baohan
/
scikit-learn
mirror of https://github.com/scikit-learn/scikit-learn.git
0
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
scikit-learn/sklearn/utils/weight_vector.pxd

30 lines
880 B
Cython
Raw Normal View History

factored out norm updates and moved them to a dedicated subclass
2012-03-08 04:18:08 +08:00
"""Efficient (dense) parameter vector implementation for linear models. """
moved WeightVector und SequentialDataset into seperate modules.
2012-03-08 03:09:05 +08:00
cimport numpy as np
cdef extern from "math.h":
cdef extern double sqrt(double x)
factored out norm updates and moved them to a dedicated subclass
2012-03-08 04:18:08 +08:00
cdef class WeightVector(object):
moved WeightVector und SequentialDataset into seperate modules.
2012-03-08 03:09:05 +08:00
cdef np.ndarray w
changing averaging to use sparse trick
2014-09-13 00:25:18 +08:00
cdef np.ndarray aw
ENH SGD Cython improvements * nogil declarations where possible (but no with nogil blocks, since the performance benefit of those is not clear) * use BLAS where appropriate * replaced DOUBLE by double, INTEGER and int32 by int/np.intc, since that's the type used by scipy.sparse * explicit checks that dataset dims don't exceed INT_MAX * replaced an np.all by an explicit loop (npymath library for C99 isfinite)
2013-09-07 22:38:53 +08:00
cdef double *w_data_ptr
changing averaging to use sparse trick
2014-09-13 00:25:18 +08:00
cdef double *aw_data_ptr
moved WeightVector und SequentialDataset into seperate modules.
2012-03-08 03:09:05 +08:00
cdef double wscale
clean up and average can now be set to an int indicating the number of iterations before averaging
2014-10-03 21:44:15 +08:00
cdef double average_a
cdef double average_b
ENH SGD Cython improvements * nogil declarations where possible (but no with nogil blocks, since the performance benefit of those is not clear) * use BLAS where appropriate * replaced DOUBLE by double, INTEGER and int32 by int/np.intc, since that's the type used by scipy.sparse * explicit checks that dataset dims don't exceed INT_MAX * replaced an np.all by an explicit loop (npymath library for C99 isfinite)
2013-09-07 22:38:53 +08:00
cdef int n_features
un-done NormedWeightVector factorization; performance decrease on RCV1 is neglectable.
2012-03-19 03:55:25 +08:00
cdef double sq_norm
moved WeightVector und SequentialDataset into seperate modules.
2012-03-08 03:09:05 +08:00
ENH SGD Cython improvements * nogil declarations where possible (but no with nogil blocks, since the performance benefit of those is not clear) * use BLAS where appropriate * replaced DOUBLE by double, INTEGER and int32 by int/np.intc, since that's the type used by scipy.sparse * explicit checks that dataset dims don't exceed INT_MAX * replaced an np.all by an explicit loop (npymath library for C99 isfinite)
2013-09-07 22:38:53 +08:00
cdef void add(self, double *x_data_ptr, int *x_ind_ptr,
adding add_average method and solving iteration bug by replacing definition of t
2014-10-03 15:51:16 +08:00
int xnnz, double c) nogil
cdef void add_average(self, double *x_data_ptr, int *x_ind_ptr,
int xnnz, double c, double num_iter) nogil
ENH SGD Cython improvements * nogil declarations where possible (but no with nogil blocks, since the performance benefit of those is not clear) * use BLAS where appropriate * replaced DOUBLE by double, INTEGER and int32 by int/np.intc, since that's the type used by scipy.sparse * explicit checks that dataset dims don't exceed INT_MAX * replaced an np.all by an explicit loop (npymath library for C99 isfinite)
2013-09-07 22:38:53 +08:00
cdef double dot(self, double *x_data_ptr, int *x_ind_ptr,
int xnnz) nogil
cdef void scale(self, double c) nogil
cdef void reset_wscale(self) nogil
cdef double norm(self) nogil