scikit-learn/sklearn/linear_model/cd_fast.pyx

656 lines
22 KiB
Cython

# Author: Alexandre Gramfort <alexandre.gramfort@inria.fr>
# Fabian Pedregosa <fabian.pedregosa@inria.fr>
# Olivier Grisel <olivier.grisel@ensta.org>
# Alexis Mignon <alexis.mignon@gmail.com>
#
# License: BSD Style.
from libc.math cimport fabs, sqrt
cimport numpy as np
import numpy as np
import numpy.linalg as linalg
cimport cython
from cpython cimport bool
import warnings
np.import_array()
cdef inline double fmax(double x, double y):
if x > y:
return x
return y
cdef inline double fsign(double f):
if f == 0:
return 0
elif f > 0:
return 1.0
else:
return -1.0
cdef extern from "cblas.h":
enum CBLAS_ORDER:
CblasRowMajor=101
CblasColMajor=102
enum CBLAS_TRANSPOSE:
CblasNoTrans=111
CblasTrans=112
CblasConjTrans=113
AtlasConj=114
void daxpy "cblas_daxpy"(int N, double alpha, double *X, int incX,
double *Y, int incY)
double ddot "cblas_ddot"(int N, double *X, int incX, double *Y, int incY)
void dger "cblas_dger"(CBLAS_ORDER Order, int M, int N, double alpha,
double *X, int incX, double *Y, int incY, double *A, int lda)
void dgemv "cblas_dgemv"(CBLAS_ORDER Order,
CBLAS_TRANSPOSE TransA, int M, int N,
double alpha, double *A, int lda,
double *X, int incX, double beta,
double *Y, int incY)
double dnrm2 "cblas_dnrm2"(int N, double *X, int incX)
void dcopy "cblas_dcopy"(int N, double *X, int incX, double *Y, int incY)
void dscal "cblas_dscal"(int N, double alpha, double *X, int incX)
ctypedef np.float64_t DOUBLE
ctypedef np.int32_t INTEGER
@cython.boundscheck(False)
@cython.wraparound(False)
@cython.cdivision(True)
def sparse_std(unsigned int n_samples,
unsigned int n_features,
np.ndarray[DOUBLE, ndim=1] X_data,
np.ndarray[INTEGER, ndim=1] X_indices,
np.ndarray[INTEGER, ndim=1] X_indptr,
np.ndarray[DOUBLE, ndim=1] X_mean=None):
cdef unsigned int ii
cdef unsigned int jj
cdef unsigned int nnz_ii
cdef double X_sum_ii
cdef double X_mean_ii
cdef double diff
cdef double X_std_ii
cdef np.ndarray[DOUBLE, ndim = 1] X_std = np.zeros(n_features, np.float64)
if X_mean is None:
X_mean = np.zeros(n_features, np.float64)
for ii in xrange(n_features):
# Computes the mean
X_sum_ii = 0.0
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
X_sum_ii += X_data[jj]
X_mean[ii] = X_sum_ii / n_samples
for ii in xrange(n_features):
X_mean_ii = X_mean[ii]
X_sum_ii = 0.0
nnz_ii = 0
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
diff = X_data[jj] - X_mean_ii
X_sum_ii += diff * diff
nnz_ii += 1
X_std[ii] = (X_sum_ii + (n_samples - nnz_ii) * X_mean_ii * X_mean_ii)
return np.sqrt(X_std)
@cython.boundscheck(False)
@cython.wraparound(False)
@cython.cdivision(True)
def enet_coordinate_descent(np.ndarray[DOUBLE, ndim=1] w,
double alpha, double beta,
np.ndarray[DOUBLE, ndim=2] X,
np.ndarray[DOUBLE, ndim=1] y,
int max_iter, double tol, bool positive=False):
"""Cython version of the coordinate descent algorithm
for Elastic-Net regression
We minimize
1 norm(y - X w, 2)^2 + alpha norm(w, 1) + beta norm(w, 2)^2
- ----
2 2
"""
# get the data information into easy vars
cdef unsigned int n_samples = X.shape[0]
cdef unsigned int n_features = X.shape[1]
# compute norms of the columns of X
cdef np.ndarray[DOUBLE, ndim=1] norm_cols_X = (X**2).sum(axis=0)
# initial value of the residuals
cdef np.ndarray[DOUBLE, ndim=1] R
cdef double tmp
cdef double w_ii
cdef double d_w_max
cdef double w_max
cdef double d_w_ii
cdef double gap = tol + 1.0
cdef double d_w_tol = tol
cdef unsigned int ii
cdef unsigned int n_iter
if alpha == 0:
warnings.warn("Coordinate descent with alpha=0 may lead to unexpected"
" results and is discouraged.")
R = y - np.dot(X, w)
tol = tol * np.dot(y, y)
for n_iter in range(max_iter):
w_max = 0.0
d_w_max = 0.0
for ii in xrange(n_features): # Loop over coordinates
if norm_cols_X[ii] == 0.0:
continue
w_ii = w[ii] # Store previous value
if w_ii != 0.0:
# R += w_ii * X[:,ii]
daxpy(n_samples, w_ii,
<DOUBLE*>(X.data + ii * n_samples * sizeof(DOUBLE)), 1,
<DOUBLE*>R.data, 1)
# tmp = (X[:,ii]*R).sum()
tmp = ddot(n_samples,
<DOUBLE*>(X.data + ii * n_samples * sizeof(DOUBLE)), 1,
<DOUBLE*>R.data, 1)
if positive and tmp < 0:
w[ii] = 0.0
else:
w[ii] = fsign(tmp) * fmax(fabs(tmp) - alpha, 0) \
/ (norm_cols_X[ii] + beta)
if w[ii] != 0.0:
# R -= w[ii] * X[:,ii] # Update residual
daxpy(n_samples, -w[ii],
<DOUBLE*>(X.data + ii * n_samples * sizeof(DOUBLE)), 1,
<DOUBLE*>R.data, 1)
# update the maximum absolute coefficient update
d_w_ii = fabs(w[ii] - w_ii)
if d_w_ii > d_w_max:
d_w_max = d_w_ii
if fabs(w[ii]) > w_max:
w_max = fabs(w[ii])
if w_max == 0.0 or d_w_max / w_max < d_w_tol or n_iter == max_iter - 1:
# the biggest coordinate update of this iteration was smaller than
# the tolerance: check the duality gap as ultimate stopping
# criterion
XtA = np.dot(X.T, R) - beta * w
if positive:
dual_norm_XtA = np.max(XtA)
else:
dual_norm_XtA = linalg.norm(XtA, np.inf)
R_norm2 = np.dot(R, R)
w_norm2 = np.dot(w, w)
if (dual_norm_XtA > alpha):
const = alpha / dual_norm_XtA
A_norm2 = R_norm2 * (const**2)
gap = 0.5 * (R_norm2 + A_norm2)
else:
const = 1.0
gap = R_norm2
gap += alpha * linalg.norm(w, 1) - const * np.dot(R.T, y) + \
0.5 * beta * (1 + const ** 2) * (w_norm2)
if gap < tol:
# return if we reached desired tolerance
break
return w, gap, tol
@cython.boundscheck(False)
@cython.wraparound(False)
@cython.cdivision(True)
def sparse_enet_coordinate_descent(np.ndarray[DOUBLE, ndim=1] w,
double alpha, double beta,
np.ndarray[DOUBLE, ndim=1] X_data,
np.ndarray[INTEGER, ndim=1] X_indices,
np.ndarray[INTEGER, ndim=1] X_indptr,
np.ndarray[DOUBLE, ndim=1] y,
np.ndarray[DOUBLE, ndim=1] X_mean,
int max_iter, double tol, bint positive=False):
"""Cython version of the coordinate descent algorithm for Elastic-Net
We minimize:
1 norm(y - X w, 2)^2 + alpha norm(w, 1) + beta norm(w, 2)^2
- ----
2 2
"""
# get the data information into easy vars
cdef unsigned int n_samples = y.shape[0]
cdef unsigned int n_features = w.shape[0]
# compute norms of the columns of X
cdef unsigned int ii
cdef np.ndarray[DOUBLE, ndim = 1] norm_cols_X = np.zeros(n_features,
np.float64)
for ii in xrange(n_features):
norm_cols_X[ii] = ((X_data[X_indptr[ii]:X_indptr[ii + 1]] - \
X_mean[ii]) ** 2).sum() + \
(n_samples - X_indptr[ii + 1] + X_indptr[ii]) * X_mean[ii] ** 2
# initial value of the residuals
cdef np.ndarray[DOUBLE, ndim = 1] R
cdef double tmp
cdef double w_ii
cdef double d_w_max
cdef double w_max
cdef double d_w_ii
cdef double X_mean_ii
cdef double R_sum
cdef double gap = tol + 1.0
cdef double d_w_tol = tol
cdef unsigned int jj
cdef unsigned int n_iter
cdef bint center = (X_mean != 0).any()
# initialize the residuals
R = y.copy()
for ii in xrange(n_features):
# sparse X column / dense w dot product
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
R[X_indices[jj]] -= X_data[jj] * w[ii]
if center:
R += X_mean[ii] * w[ii]
tol = tol * np.dot(y, y)
for n_iter in range(max_iter):
w_max = 0.0
d_w_max = 0.0
for ii in xrange(n_features): # Loop over coordinates
if norm_cols_X[ii] == 0.0:
continue
w_ii = w[ii] # Store previous value
X_mean_ii = X_mean[ii]
if w_ii != 0.0:
# R += w_ii * X[:,ii]
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
R[X_indices[jj]] += X_data[jj] * w_ii
if center:
for jj in xrange(n_samples):
R[jj] -= X_mean_ii * w[ii]
# tmp = (X[:,ii] * R).sum()
tmp = 0.0
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
tmp += R[X_indices[jj]] * X_data[jj]
if center:
R_sum = 0.0
for jj in xrange(n_samples):
R_sum += R[jj]
tmp -= R_sum * X_mean_ii
if positive and tmp < 0.0:
w[ii] = 0.0
else:
w[ii] = fsign(tmp) * fmax(fabs(tmp) - alpha, 0) \
/ (norm_cols_X[ii] + beta)
if w[ii] != 0.0:
# R -= w[ii] * X[:,ii] # Update residual
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
R[X_indices[jj]] -= X_data[jj] * w[ii]
if center:
for jj in xrange(n_samples):
R[jj] += X_mean_ii * w[ii]
# update the maximum absolute coefficient update
d_w_ii = fabs(w[ii] - w_ii)
if d_w_ii > d_w_max:
d_w_max = d_w_ii
if w[ii] > w_max:
w_max = w[ii]
if w_max == 0.0 or d_w_max / w_max < d_w_tol or n_iter == max_iter - 1:
# the biggest coordinate update of this iteration was smaller than
# the tolerance: check the duality gap as ultimate stopping
# criterion
# sparse X.T / dense R dot product
X_T_R = np.zeros(n_features)
for ii in xrange(n_features):
for jj in xrange(X_indptr[ii], X_indptr[ii + 1]):
X_T_R[ii] += X_data[jj] * R[X_indices[jj]]
X_T_R[ii] -= X_mean[ii] * R.sum()
XtA = X_T_R - beta * w
if positive:
dual_norm_XtA = np.max(XtA)
else:
dual_norm_XtA = linalg.norm(XtA, np.inf)
R_norm2 = np.dot(R, R)
w_norm2 = np.dot(w, w)
if (dual_norm_XtA > alpha):
const = alpha / dual_norm_XtA
A_norm2 = R_norm2 * const**2
gap = 0.5 * (R_norm2 + A_norm2)
else:
const = 1.0
gap = R_norm2
gap += alpha * linalg.norm(w, 1) - const * np.dot(R.T, y) + \
0.5 * beta * (1 + const ** 2) * (w_norm2)
if gap < tol:
# return if we reached desired tolerance
break
return w, gap, tol
@cython.boundscheck(False)
@cython.wraparound(False)
@cython.cdivision(True)
def enet_coordinate_descent_gram(np.ndarray[DOUBLE, ndim=1] w,
double alpha, double beta,
np.ndarray[DOUBLE, ndim=2] Q,
np.ndarray[DOUBLE, ndim=1] q,
np.ndarray[DOUBLE, ndim=1] y,
int max_iter, double tol, bool positive=False):
"""Cython version of the coordinate descent algorithm
for Elastic-Net regression
We minimize
1 w^T Q w - q^T w + alpha norm(w, 1) + beta norm(w, 2)^2
- ----
2 2
which amount to the Elastic-Net problem when:
Q = X^T X (Gram matrix)
q = X^T y
"""
# get the data information into easy vars
cdef unsigned int n_samples = y.shape[0]
cdef unsigned int n_features = Q.shape[0]
# initial value "Q w" which will be kept of up to date in the iterations
cdef np.ndarray[DOUBLE, ndim=1] H = np.dot(Q, w)
cdef double tmp
cdef double w_ii
cdef double d_w_max
cdef double w_max
cdef double d_w_ii
cdef double gap = tol + 1.0
cdef double d_w_tol = tol
cdef unsigned int ii
cdef unsigned int n_iter
cdef double y_norm2 = np.dot(y, y)
tol = tol * y_norm2
if alpha == 0:
warnings.warn("Coordinate descent with alpha=0 may lead to unexpected"
" results and is discouraged.")
for n_iter in range(max_iter):
w_max = 0.0
d_w_max = 0.0
for ii in xrange(n_features): # Loop over coordinates
if Q[ii, ii] == 0.0:
continue
w_ii = w[ii] # Store previous value
if w_ii != 0.0:
# H -= w_ii * Q[ii]
daxpy(n_features, -w_ii,
<DOUBLE*>(Q.data + ii * n_features * sizeof(DOUBLE)), 1,
<DOUBLE*>H.data, 1)
tmp = q[ii] - H[ii]
if positive and tmp < 0:
w[ii] = 0.0
else:
w[ii] = fsign(tmp) * fmax(fabs(tmp) - alpha, 0) \
/ (Q[ii, ii] + beta)
if w[ii] != 0.0:
# H += w[ii] * Q[ii] # Update H = X.T X w
daxpy(n_features, w[ii],
<DOUBLE*>(Q.data + ii * n_features * sizeof(DOUBLE)), 1,
<DOUBLE*>H.data, 1)
# update the maximum absolute coefficient update
d_w_ii = fabs(w[ii] - w_ii)
if d_w_ii > d_w_max:
d_w_max = d_w_ii
if fabs(w[ii]) > w_max:
w_max = fabs(w[ii])
if w_max == 0.0 or d_w_max / w_max < d_w_tol or n_iter == max_iter - 1:
# the biggest coordinate update of this iteration was smaller than
# the tolerance: check the duality gap as ultimate stopping
# criterion
q_dot_w = np.dot(w, q)
XtA = q - H - beta * w
if positive:
dual_norm_XtA = np.max(XtA)
else:
dual_norm_XtA = linalg.norm(XtA, np.inf)
R_norm2 = y_norm2 + np.sum(w * H) - 2.0 * q_dot_w
w_norm2 = np.dot(w, w)
if (dual_norm_XtA > alpha):
const = alpha / dual_norm_XtA
A_norm2 = R_norm2 * (const ** 2)
gap = 0.5 * (R_norm2 + A_norm2)
else:
const = 1.0
gap = R_norm2
gap += alpha * linalg.norm(w, 1) \
- const * y_norm2 \
+ const * q_dot_w + \
0.5 * beta * (1 + const ** 2) * (w_norm2)
if gap < tol:
# return if we reached desired tolerance
break
return w, gap, tol
cdef double abs_max(int n, double* a):
"""np.max(np.abs(a))"""
cdef int i
cdef double m = fabs(a[0])
cdef double d
for i in xrange(1, n):
d = fabs(a[i])
if d > m:
m = d
return m
cdef double diff_abs_max(int n, double* a, double* b):
"""np.max(np.abs(a - b))"""
cdef int i
cdef double m = fabs(a[0] - b[0])
cdef double d
for i in xrange(1, n):
d = fabs(a[i] - b[i])
if d > m:
m = d
return m
@cython.boundscheck(False)
@cython.wraparound(False)
@cython.cdivision(True)
def enet_coordinate_descent_multi_task(np.ndarray[DOUBLE, ndim=2, mode='fortran'] W,
double l1_reg, double l2_reg,
np.ndarray[DOUBLE, ndim=2, mode='fortran'] X,
np.ndarray[DOUBLE, ndim=2] Y,
int max_iter, double tol):
"""Cython version of the coordinate descent algorithm
for Elastic-Net mult-task regression
We minimize
1 norm(y - X w, 2)^2 + l1_reg ||w||_21 + l2_reg norm(w, 2)^2
- ----
2 2
"""
# get the data information into easy vars
cdef unsigned int n_samples = X.shape[0]
cdef unsigned int n_features = X.shape[1]
cdef unsigned int n_tasks = Y.shape[1]
# compute norms of the columns of X
cdef np.ndarray[DOUBLE, ndim=1] norm_cols_X = (X ** 2).sum(axis=0)
# initial value of the residuals
cdef np.ndarray[DOUBLE, ndim=2, mode='c'] R
cdef np.ndarray[DOUBLE, ndim=1, mode='c'] tmp = np.zeros(n_tasks, dtype=np.float)
cdef np.ndarray[DOUBLE, ndim=1] w_ii = np.zeros(n_tasks, dtype=np.float)
cdef double d_w_max
cdef double w_max
cdef double d_w_ii
cdef double nn
cdef double W_ii_abs_max
cdef double gap = tol + 1.0
cdef double d_w_tol = tol
cdef unsigned int ii
cdef unsigned int n_iter
if l1_reg == 0:
warnings.warn("Coordinate descent with l1_reg=0 may lead to unexpected"
" results and is discouraged.")
R = Y - np.dot(X, W.T)
R = np.asarray(R, order='C')
# tol = tol * linalg.norm(Y, ord='fro') ** 2
tol = tol * dnrm2(n_samples * n_tasks, <DOUBLE*>Y.data, 1) ** 2
for n_iter in range(max_iter):
w_max = 0.0
d_w_max = 0.0
for ii in xrange(n_features): # Loop over coordinates
if norm_cols_X[ii] == 0.0:
continue
# w_ii = W[:, ii] # Store previous value
dcopy(n_tasks, <DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)),
1, <DOUBLE*>w_ii.data, 1)
# if np.sum(w_ii ** 2) != 0.0: # can do better
if dnrm2(n_tasks, <DOUBLE*>w_ii.data, 1) != 0.0:
# R += np.dot(X[:, ii][:, None], w_ii[None, :]) # rank 1 update
dger(CblasRowMajor, n_samples, n_tasks, 1.0,
<DOUBLE*>(X.data + ii * n_samples * sizeof(DOUBLE)), 1,
<DOUBLE*>w_ii.data, 1,
<DOUBLE*>R.data, n_tasks)
# tmp = np.dot(X[:, ii][None, :], R).ravel()
dgemv(CblasRowMajor, CblasTrans,
n_samples, n_tasks, 1.0, <DOUBLE*>R.data,
n_tasks, <DOUBLE*>(X.data + ii * n_samples * sizeof(DOUBLE)),
1, 0.0, <DOUBLE*>tmp.data, 1)
# nn = sqrt(np.sum(tmp ** 2))
nn = dnrm2(n_tasks, <DOUBLE*>tmp.data, 1)
# W[:, ii] = tmp * fmax(1. - l1_reg / nn, 0) / (norm_cols_X[ii] + l2_reg)
dcopy(n_tasks, <DOUBLE*>tmp.data,
1, <DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)), 1)
dscal(n_tasks, fmax(1. - l1_reg / nn, 0) / (norm_cols_X[ii] + l2_reg),
<DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)), 1)
# if np.sum(W[:, ii] ** 2) != 0.0: # can do better
if dnrm2(n_tasks, <DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)), 1) != 0.0:
# R -= np.dot(X[:, ii][:, None], W[:, ii][None, :]) # Update residual : rank 1 update
dger(CblasRowMajor, n_samples, n_tasks, -1.0,
<DOUBLE*>(X.data + ii * n_samples * sizeof(DOUBLE)), 1,
<DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)), 1,
<DOUBLE*>R.data, n_tasks)
# update the maximum absolute coefficient update
d_w_ii = diff_abs_max(n_tasks,
<DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)),
<DOUBLE*>w_ii.data)
if d_w_ii > d_w_max:
d_w_max = d_w_ii
W_ii_abs_max = abs_max(n_tasks,
<DOUBLE*>(W.data + ii * n_tasks * sizeof(DOUBLE)))
if W_ii_abs_max > w_max:
w_max = W_ii_abs_max
if w_max == 0.0 or d_w_max / w_max < d_w_tol or n_iter == max_iter - 1:
# the biggest coordinate update of this iteration was smaller than
# the tolerance: check the duality gap as ultimate stopping
# criterion
XtA = np.dot(X.T, R) - l2_reg * W.T
dual_norm_XtA = np.max(np.sqrt(np.sum(XtA ** 2, axis=1)))
# TODO: use squared L2 norm directly
# R_norm = linalg.norm(R, ord='fro')
# w_norm = linalg.norm(W, ord='fro')
R_norm = dnrm2(n_samples * n_tasks, <DOUBLE*>R.data, 1)
w_norm = dnrm2(n_features * n_tasks, <DOUBLE*>W.data, 1)
if (dual_norm_XtA > l1_reg):
const = l1_reg / dual_norm_XtA
A_norm = R_norm * const
gap = 0.5 * (R_norm ** 2 + A_norm ** 2)
else:
const = 1.0
gap = R_norm ** 2
gap += l1_reg * np.sqrt(np.sum(W ** 2, axis=0)).sum() - const * np.sum(R * Y) + \
0.5 * l2_reg * (1 + const ** 2) * (w_norm ** 2)
if gap < tol:
# return if we reached desired tolerance
break
return W, gap, tol