand(unique_y.size))
#print('Check grad: %s' % optimize.check_grad(f_obj, f_grad, x0, X, y))
#print(optimize.approx_fprime(x0, f_obj, 1e-6, X, y))
#print(f_grad(x0, X, y))
#print(optimize.approx_fprime(x0, f_obj, 1e-6, X, y) - f_grad(x0, X, y))
#import ipdb; ipdb.set_trace()
def callback(x0):
x0 = np.asarray(x0)
print('Check grad: %s' % optimize.check_grad(f_obj, f_grad, x0, X, y))
if verbose:
# check that gradient is correctly computed
print('OBJ: %s' % f_obj(x0, X, y))
if solver == 'TRON':
import pytron
out = pytron.minimize(f_obj, grad_hess, x0, args=(X, y))
else:
options = {'maxiter' : max_iter, 'disp': 0, 'maxfun':10000}
out = optimize.minimize(f_obj, x0, args=(X, y), method=solver,
jac=f_grad, hessp=f_hess, options=options, callback=callback)
if not out.success:
warnings.warn(out.message)
w, theta = np.split(out.x, [X.shape[1]])
return w, theta
def fit(self, X, y, sample_weight=1.0):
"""Fit Ridge regression model
Parameters
----------
X : {array-like, sparse matrix}, shape = [n_samples, n_features]
Training data
y : array-like, shape = [n_samples] or [n_samples, n_responses]
Target values
sample_weight : float or array-like of shape [n_samples]
Sample weight
Returns
-------
self : Returns self.
"""
X = safe_asarray(X, dtype=np.float)
y = np.asarray(y, dtype=np.float)
n_samples, n_features = X.shape
X, y, X_mean, y_mean, X_std = LinearModel._center_data(X, y,
self.fit_intercept, self.normalize, self.copy_X)
gcv_mode = self.gcv_mode
with_sw = len(np.shape(sample_weight))
if gcv_mode is None or gcv_mode == 'auto':
if n_features > n_samples or with_sw:
gcv_mode = 'eigen'
else:
gcv_mode = 'svd'
elif gcv_mode == "svd" and with_sw:
# FIXME non-uniform sample weights not yet supported
warnings.warn("non-uniform sample weights unsupported for svd, "
"forcing usage of eigen")
gcv_mode = 'eigen'
if gcv_mode == 'eigen':
_pre_compute = self._pre_compute
_errors = self._errors
_values = self._values
elif gcv_mode == 'svd':
# assert n_samples >= n_features
_pre_compute = self._pre_compute_svd
_errors = self._errors_svd
_values = self._values_svd
else:
raise ValueError('bad gcv_mode "%s"' % gcv_mode)
v, Q, QT_y = _pre_compute(X, y)
n_y = 1 if len(y.shape) == 1 else y.shape[1]
cv_values = np.zeros((n_samples * n_y, len(self.alphas)))
C = []
error = self.score_func is None and self.loss_func is None
for i, alpha in enumerate(self.alphas):
if error:
out, c = _errors(sample_weight * alpha, y, v, Q, QT_y)
else:
out, c = _values(sample_weight * alpha, y, v, Q, QT_y)
cv_values[:, i] = out.ravel()
C.append(c)
if error:
best = cv_values.mean(axis=0).argmin()
else:
func = self.score_func if self.score_func else self.loss_func
out = [func(y.ravel(), cv_values[:, i])
for i in range(len(self.alphas))]
best = np.argmax(out) if self.score_func else np.argmin(out)
self.alpha_ = self.alphas[best]
self.dual_coef_ = C[best]
self.coef_ = safe_sparse_dot(self.dual_coef_.T, X)
self._set_intercept(X_mean, y_mean, X_std)
if self.store_cv_values:
if len(y.shape) == 1:
cv_values_shape = n_samples, len(self.alphas)
else:
cv_values_shape = n_samples, n_y, len(self.alphas)
self.cv_values_ = cv_values.reshape(cv_values_shape)
return self
yKrn = np.zeros( shape=(1,20000) )
#yKrn = np.array( []) #, dtype=np.float64 )
print("yKrn type", type(yKrn) )
#yKrn = [ entropy(xk) for xk in xKrn] #patData.shape([0]) ] #.ravel()
#for i in np.arange(0,1,0.1):
# print i
for i in xKrn:
print i
yKrn[i]= entropy( xKrn[ i:i+clmns ] )
print xKrn[i]
i+=clmns
print("##yKrn##", shape(yKrn) )
print yKrn
np.ravel(yKrn)
safe_asarray(yKrn) #.ravel()
np.asarray_chkfinite(yKrn)
print("l885: yKrn", type(yKrn) )
#yKrn = yKrn.astype(numpy.float32, copy=False)
#safe_asarray(yKrn).ravel()#print(len(yKrn) )
np.array(yKrn,float); as_float_array(XKrn); as_float_array(yKrn)
#yKrn.astype(float)
np.float64(yKrn); np.asarray( yKrn )
#warn_if_not_float(yKrn)
XKrn = XKrn[np.logical_not(np.isnan(XKrn))]; yKrn = XKrn[np.logical_not(np.isnan(yKrn))]
assert_all_finite(XKrn); assert_all_finite(yKrn) #X_vec = np.vectorize(XKrn) #y_vec = np.vectorize(yKrn)
XKrn.ravel(), yKrn.ravel()
print("912: XKrn row, yKrn row", XKrn.shape, yKrn.shape )
def fit(self, X, y, sample_weight=1.0):
"""Fit Ridge regression model
Parameters
----------
X : {array-like, sparse matrix}, shape = [n_samples, n_features]
Training data
y : array-like, shape = [n_samples] or [n_samples, n_responses]
Target values
sample_weight : float or array-like of shape [n_samples]
Sample weight
Returns
-------
self : Returns self.
"""
X = safe_asarray(X, dtype=np.float)
y = np.asarray(y, dtype=np.float)
n_samples, n_features = X.shape
X, y, X_mean, y_mean, X_std = LinearModel._center_data(
X, y, self.fit_intercept, self.normalize, self.copy_X)
gcv_mode = self.gcv_mode
with_sw = len(np.shape(sample_weight))
if gcv_mode is None or gcv_mode == 'auto':
if n_features > n_samples or with_sw:
gcv_mode = 'eigen'
else:
gcv_mode = 'svd'
elif gcv_mode == "svd" and with_sw:
# FIXME non-uniform sample weights not yet supported
warnings.warn("non-uniform sample weights unsupported for svd, "
"forcing usage of eigen")
gcv_mode = 'eigen'
if gcv_mode == 'eigen':
_pre_compute = self._pre_compute
_errors = self._errors
_values = self._values
elif gcv_mode == 'svd':
# assert n_samples >= n_features
_pre_compute = self._pre_compute_svd
_errors = self._errors_svd
_values = self._values_svd
else:
raise ValueError('bad gcv_mode "%s"' % gcv_mode)
v, Q, QT_y = _pre_compute(X, y)
n_y = 1 if len(y.shape) == 1 else y.shape[1]
cv_values = np.zeros((n_samples * n_y, len(self.alphas)))
C = []
error = self.score_func is None and self.loss_func is None
for i, alpha in enumerate(self.alphas):
if error:
out, c = _errors(sample_weight * alpha, y, v, Q, QT_y)
else:
out, c = _values(sample_weight * alpha, y, v, Q, QT_y)
cv_values[:, i] = out.ravel()
C.append(c)
if error:
best = cv_values.mean(axis=0).argmin()
else:
func = self.score_func if self.score_func else self.loss_func
out = [
func(y.ravel(), cv_values[:, i])
for i in range(len(self.alphas))
]
best = np.argmax(out) if self.score_func else np.argmin(out)
self.best_alpha = self.alphas[best]
self.dual_coef_ = C[best]
self.coef_ = safe_sparse_dot(self.dual_coef_.T, X)
self._set_intercept(X_mean, y_mean, X_std)
if self.store_cv_values:
if len(y.shape) == 1:
cv_values_shape = n_samples, len(self.alphas)
else:
cv_values_shape = n_samples, n_y, len(self.alphas)
self.cv_values_ = cv_values.reshape(cv_values_shape)
return self
[ 0.26551972, -0.52748602],
[-0.71613885, -1.67219803],
[-1.1508754, -0.84463093],
[ 0.57204618, -0.11031915],
[ 2.32416009, -1.88625748],
[ 1.88126616, 1.54628757],
[ 0.71425139, 2.07374798],
[-0.52405951, -1.02460704],
[ 1.00767888, 0.93742334],
[-1.41500546, -1.68304715],
[ 0.23472175, 0.75076249],
[ 0.38353232, 0.52369016],
[-1.08196752, -1.00545824],
[-0.0435215, -0.9002513 ]];
#X = [[ 0.71425139, 2.07374798],
#[-0.52405951, -1.02460704],
#[ 1.00767888, 0.93742334],
#[-1.41500546, -1.68304715],
#[ 0.23472175, 0.75076249],
#[ 0.38353232, 0.52369016],
#[-1.08196752, -1.00545824],
#[-0.0435215, -0.9002513 ]];
X = safe_asarray(X);
#bandwidth = estimate_bandwidth(X, quantile=0.2, n_samples=10)
bandwidth = 0.830430037949;
ms = MeanShift(bandwidth=bandwidth, bin_seeding=True);
ms.fit(X)