"Copy neighbors into the output matrix."
used = np.zeros(dst.nrows, dtype=np.int32)
for i in range(src.nrows):
sp, ep = src.row_extent(i)
for j in range(sp, ep):
c = src.colinds[j]
v = src.values[j]
if c != i and v >= thresh:
_insert(dst, used, limits, i, c, v)
return used
@njit(nogil=True, parallel=not is_mp_worker())
def _sort_nbrs(smat):
for i in prange(smat.nrows):
sp, ep = smat.row_extent(i)
kvp_minheap_sort(sp, ep, smat.colinds, smat.values)
@njit(nogil=True)
def _make_sim_block(nitems, bsp, bitems, r_sp, r_ep, r_cs, r_vs, min_sim, max_nbrs):
# pass 1: compute the size of each row
sizes = np.zeros(bitems, np.int32)
for i in range(nitems):
for j in range(r_sp[i], r_ep[i]):
# we accept the neighbor if it passes threshold and isn't a self-similarity
r = r_cs[j]
if i != bsp + r and r_vs[j] >= min_sim:
from lenskit import util, DataWarning, ConfigWarning
from lenskit.data import sparse_ratings
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__)
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)]
@njit(parallel=not is_mp_worker())
def _sort_nbrs(smat: CSR):
for i in prange(smat.nrows):
sp, ep = smat.row_extent(i)
kvp_minheap_sort(sp, ep, smat.colinds, smat.values)
@njit
def _trim_sim_block(nitems, bsp, bitems, block, min_sim, max_nbrs):
# pass 1: compute the size of each row
sizes = np.zeros(bitems, np.int32)
for i in range(nitems):
sp, ep = block.row_extent(i)
for j in range(sp, ep):
# we accept the neighbor if it passes threshold and isn't a self-similarity
r = block.colinds[j]
"""
TensorFlow-based algorithms.
"""
import logging
from .util import have_usable_tensorflow
from .biasedmf import BiasedMF # noqa: F401
from .ibmf import IntegratedBiasMF # noqa: F401
from .bpr import BPR # noqa: F401
from lenskit.util.parallel import is_mp_worker
TF_AVAILABLE = have_usable_tensorflow()
_log = logging.getLogger(__name__)
if TF_AVAILABLE and is_mp_worker():
import tensorflow as _tf
_log.info('disabling GPUs in worker process')
_tf.config.set_visible_devices([], 'GPU')
_log.info('disabling multithreading in worker')
_tf.config.threading.set_inter_op_parallelism_threads(1)
_tf.config.threading.set_intra_op_parallelism_threads(1)
"""
TensorFlow-based algorithms.
"""
import logging
from .biasedmf import BiasedMF # noqa: F401
from .ibmf import IntegratedBiasMF # noqa: F401
from .bpr import BPR # noqa: F401
from lenskit.util.parallel import is_mp_worker
try:
import tensorflow as _tf
TF_AVAILABLE = True
except ImportError:
TF_AVAILABLE = False
_log = logging.getLogger(__name__)
if is_mp_worker():
_log.info('disabling GPUs in worker process')
_tf.config.set_visible_devices([], 'GPU')
def _worker_status(blob, *args):
_log.info('in worker %s', mp.current_process().name)
return os.getpid(), is_worker(), is_mp_worker()
def _worker_status(blob, *args):
return os.getpid(), is_worker(), is_mp_worker()
