def _compute_similarities(self, rmat, items, users):
mkl = matrix.mkl_ops()
mkl = None
if mkl is None:
return self._scipy_similarities(rmat, items, users)
else:
return self._mkl_similarities(mkl, rmat)
def test_mkl_available():
if 'CONDA_PREFIX' in os.environ:
ops = lm.mkl_ops()
assert ops is not None
assert ops.clib is not None
else:
skip('only require MKL availability in Conda')
def test_mkl_available():
if 'CONDA_PREFIX' in os.environ:
try:
import mkl
except ImportError:
skip('in MKL-free Conda environment')
ops = lm.mkl_ops()
assert ops is not None
assert ops.clib is not None
else:
skip('only require MKL availability in Conda')
import pandas as pd
import numpy as np
import scipy.sparse as sps
import scipy.sparse.linalg as spla
from numba import jit, njit, prange
from numba.typed import List
from lenskit import util, matrix, DataWarning
from lenskit.sharing import in_share_context
from lenskit.util.parallel import is_mp_worker
from lenskit.util.accum import kvp_minheap_insert, kvp_minheap_sort
from . import Predictor
_logger = logging.getLogger(__name__)
_mkl_ops = matrix.mkl_ops()
_empty_csr = matrix._impl_mod()._empty_csr
_subset_rows = matrix._impl_mod()._subset_rows
if _mkl_ops is not None:
# we have to import LK CFFI utils into this module
for lkv in dir(_mkl_ops):
if lkv.startswith('_lk_mkl'):
globals()[lkv] = getattr(_mkl_ops, lkv)
def _make_blocks(n, size):
"Create blocks for the range 0..n."
return [(s, min(s + size, n)) for s in range(0, n, size)]
def _compute_similarities(self, rmat):
mkl = matrix.mkl_ops()
if mkl is None:
return self._scipy_similarities(rmat)
else:
return self._mkl_similarities(mkl, rmat)
def mkl_ops():
ops = lm.mkl_ops()
if ops is None:
skip('MKL not available')
return ops
import os
import numpy as np
import scipy.sparse as sps
import cffi
from pytest import mark, approx, skip
import lenskit.matrix as lm
import lenskit.util.test as lktu
mkl_ops = lm.mkl_ops()
def test_mkl_available():
if 'CONDA_PREFIX' in os.environ:
assert mkl_ops is not None
assert mkl_ops.clib is not None
else:
skip('only require MKL availability in Conda')
@mark.skipif(mkl_ops is None, reason='MKL not available')
def test_mkl_mult_vec():
for i in range(50):
m = np.random.randint(5, 100)
n = np.random.randint(5, 100)
M = np.random.randn(m, n)
M[M <= 0] = 0
s = sps.csr_matrix(M)
assert s.nnz == np.sum(M > 0)
