scikit-learn/sklearn/utils/seq_dataset.pyx

183 lines
7.1 KiB
Cython

# cython: cdivision=True
# cython: boundscheck=False
# cython: wraparound=False
#
# Author: Peter Prettenhofer <peter.prettenhofer@gmail.com>
#
# Licence: BSD 3 clause
cimport cython
from libc.limits cimport INT_MAX
cimport numpy as np
import numpy as np
np.import_array()
cdef class SequentialDataset:
"""Base class for datasets with sequential data access. """
cdef void next(self, double **x_data_ptr, int **x_ind_ptr,
int *nnz, double *y, double *sample_weight) nogil:
"""Get the next example ``x`` from the dataset.
Parameters
----------
x_data_ptr : double**
A pointer to the double array which holds the feature
values of the next example.
x_ind_ptr : np.intc**
A pointer to the int array which holds the feature
indices of the next example.
nnz : int*
A pointer to an int holding the number of non-zero
values of the next example.
y : double*
The target value of the next example.
sample_weight : double*
The weight of the next example.
"""
with gil:
raise NotImplementedError()
cdef void shuffle(self, seed):
"""Permutes the ordering of examples. """
raise NotImplementedError()
cdef class ArrayDataset(SequentialDataset):
"""Dataset backed by a two-dimensional numpy array.
The dtype of the numpy array is expected to be ``np.float64`` (double)
and C-style memory layout.
"""
def __cinit__(self, np.ndarray[double, ndim=2, mode='c'] X,
np.ndarray[double, ndim=1, mode='c'] Y,
np.ndarray[double, ndim=1, mode='c'] sample_weights):
"""A ``SequentialDataset`` backed by a two-dimensional numpy array.
Parameters
----------
X : ndarray, dtype=double, ndim=2, mode='c'
The samples; a two-dimensional c-continuous numpy array of
dtype double.
Y : ndarray, dtype=double, ndim=1, mode='c'
The target values; a one-dimensional c-continuous numpy array of
dtype double.
sample_weights : ndarray, dtype=double, ndim=1, mode='c'
The weight of each sample; a one-dimensional c-continuous numpy
array of dtype double.
"""
if X.shape[0] > INT_MAX or X.shape[1] > INT_MAX:
raise ValueError("More than %d samples or features not supported;"
" got (%d, %d)."
% (INT_MAX, X.shape[0], X.shape[1]))
self.n_samples = X.shape[0]
self.n_features = X.shape[1]
cdef np.ndarray[int, ndim=1,
mode='c'] feature_indices = np.arange(0, self.n_features,
dtype=np.intc)
self.feature_indices = feature_indices
self.feature_indices_ptr = <int *> feature_indices.data
self.current_index = -1
self.stride = X.strides[0] / X.itemsize
self.X_data_ptr = <double *>X.data
self.Y_data_ptr = <double *>Y.data
self.sample_weight_data = <double *>sample_weights.data
# Use index array for fast shuffling
cdef np.ndarray[int, ndim=1,
mode='c'] index = np.arange(0, self.n_samples,
dtype=np.intc)
self.index = index
self.index_data_ptr = <int *> index.data
cdef void next(self, double **x_data_ptr, int **x_ind_ptr,
int *nnz, double *y, double *sample_weight) nogil:
cdef int current_index = self.current_index
if current_index >= (self.n_samples - 1):
current_index = -1
current_index += 1
cdef int sample_idx = self.index_data_ptr[current_index]
cdef int offset = sample_idx * self.stride
y[0] = self.Y_data_ptr[sample_idx]
x_data_ptr[0] = self.X_data_ptr + offset
x_ind_ptr[0] = self.feature_indices_ptr
nnz[0] = self.n_features
sample_weight[0] = self.sample_weight_data[sample_idx]
self.current_index = current_index
cdef void shuffle(self, seed):
np.random.RandomState(seed).shuffle(self.index)
cdef class CSRDataset(SequentialDataset):
"""A ``SequentialDataset`` backed by a scipy sparse CSR matrix. """
def __cinit__(self, np.ndarray[double, ndim=1, mode='c'] X_data,
np.ndarray[int, ndim=1, mode='c'] X_indptr,
np.ndarray[int, ndim=1, mode='c'] X_indices,
np.ndarray[double, ndim=1, mode='c'] Y,
np.ndarray[double, ndim=1, mode='c'] sample_weight):
"""Dataset backed by a scipy sparse CSR matrix.
The feature indices of ``x`` are given by x_ind_ptr[0:nnz].
The corresponding feature values are given by
x_data_ptr[0:nnz].
Parameters
----------
X_data : ndarray, dtype=double, ndim=1, mode='c'
The data array of the CSR matrix; a one-dimensional c-continuous
numpy array of dtype double.
X_indptr : ndarray, dtype=np.intc, ndim=1, mode='c'
The index pointer array of the CSR matrix; a one-dimensional
c-continuous numpy array of dtype np.intc.
X_indices : ndarray, dtype=np.intc, ndim=1, mode='c'
The column indices array of the CSR matrix; a one-dimensional
c-continuous numpy array of dtype np.intc.
Y : ndarray, dtype=double, ndim=1, mode='c'
The target values; a one-dimensional c-continuous numpy array of
dtype double.
sample_weights : ndarray, dtype=double, ndim=1, mode='c'
The weight of each sample; a one-dimensional c-continuous numpy
array of dtype double.
"""
self.n_samples = Y.shape[0]
self.current_index = -1
self.X_data_ptr = <double *>X_data.data
self.X_indptr_ptr = <int *>X_indptr.data
self.X_indices_ptr = <int *>X_indices.data
self.Y_data_ptr = <double *>Y.data
self.sample_weight_data = <double *>sample_weight.data
# Use index array for fast shuffling
cdef np.ndarray[int, ndim=1, mode='c'] idx = np.arange(self.n_samples,
dtype=np.intc)
self.index = idx
self.index_data_ptr = <int *>idx.data
cdef void next(self, double **x_data_ptr, int **x_ind_ptr,
int *nnz, double *y, double *sample_weight) nogil:
cdef int current_index = self.current_index
if current_index >= (self.n_samples - 1):
current_index = -1
current_index += 1
cdef int sample_idx = self.index_data_ptr[current_index]
cdef int offset = self.X_indptr_ptr[sample_idx]
y[0] = self.Y_data_ptr[sample_idx]
x_data_ptr[0] = self.X_data_ptr + offset
x_ind_ptr[0] = self.X_indices_ptr + offset
nnz[0] = self.X_indptr_ptr[sample_idx + 1] - offset
sample_weight[0] = self.sample_weight_data[sample_idx]
self.current_index = current_index
cdef void shuffle(self, seed):
np.random.RandomState(seed).shuffle(self.index)