# This script is part of navis (http://www.github.com/navis-org/navis).
# Copyright (C) 2018 Philipp Schlegel
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
"""Module contains functions implementing NBLAST."""
import time
import numbers
import os
import operator
from functools import partial
import numpy as np
import pandas as pd
import multiprocessing as mp
from concurrent.futures import ProcessPoolExecutor, as_completed
from typing import Callable, Dict, Union, Optional
from typing_extensions import Literal
from navis.nbl.smat import Lookup2d, smat_fcwb, _nblast_v1_scoring
from .. import utils, config
from ..core import NeuronList, Dotprops, make_dotprops
from .base import Blaster, NestedIndices
__all__ = ['nblast', 'nblast_smart', 'nblast_allbyall', 'sim_to_dist']
fp = os.path.dirname(__file__)
smat_path = os.path.join(fp, 'score_mats')
logger = config.get_logger(__name__)
# This multiplier controls job size for NBLASTs (only relevant for
# multiprocessing and if progress=True).
# Larger multiplier = larger job sizes = fewer jobs = slower updates & less overhead
# Smaller multiplier = smaller job sizes = more jobs = faster updates & more overhead
JOB_SIZE_MULTIPLIER = 1
JOB_MAX_TIME_SECONDS = 60 * 30
# This controls how many threads we allow pykdtree to use during multi-core
# NBLAST
OMP_NUM_THREADS_LIMIT = 1
ALLOWED_SCORES = ('forward', 'mean', 'min', 'max', 'both')
class NBlaster(Blaster):
"""Implements version 2 of the NBLAST algorithm.
Please note that some properties are computed on initialization and
changing parameters (e.g. ``use_alpha``) at a later stage will mess things
up!
Parameters
----------
use_alpha : bool
Whether or not to use alpha values for the scoring.
If True, the dotproduct of nearest neighbor vectors will
be scaled by ``sqrt(alpha1 * alpha2)``.
normalized : bool
If True, will normalize scores by the best possible score
(i.e. self-self) of the query neuron.
smat : navis.nbl.smat.Lookup2d | pd.DataFrame | str | Callable
How to convert the point match pairs into an NBLAST score,
usually by a lookup table:
- if 'auto' (default), will use the "official" NBLAST scoring
matrices based on FCWB data. Same behaviour as in R's
nat.nblast implementation.
- if ``smat='v1'`` it uses the analytic formulation of the
NBLAST scoring from Kohl et. al (2013). You can adjust parameter
``sigma_scaling`` (default to 10) using ``smat_kwargs``.
- DataFrames will be used to build a ``Lookup2d``
- if ``Callable`` given, it passes distance and dot products as
first and second argument respectively
- if ``smat=None`` the scores will be generated as the
product of the distances and the dotproduct of the vectors
of nearest-neighbor pairs
smat_kwargs: Dictionary with additional parameters passed to scoring
functions. For example: ``smat_kwargs["sigma_scoring"] = 10``.
limit_dist : float | "auto" | None
Sets the max distance for the nearest neighbor search
(`distance_upper_bound`). Typically this should be the
highest distance considered by the scoring function. If
"auto", will extract that value from the first axis of the
scoring matrix.
progress : bool
If True, will show a progress bar.
"""
def __init__(self, use_alpha=False, normalized=True, smat='auto',
limit_dist=None, approx_nn=False, dtype=np.float64,
progress=True, smat_kwargs=dict()):
"""Initialize class."""
super().__init__(progress=progress, dtype=dtype)
self.use_alpha = use_alpha
self.normalized = normalized
self.approx_nn = approx_nn
self.desc = "NBlasting"
if smat is None:
self.score_fn = operator.mul
elif smat == 'auto':
self.score_fn = smat_fcwb(self.use_alpha)
elif smat == 'v1':
self.score_fn = partial(
_nblast_v1_scoring, sigma_scoring = smat_kwargs.get('sigma_scoring', 10)
)
elif isinstance(smat, pd.DataFrame):
self.score_fn = Lookup2d.from_dataframe(smat)
else:
self.score_fn = smat
if limit_dist == "auto":
try:
if self.score_fn.axes[0].boundaries[-1] != np.inf:
self.distance_upper_bound = self.score_fn.axes[0].boundaries[-1]
else:
# If the right boundary is open (i.e. infinity), we will use
# the second highest boundary plus a 5% offset
self.distance_upper_bound = self.score_fn.axes[0].boundaries[-2] * 1.05
except AttributeError:
logger.warning("Could not infer distance upper bound from scoring function")
self.distance_upper_bound = None
else:
self.distance_upper_bound = limit_dist
def append(self, dotprops: Dotprops, self_hit: Optional[float] = None) -> NestedIndices:
"""Append dotprops.
Returns the numerical index appended dotprops.
If dotprops is a (possibly nested) sequence of dotprops,
return a (possibly nested) list of indices.
Note that `self_hit` is ignored (and hence calculated from scratch)
when `dotprops` is a nested list of dotprops.
"""
if isinstance(dotprops, Dotprops):
return self._append_dotprops(dotprops, self_hit=self_hit)
try:
return [self.append(n) for n in dotprops]
except TypeError: # i.e. not iterable
raise ValueError(f"Expected Dotprops or iterable thereof; got {type(dotprops)}")
def _append_dotprops(self, dotprops: Dotprops, self_hit: Optional[float] = None) -> int:
next_id = len(self)
self.neurons.append(dotprops)
self.ids.append(dotprops.id)
# Calculate score for self hit
if not self_hit:
self.self_hits.append(self.calc_self_hit(dotprops))
else:
self.self_hits.append(self_hit)
return next_id
def calc_self_hit(self, dotprops):
"""Non-normalized value for self hit."""
if not self.use_alpha:
return len(dotprops.points) * self.score_fn(0, 1.0)
else:
dists = np.repeat(0, len(dotprops.points))
alpha = dotprops.alpha * dotprops.alpha
dots = np.repeat(1, len(dotprops.points)) * np.sqrt(alpha)
return self.score_fn(dists, dots).sum()
def single_query_target(self, q_idx: int, t_idx: int, scores='forward'):
"""Query single target against single target."""
# Take a short-cut if this is a self-self comparison
if q_idx == t_idx:
if self.normalized:
return 1
return self.self_hits[q_idx]
# Run nearest-neighbor search for query against target
data = self.neurons[q_idx].dist_dots(self.neurons[t_idx],
alpha=self.use_alpha,
# eps=0.1 means we accept 10% inaccuracy
eps=.1 if self.approx_nn else 0,
distance_upper_bound=self.distance_upper_bound)
if self.use_alpha:
dists, dots, alpha = data
dots *= np.sqrt(alpha)
else:
dists, dots = data
scr = self.score_fn(dists, dots).sum()
# Normalize against best hit
if self.normalized:
scr /= self.self_hits[q_idx]
# For the mean score we also have to produce the reverse score
if scores in ('mean', 'min', 'max', 'both'):
reverse = self.single_query_target(t_idx, q_idx, scores='forward')
if scores == 'mean':
scr = (scr + reverse) / 2
elif scores == 'min':
scr = min(scr, reverse)
elif scores == 'max':
scr = max(scr, reverse)
elif scores == 'both':
# If both scores are requested
scr = [scr, reverse]
return scr
[docs]
def nblast_smart(query: Union[Dotprops, NeuronList],
target: Optional[str] = None,
t: Union[int, float] = 90,
criterion: Union[Literal['percentile'],
Literal['score'],
Literal['N']] = 'percentile',
scores: Union[Literal['forward'],
Literal['mean'],
Literal['min'],
Literal['max']] = 'forward',
return_mask: bool = False,
normalized: bool = True,
use_alpha: bool = False,
smat: Optional[Union[str, pd.DataFrame]] = 'auto',
limit_dist: Optional[Union[Literal['auto'], int, float]] = 'auto',
approx_nn: bool = False,
precision: Union[int, str, np.dtype] = 64,
n_cores: int = os.cpu_count() // 2,
progress: bool = True,
smat_kwargs: Optional[Dict] = dict()) -> pd.DataFrame:
"""Smart(er) NBLAST query against target neurons.
In contrast to :func:`navis.nblast` this function will first run a
"pre-NBLAST" in which only 10% of the query dotprops' points are used.
Using those initial scores, we select for each query the highest scoring
targets and run the full NBLAST only on those query-target pairs (see
``t`` and ``criterion`` for fine-tuning).
Parameters
----------
query : Dotprops | NeuronList
Query neuron(s) to NBLAST against the targets. Neurons
should be in microns as NBLAST is optimized for that and
have similar sampling resolutions.
target : Dotprops | NeuronList, optional
Target neuron(s) to NBLAST against. Neurons should be in
microns as NBLAST is optimized for that and have
similar sampling resolutions. If not provided, will NBLAST
queries against themselves.
t : int | float
Determines for which pairs we will run a full NBLAST. See
``criterion`` parameter for details.
criterion : "percentile" | "score" | "N"
Criterion for selecting query-target pairs for full NBLAST:
- "percentile" runs full NBLAST on the ``t``-th percentile
- "score" runs full NBLAST on all scores above ``t``
- "N" runs full NBLAST on top ``t`` targets
return_mask : bool
If True, will also return a boolean mask that shows which
scores are based on a full NBLAST and which ones only on
the pre-NBLAST.
scores : 'forward' | 'mean' | 'min' | 'max'
Determines the final scores:
- 'forward' (default) returns query->target scores
- 'mean' returns the mean of query->target and
target->query scores
- 'min' returns the minium between query->target and
target->query scores
- 'max' returns the maximum between query->target and
target->query scores
use_alpha : bool, optional
Emphasizes neurons' straight parts (backbone) over parts
that have lots of branches.
normalized : bool, optional
Whether to return normalized NBLAST scores.
smat : str | pd.DataFrame | Callable
Score matrix. If 'auto' (default), will use scoring matrices
from FCWB. Same behaviour as in R's nat.nblast
implementation.
If ``smat='v1'`` it uses the analytic formulation of the
NBLAST scoring from Kohl et. al (2013). You can adjust parameter
``sigma_scaling`` (default to 10) using ``smat_kwargs``.
If ``smat=None`` the scores will be
generated as the product of the distances and the dotproduct
of the vectors of nearest-neighbor pairs.
If ``Callable`` given, it passes distance and dot products as
first and second argument respectively.
smat_kwargs: Dictionary with additional parameters passed to scoring
functions.
limit_dist : float | "auto" | None
Sets the max distance for the nearest neighbor search
(`distance_upper_bound`). Typically this should be the
highest distance considered by the scoring function. If
"auto", will extract that value from the scoring matrix.
While this can give a ~2X speed up, it will introduce slight
inaccuracies because we won't have a vector component for
points without a nearest neighbour within the distance
limits. The impact depends on the scoring function but with
the default FCWB ``smat``, this is typically limited to the
third decimal (0.0086 +/- 0.0027 for an all-by-all of 1k
neurons).
approx_nn : bool
If True, will use approximate nearest neighbors. This gives
a >2X speed up but also produces only approximate scores.
Impact depends on the use case - testing highly recommended!
precision : int [16, 32, 64] | str [e.g. "float64"] | np.dtype
Precision for scores. Defaults to 64 bit (double) floats.
This is useful to reduce the memory footprint for very large
matrices. In real-world scenarios 32 bit (single)- and
depending on the purpose even 16 bit (half) - are typically
sufficient.
n_cores : int, optional
Max number of cores to use for nblasting. Default is
``os.cpu_count() // 2``. This should ideally be an even
number as that allows optimally splitting queries onto
individual processes.
progress : bool
Whether to show progress bars.
Returns
-------
scores : pandas.DataFrame
Matrix with NBLAST scores. Rows are query neurons, columns
are targets. The order is the same as in ``query``/``target``
and the labels are based on the neurons' ``.id`` property.
mask : np.ndarray
Only if ``return_mask=True``: a boolean mask with same shape
as ``scores`` that shows which scores are based on a full
NBLAST and which ones only on the pre-NBLAST.
References
----------
Costa M, Manton JD, Ostrovsky AD, Prohaska S, Jefferis GS. NBLAST: Rapid,
Sensitive Comparison of Neuronal Structure and Construction of Neuron
Family Databases. Neuron. 2016 Jul 20;91(2):293-311.
doi: 10.1016/j.neuron.2016.06.012.
Examples
--------
>>> import navis
>>> nl = navis.example_neurons(n=5)
>>> nl.units
<Quantity([8 8 8 8 8], 'nanometer')>
>>> # Convert to microns
>>> nl_um = nl * (8 / 1000)
>>> # Convert to dotprops
>>> dps = navis.make_dotprops(nl_um)
>>> # Run a NBLAST where only the top target from the pre-NBLAST is run
>>> # through a full NBLAST
>>> scores = navis.nblast_smart(dps[:3], dps[3:], t=1, criterion='N')
See Also
--------
:func:`navis.nblast`
The conventional full NBLAST.
:func:`navis.nblast_allbyall`
A more efficient way than ``nblast(query=x, target=x)``.
:func:`navis.synblast`
A synapse-based variant of NBLAST.
"""
utils.eval_param(criterion, name='criterion',
allowed_values=("percentile", "score", "N"))
utils.eval_param(scores, name='scores', allowed_values=ALLOWED_SCORES)
# We will make a couple tweaks for speed things up if this is
# an all-by-all NBLAST
aba = False
pre_scores = scores
if isinstance(target, type(None)):
target = query
aba = True
# For all-by-all's we can compute only forward scores and
# produce the mean later
if scores == 'mean':
pre_scores = 'forward'
try:
t = int(t)
except BaseException:
raise TypeError(f'`t` must be (convertable to) integer - got "{type(t)}"')
if criterion == 'percentile':
if (t <= 0 or t >= 100):
raise ValueError('Expected `t` to be integer between 0 and 100 for '
f'criterion "percentile", got {t}')
elif criterion == 'N':
if (t < 0 or t > len(target)):
raise ValueError('`t` must be between 0 and the total number of '
f'targets ({len(target)}) for criterion "N", '
f'got {t}')
# Make sure we're working on NeuronLists
query_dps = NeuronList(query)
target_dps = NeuronList(target)
# Run NBLAST preflight checks
nblast_preflight(query_dps, target_dps, n_cores,
req_unique_ids=True,
req_microns=isinstance(smat, str) and smat=='auto')
# Make simplified dotprops
query_dps_simp = query_dps.downsample(10, inplace=False)
if not aba:
target_dps_simp = target_dps.downsample(10, inplace=False)
else:
target_dps_simp = query_dps_simp
# Find a partition that produces batches that each run in approximately
# 10 seconds
if n_cores and n_cores > 1:
if progress:
# If progress bar, we need to make smaller mini batches.
# These mini jobs must not be too small - otherwise the overhead
# from spawning and sending results between processes slows things
# down dramatically. Hence we want to make sure that each job runs
# for >10s. The run time depends on the system and how big the neurons
# are. Here, we run a quick test and try to extrapolate from there
n_rows, n_cols = find_batch_partition(query_dps_simp, target_dps_simp,
T=10 * JOB_SIZE_MULTIPLIER)
else:
# If no progress bar needed, we can just split neurons evenly across
# all available cores
n_rows, n_cols = find_optimal_partition(n_cores, query_dps_simp, target_dps_simp)
else:
n_rows = n_cols = 1
# Calculate self-hits once for all neurons
nb = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
query_self_hits = np.array([nb.calc_self_hit(n) for n in query_dps_simp])
target_self_hits = np.array([nb.calc_self_hit(n) for n in target_dps_simp])
# This makes sure we don't run into multiple layers of concurrency
with set_omp_flag(limits=OMP_NUM_THREADS_LIMIT if n_cores and (n_cores > 1) else None):
# Initialize a pool of workers
# Note that we're forcing "spawn" instead of "fork" (default on linux)!
# This is to reduce the memory footprint since "fork" appears to inherit all
# variables (including all neurons) while "spawn" appears to get only
# what's required to run the job?
with ProcessPoolExecutor(max_workers=n_cores,
mp_context=mp.get_context('spawn')) as pool:
with config.tqdm(desc='Prep. pre-NBLAST',
total=n_rows * n_cols,
leave=False,
disable=not progress) as pbar:
futures = {}
nblasters = []
for qix in np.array_split(np.arange(len(query_dps_simp)), n_rows):
for tix in np.array_split(np.arange(len(target_dps_simp)), n_cols):
# Initialize NBlaster
this = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
# Add queries and targets
for i, ix in enumerate(qix):
this.append(query_dps_simp[ix], query_self_hits[ix])
for i, ix in enumerate(tix):
this.append(target_dps_simp[ix], target_self_hits[ix])
# Keep track of indices of queries and targets
this.queries = np.arange(len(qix))
this.targets = np.arange(len(tix)) + len(qix)
this.queries_ix = qix # this facilitates filling in the big matrix later
this.targets_ix = tix # this facilitates filling in the big matrix later
this.pbar_position = len(nblasters) if not utils.is_jupyter() else None
nblasters.append(this)
pbar.update()
# If multiple cores requested, submit job to the pool right away
if n_cores and n_cores > 1 and (n_cols > 1 or n_rows > 1):
this.progress=False # no progress bar for individual NBLASTERs
futures[pool.submit(this.multi_query_target,
q_idx=this.queries,
t_idx=this.targets,
scores=pre_scores)] = this
# Collect results
if futures and len(futures) > 1:
# Prepare empty score matrix
scr = pd.DataFrame(np.empty((len(query_dps_simp),
len(target_dps_simp)),
dtype=this.dtype),
index=query_dps_simp.id,
columns=target_dps_simp.id)
scr.index.name = 'query'
scr.columns.name = 'target'
# Collect results
# We're dropping the "N / N_total" bit from the progress bar because
# it's not helpful here
fmt = ('{desc}: {percentage:3.0f}%|{bar}| [{elapsed}<{remaining}]')
for f in config.tqdm(as_completed(futures),
desc='Pre-NBLASTs',
bar_format=fmt,
total=len(futures),
smoothing=0,
disable=not progress,
leave=False):
res = f.result()
this = futures[f]
# Fill-in big score matrix
scr.iloc[this.queries_ix, this.targets_ix] = res.values
else:
scr = this.multi_query_target(this.queries,
this.targets,
scores=scores)
# If this is an all-by-all and we would have computed only forward scores
# during pre-NBLAST
if aba and scores == 'mean':
scr = (scr + scr.T.values) / 2
# Now select targets of interest for each query
if criterion == 'percentile':
# Generate a mask for the scores we want to recalculate from full dotprops
sel = np.percentile(scr, q=t, axis=1)
mask = scr >= sel.reshape(-1, 1)
elif criterion == 'score':
# Generate a mask for the scores we want to recalculate from full dotprops
sel = np.full(scr.shape[0], fill_value=t)
mask = scr >= sel.reshape(-1, 1)
else:
# Sort such that the top hit is to the left
srt = np.argsort(scr.values, axis=1)[:, ::-1]
# Generate the mask
mask = pd.DataFrame(np.zeros(scr.shape, dtype=bool),
columns=scr.columns, index=scr.index)
_ = np.arange(mask.shape[0])
for N in range(t):
mask.values[_, srt[:, N]] = True
# Calculate self-hits for full neurons
query_self_hits = np.array([nb.calc_self_hit(n) for n in query_dps])
target_self_hits = np.array([nb.calc_self_hit(n) for n in target_dps])
# This makes sure we don't run into multiple layers of concurrency
with set_omp_flag(limits=OMP_NUM_THREADS_LIMIT if n_cores and (n_cores > 1) else None):
# Initialize a pool of workers
# Note that we're forcing "spawn" instead of "fork" (default on linux)!
# This is to reduce the memory footprint since "fork" appears to inherit all
# variables (including all neurons) while "spawn" appears to get only
# what's required to run the job?
with ProcessPoolExecutor(max_workers=n_cores,
mp_context=mp.get_context('spawn')) as pool:
with config.tqdm(desc='Prep. full NBLAST',
total=n_rows * n_cols,
leave=False,
disable=not progress) as pbar:
futures = {}
nblasters = []
for qix in np.array_split(np.arange(len(query_dps)), n_rows):
for tix in np.array_split(np.arange(len(target_dps)), n_cols):
# Initialize NBlaster
this = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
# Add queries and targets
for i, ix in enumerate(qix):
this.append(query_dps[ix], query_self_hits[ix])
for i, ix in enumerate(tix):
this.append(target_dps[ix], target_self_hits[ix])
# Find the pairs to NBLAST in this part of the matrix
submask = mask.loc[query_dps[qix].id,
target_dps[tix].id]
# `pairs` is an array of `[[query, target], [...]]` pairs
this.pairs = np.vstack(np.where(submask)).T
# Offset the query indices
this.pairs[:, 1] += len(qix)
# Track this NBLASTER's mask relative to the original big one
this.mask = np.zeros(mask.shape, dtype=bool)
this.mask[qix[0]:qix[-1]+1, tix[0]:tix[-1]+1] = submask
# Make sure position of progress bar checks out
this.pbar_position = len(nblasters) if not utils.is_jupyter() else None
this.desc = 'Full NBLAST'
nblasters.append(this)
pbar.update()
# If multiple cores requested, submit job to the pool right away
if n_cores and n_cores > 1 and (n_cols > 1 or n_rows > 1):
this.progress=False # no progress bar for individual NBLASTERs
futures[pool.submit(this.pair_query_target,
pairs=this.pairs,
scores=scores)] = this
# Collect results
if futures and len(futures) > 1:
# Collect results
# We're dropping the "N / N_total" bit from the progress bar because
# it's not helpful here
fmt = ('{desc}: {percentage:3.0f}%|{bar}| [{elapsed}<{remaining}]')
for f in config.tqdm(as_completed(futures),
desc='NBLASTing',
bar_format=fmt,
total=len(futures),
smoothing=0,
disable=not progress,
leave=False):
res = f.result()
this = futures[f]
# Fill-in big score matrix
scr[this.mask] = res
else:
scr[mask] = this.pair_query_target(this.pairs, scores=scores)
if return_mask:
return scr, mask
return scr
[docs]
def nblast(query: Union[Dotprops, NeuronList],
target: Optional[str] = None,
scores: Union[Literal['forward'],
Literal['mean'],
Literal['min'],
Literal['max']] = 'forward',
normalized: bool = True,
use_alpha: bool = False,
smat: Optional[Union[str, pd.DataFrame, Callable]] = 'auto',
limit_dist: Optional[Union[Literal['auto'], int, float]] = None,
approx_nn: bool = False,
precision: Union[int, str, np.dtype] = 64,
n_cores: int = os.cpu_count() // 2,
progress: bool = True,
smat_kwargs: Optional[Dict] = dict()) -> pd.DataFrame:
"""NBLAST query against target neurons.
This implements the NBLAST algorithm from Costa et al. (2016) (see
references) and mirror the implementation in R's ``nat.nblast``
(https://github.com/natverse/nat.nblast).
Parameters
----------
query : Dotprops | NeuronList
Query neuron(s) to NBLAST against the targets. Neurons
should be in microns as NBLAST is optimized for that and
have similar sampling resolutions.
target : Dotprops | NeuronList, optional
Target neuron(s) to NBLAST against. Neurons should be in
microns as NBLAST is optimized for that and have
similar sampling resolutions. If not provided, will NBLAST
queries against themselves.
scores : 'forward' | 'mean' | 'min' | 'max' | 'both'
Determines the final scores:
- 'forward' (default) returns query->target scores
- 'mean' returns the mean of query->target and
target->query scores
- 'min' returns the minium between query->target and
target->query scores
- 'max' returns the maximum between query->target and
target->query scores
- 'both' will return foward and reverse scores as
multi-index DataFrame
use_alpha : bool, optional
Emphasizes neurons' straight parts (backbone) over parts
that have lots of branches.
normalized : bool, optional
Whether to return normalized NBLAST scores.
smat : str | pd.DataFrame | Callable
Score matrix. If 'auto' (default), will use scoring matrices
from FCWB. Same behaviour as in R's nat.nblast
implementation.
If ``smat='v1'`` it uses the analytic formulation of the
NBLAST scoring from Kohl et. al (2013). You can adjust parameter
``sigma_scaling`` (default to 10) using ``smat_kwargs``.
If ``Callable`` given, it passes distance and dot products as
first and second argument respectively.
If ``smat=None`` the scores will be
generated as the product of the distances and the dotproduct
of the vectors of nearest-neighbor pairs.
limit_dist : float | "auto" | None
Sets the max distance for the nearest neighbor search
(`distance_upper_bound`). Typically this should be the
highest distance considered by the scoring function. If
"auto", will extract that value from the scoring matrix.
While this can give a ~2X speed up, it will introduce slight
inaccuracies because we won't have a vector component for
points without a nearest neighbour within the distance
limits. The impact depends on the scoring function but with
the default FCWB ``smat``, this is typically limited to the
third decimal (0.0086 +/- 0.0027 for an all-by-all of 1k
neurons).
approx_nn : bool
If True, will use approximate nearest neighbors. This gives
a >2X speed up but also produces only approximate scores.
Impact depends on the use case - testing highly recommended!
n_cores : int, optional
Max number of cores to use for nblasting. Default is
``os.cpu_count() // 2``. This should ideally be an even
number as that allows optimally splitting queries onto
individual processes.
precision : int [16, 32, 64] | str [e.g. "float64"] | np.dtype
Precision for scores. Defaults to 64 bit (double) floats.
This is useful to reduce the memory footprint for very large
matrices. In real-world scenarios 32 bit (single)- and
depending on the purpose even 16 bit (half) - are typically
sufficient.
progress : bool
Whether to show progress bars. This may cause some overhead,
so switch off if you don't really need it.
smat_kwargs: Dictionary with additional parameters passed to scoring
functions.
Returns
-------
scores : pandas.DataFrame
Matrix with NBLAST scores. Rows are query neurons, columns
are targets. The order is the same as in ``query``/``target``
and the labels are based on the neurons' ``.id`` property.
References
----------
Costa M, Manton JD, Ostrovsky AD, Prohaska S, Jefferis GS. NBLAST: Rapid,
Sensitive Comparison of Neuronal Structure and Construction of Neuron
Family Databases. Neuron. 2016 Jul 20;91(2):293-311.
doi: 10.1016/j.neuron.2016.06.012.
Examples
--------
>>> import navis
>>> nl = navis.example_neurons(n=5)
>>> nl.units
<Quantity([8 8 8 8 8], 'nanometer')>
>>> # Convert to microns
>>> nl_um = nl * (8 / 1000)
>>> # Convert to dotprops
>>> dps = navis.make_dotprops(nl_um)
>>> # Run the nblast
>>> scores = navis.nblast(dps[:3], dps[3:])
See Also
--------
:func:`navis.nblast_allbyall`
A more efficient way than ``nblast(query=x, target=x)``.
:func:`navis.nblast_smart`
A smart(er) NBLAST suited for very large NBLAST.
:func:`navis.synblast`
A synapse-based variant of NBLAST.
"""
utils.eval_param(scores, name='scores', allowed_values=ALLOWED_SCORES)
if isinstance(target, type(None)):
target = query
# Make sure we're working on NeuronLists
query_dps = NeuronList(query)
target_dps = NeuronList(target)
# Run NBLAST preflight checks
nblast_preflight(query_dps, target_dps, n_cores,
req_unique_ids=True,
req_microns=isinstance(smat, str) and smat=='auto')
# Find a partition that produces batches that each run in approximately
# 10 seconds
if n_cores and n_cores > 1:
if progress:
# If progress bar, we need to make smaller mini batches.
# These mini jobs must not be too small - otherwise the overhead
# from spawning and sending results between processes slows things
# down dramatically. Hence we want to make sure that each job runs
# for >10s. The run time depends on the system and how big the neurons
# are. Here, we run a quick test and try to extrapolate from there
n_rows, n_cols = find_batch_partition(query_dps, target_dps,
T=10 * JOB_SIZE_MULTIPLIER)
else:
# If no progress bar needed, we could just split neurons evenly across
# all available cores but that can lead to one core lagging behind
# and finishing much later than all the others. To avoid this, we
# should aim for each batch to finish in a certain amount of time
n_rows, n_cols = find_batch_partition(query_dps, target_dps,
T=JOB_MAX_TIME_SECONDS)
if (n_rows * n_cols) < n_cores:
n_rows, n_cols = find_optimal_partition(n_cores, query_dps, target_dps)
else:
n_rows = n_cols = 1
# Calculate self-hits once for all neurons
nb = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
query_self_hits = np.array([nb.calc_self_hit(n) for n in query_dps])
target_self_hits = np.array([nb.calc_self_hit(n) for n in target_dps])
# This makes sure we don't run into multiple layers of concurrency
with set_omp_flag(limits=OMP_NUM_THREADS_LIMIT if n_cores and (n_cores > 1) else None):
# Initialize a pool of workers
# Note that we're forcing "spawn" instead of "fork" (default on linux)!
# This is to reduce the memory footprint since "fork" appears to inherit all
# variables (including all neurons) while "spawn" appears to get only
# what's required to run the job?
with ProcessPoolExecutor(max_workers=n_cores,
mp_context=mp.get_context('spawn')) as pool:
with config.tqdm(desc='Preparing',
total=n_rows * n_cols,
leave=False,
disable=not progress) as pbar:
futures = {}
nblasters = []
for qix in np.array_split(np.arange(len(query_dps)), n_rows):
for tix in np.array_split(np.arange(len(target_dps)), n_cols):
# Initialize NBlaster
this = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
# Add queries and targets
for i, ix in enumerate(qix):
this.append(query_dps[ix], query_self_hits[ix])
for i, ix in enumerate(tix):
this.append(target_dps[ix], target_self_hits[ix])
# Keep track of indices of queries and targets
this.queries = np.arange(len(qix))
this.targets = np.arange(len(tix)) + len(qix)
this.queries_ix = qix # this facilitates filling in the big matrix later
this.targets_ix = tix # this facilitates filling in the big matrix later
this.pbar_position = len(nblasters) if not utils.is_jupyter() else None
nblasters.append(this)
pbar.update()
# If multiple cores requested, submit job to the pool right away
if n_cores and n_cores > 1 and (n_cols > 1 or n_rows > 1):
this.progress=False # no progress bar for individual NBLASTERs
futures[pool.submit(this.multi_query_target,
q_idx=this.queries,
t_idx=this.targets,
scores=scores)] = this
# Collect results
if futures and len(futures) > 1:
# Prepare empty score matrix
scores = pd.DataFrame(np.empty((len(query_dps), len(target_dps)),
dtype=this.dtype),
index=query_dps.id, columns=target_dps.id)
scores.index.name = 'query'
scores.columns.name = 'target'
# Collect results
# We're dropping the "N / N_total" bit from the progress bar because
# it's not helpful here
fmt = ('{desc}: {percentage:3.0f}%|{bar}| [{elapsed}<{remaining}]')
for f in config.tqdm(as_completed(futures),
desc='NBLASTing',
bar_format=fmt,
total=len(futures),
smoothing=0,
disable=not progress,
leave=False):
res = f.result()
this = futures[f]
# Fill-in big score matrix
scores.iloc[this.queries_ix, this.targets_ix] = res.values
else:
scores = this.multi_query_target(this.queries,
this.targets,
scores=scores)
return scores
[docs]
def nblast_allbyall(x: NeuronList,
normalized: bool = True,
use_alpha: bool = False,
smat: Optional[Union[str, pd.DataFrame, Callable]] = 'auto',
limit_dist: Optional[Union[Literal['auto'], int, float]] = None,
approx_nn: bool = False,
precision: Union[int, str, np.dtype] = 64,
n_cores: int = os.cpu_count() // 2,
progress: bool = True,
smat_kwargs: Optional[Dict] = dict()) -> pd.DataFrame:
"""All-by-all NBLAST of inputs neurons.
A more efficient way than running ``nblast(query=x, target=x)``.
Parameters
----------
x : Dotprops | NeuronList
Neuron(s) to NBLAST against each other. Neurons should
be in microns as NBLAST is optimized for that and have
similar sampling resolutions.
n_cores : int, optional
Max number of cores to use for nblasting. Default is
``os.cpu_count() // 2``. This should ideally be an even
number as that allows optimally splitting queries onto
individual processes.
use_alpha : bool, optional
Emphasizes neurons' straight parts (backbone) over parts
that have lots of branches.
normalized : bool, optional
Whether to return normalized NBLAST scores.
smat : str | pd.DataFrame | Callable, optional
Score matrix/function:
- If ``smat='auto'`` (default), will use scoring matrices
based on flycircuit data. Same behaviour as in R's
nat.nblast implementation.
- For ``smat='v1'``, uses the analytic formulation of the
NBLAST scoring from Kohl et. al (2013). You can adjust
parameter ``sigma_scaling`` (default to 10) using ``smat_kwargs``.
- For ``smat=None`` the scores will be generated as the product
of the distances and the dotproduct of the vectors of
nearest-neighbor pairs.
- If function, must consume distance and dot products as
first and second argument, respectively and return float.
limit_dist : float | "auto" | None
Sets the max distance for the nearest neighbor search
(`distance_upper_bound`). Typically this should be the
highest distance considered by the scoring function. If
"auto", will extract that value from the scoring matrix.
While this can give a ~2X speed up, it will introduce slight
inaccuracies because we won't have a vector component for
points without a nearest neighbour within the distance
limits. The impact depends on the scoring function but with
the default FCWB ``smat``, this is typically limited to the
third decimal (0.0086 +/- 0.0027 for an all-by-all of 1k
neurons).
approx_nn : bool
If True, will use approximate nearest neighbors. This gives
a >2X speed up but also produces only approximate scores.
Impact depends on the use case - testing highly recommended!
precision : int [16, 32, 64] | str [e.g. "float64"] | np.dtype
Precision for scores. Defaults to 64 bit (double) floats.
This is useful to reduce the memory footprint for very large
matrices. In real-world scenarios 32 bit (single)- and
depending on the purpose even 16 bit (half) - are typically
sufficient.
progress : bool
Whether to show progress bars. This cause may some overhead,
so switch off if you don't really need it.
smat_kwargs: Dictionary with additional parameters passed to scoring
functions.
Returns
-------
scores : pandas.DataFrame
Matrix with NBLAST scores. Rows are query neurons, columns
are targets. The order is the same as in ``x``
and the labels are based on the neurons' ``.id`` property.
References
----------
Costa M, Manton JD, Ostrovsky AD, Prohaska S, Jefferis GS. NBLAST: Rapid,
Sensitive Comparison of Neuronal Structure and Construction of Neuron
Family Databases. Neuron. 2016 Jul 20;91(2):293-311.
doi: 10.1016/j.neuron.2016.06.012.
Examples
--------
>>> import navis
>>> nl = navis.example_neurons(n=5)
>>> nl.units
<Quantity([8 8 8 8 8], 'nanometer')>
>>> # Convert to microns
>>> nl_um = nl * (8 / 1000)
>>> # Make dotprops
>>> dps = navis.make_dotprops(nl_um)
>>> # Run the nblast
>>> scores = navis.nblast_allbyall(dps)
See Also
--------
:func:`navis.nblast`
For generic query -> target nblasts.
"""
# Check if pykdtree flag needed to be set
#if n_cores and n_cores > 1:
# check_pykdtree_flag()
# Make sure we're working on NeuronLists
dps = NeuronList(x)
# Run NBLAST preflight checks
# Note that we are passing the same dotprops twice to avoid having to
# change the function's signature. Should have little to no overhead.
nblast_preflight(dps, dps, n_cores,
req_unique_ids=True,
req_microns=isinstance(smat, str) and smat=='auto')
# Find a partition that produces batches that each run in approximately
# 10 seconds
if n_cores and n_cores > 1:
if progress:
# If progress bar, we need to make smaller mini batches.
# These mini jobs must not be too small - otherwise the overhead
# from spawning and sending results between processes slows things
# down dramatically. Hence we want to make sure that each job runs
# for >10s. The run time depends on the system and how big the neurons
# are. Here, we run a quick test and try to extrapolate from there:
n_rows, n_cols = find_batch_partition(dps, dps,
T=10 * JOB_SIZE_MULTIPLIER)
else:
# If no progress bar needed, we can just split neurons evenly across
# all available cores
n_rows, n_cols = find_optimal_partition(n_cores, dps, dps)
else:
n_rows = n_cols = 1
# Calculate self-hits once for all neurons
nb = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
self_hits = np.array([nb.calc_self_hit(n) for n in dps])
# This makes sure we don't run into multiple layers of concurrency
with set_omp_flag(limits=OMP_NUM_THREADS_LIMIT if n_cores and (n_cores > 1) else None):
# Initialize a pool of workers
# Note that we're forcing "spawn" instead of "fork" (default on linux)!
# This is to reduce the memory footprint since "fork" appears to inherit all
# variables (including all neurons) while "spawn" appears to get only
# what's required to run the job?
with ProcessPoolExecutor(max_workers=n_cores,
mp_context=mp.get_context('spawn')) as pool:
with config.tqdm(desc='Preparing',
total=n_rows * n_cols,
leave=False,
disable=not progress) as pbar:
futures = {}
nblasters = []
for qix in np.array_split(np.arange(len(dps)), n_rows):
for tix in np.array_split(np.arange(len(dps)), n_cols):
# Initialize NBlaster
this = NBlaster(use_alpha=use_alpha,
normalized=normalized,
smat=smat,
limit_dist=limit_dist,
dtype=precision,
approx_nn=approx_nn,
progress=progress,
smat_kwargs=smat_kwargs)
# Make sure we don't add the same neuron twice
# Map indices to neurons
to_add = list(set(qix) | set(tix))
# Add neurons
ixmap = {}
for i, ix in enumerate(to_add):
this.append(dps[ix], self_hits[ix])
ixmap[ix] = i
# Keep track of indices of queries and targets
this.queries = [ixmap[ix] for ix in qix]
this.targets = [ixmap[ix] for ix in tix]
this.queries_ix = qix # this facilitates filling in the big matrix later
this.targets_ix = tix # this facilitates filling in the big matrix later
this.pbar_position = len(nblasters) if not utils.is_jupyter() else None
nblasters.append(this)
pbar.update()
# If multiple cores requested, submit job to the pool right away
if n_cores and n_cores > 1 and (n_cols > 1 or n_rows > 1):
this.progress=False # no progress bar for individual NBLASTERs
futures[pool.submit(this.multi_query_target,
q_idx=this.queries,
t_idx=this.targets,
scores='forward')] = this
# Collect results
if futures and len(futures) > 1:
# Prepare empty score matrix
scores = pd.DataFrame(np.empty((len(dps), len(dps)),
dtype=this.dtype),
index=dps.id, columns=dps.id)
scores.index.name = 'query'
scores.columns.name = 'target'
# Collect results
# We're dropping the "N / N_total" bit from the progress bar because
# it's not helpful here
fmt = ('{desc}: {percentage:3.0f}%|{bar}| [{elapsed}<{remaining}]')
for f in config.tqdm(as_completed(futures),
desc='NBLASTing',
bar_format=fmt,
total=len(futures),
smoothing=0,
disable=not progress,
leave=False):
res = f.result()
this = futures[f]
# Fill-in big score matrix
scores.iloc[this.queries_ix, this.targets_ix] = res.values
else:
scores = this.all_by_all()
return scores
def test_single_query_time(q, t, it=100):
"""Test average time of a single NBLAST query."""
# Get a median-sized query and target
q_ix = np.argsort(q.n_points)[len(q)//2]
t_ix = np.argsort(t.n_points)[len(t)//2]
# Run a quick single query benchmark
timings = []
for i in range(it): # Run N tests
s = time.time()
_ = t[t_ix].dist_dots(q[q_ix]) # Dist dot (ignore scoring / normalizing)
timings.append(time.time() - s)
return np.mean(timings) # seconds per medium sized query
def find_batch_partition(q, t, T=10, n_cores=None):
"""Find partitions such that each batch takes about `T` seconds.
Parameters
----------
q,t : NeuronList of Dotprops
Query and targets, respectively.
T : int
Time (in seconds) to aim for.
n_cores : int, optional
Number of cores that will be used. If provided, will try to
make sure that (n_rows * n_cols) is a multiple of n_cores by
increasing the number of rows (thereby decreasing the time
per batch).
Returns
-------
n_rows, n_cols
"""
# Test a single query
time_per_query = test_single_query_time(q, t)
# Number of queries per job such that each job runs in `T` second
queries_per_batch = T / time_per_query
# Number of neurons per batch
neurons_per_batch = max(1, int(np.sqrt(queries_per_batch)))
n_rows = max(1, len(q) // neurons_per_batch)
n_cols = max(1, len(t) // neurons_per_batch)
if n_cores and ((n_rows * n_cols) > n_cores):
while (n_rows * n_cols) % n_cores:
n_rows += 1
return n_rows, n_cols
def find_optimal_partition(N_cores, q, t):
"""Find an optimal partition for given NBLAST query.
Parameters
----------
N_cores : int
Number of available cores.
q,t : NeuronList of Dotprops
Query and targets, respectively.
Returns
-------
n_rows, n_cols
"""
neurons_per_query = []
for n_rows in range(1, N_cores + 1):
# Skip splits we can't make
if N_cores % n_rows:
continue
if n_rows > len(q):
continue
n_cols = min(int(N_cores / n_rows), len(t))
n_queries = len(q) / n_rows
n_targets = len(t) / n_cols
neurons_per_query.append([n_rows, n_cols, n_queries + n_targets])
# Find the optimal partition
neurons_per_query = np.array(neurons_per_query)
n_rows, n_cols = neurons_per_query[np.argmin(neurons_per_query[:, 2]), :2]
return int(n_rows), int(n_cols)
def force_dotprops(x, k, resample, progress=False):
"""Force data into Dotprops."""
if isinstance(x, (NeuronList, list)):
dp = [force_dotprops(n, k, resample) for n in config.tqdm(x,
desc='Dotprops',
disable=not progress,
leave=False)]
return NeuronList(dp)
# Try converting non-Dotprops
if not isinstance(x, Dotprops):
return make_dotprops(x, k=k, resample=resample)
# Return Dotprops
return x
def check_microns(x):
"""Check if neuron data is in microns.
Returns either [True, None (=unclear), False]
"""
microns = [config.ureg.Unit('microns'),
config.ureg.Unit('um'),
config.ureg.Unit('micrometer'),
config.ureg.Unit('dimensionless')]
if not isinstance(x, NeuronList):
x = NeuronList(x)
# For very large NeuronLists converting the unit string to pint units is
# the time consuming step. Here we will first reduce to unique units:
unit_str = []
for n in x:
if isinstance(n._unit_str, str):
unit_str.append(n._unit_str)
else:
unit_str += list(n._unit_str)
unit_str = np.unique(unit_str)
any_not_microns = False
all_units = True
for u in unit_str:
# If not a unit (i.e. `None`)
if not u:
all_units = False
continue
# Convert to proper unit
u = config.ureg(u).to_compact().units
if u not in microns:
any_not_microns = True
if any_not_microns:
return False
elif all_units:
return True
return None
def align_dtypes(*x, downcast=False, inplace=True):
"""Align data types of dotprops.
Parameters
----------
*x : Dotprops | NeuronLists thereof
downcast : bool
If True, will downcast all points to the lowest precision
dtype.
inplace : bool
If True, will modify the original neuron objects. If False, will
make a copy before changing dtypes.
Returns
-------
*x
Input data with aligned dtypes.
"""
dtypes = get_dtypes(x)
if len(dtypes) == 1:
return x
if not inplace:
for i in range(x):
x[i] = x[i].copy()
if downcast:
target = lowest_type(*x)
else:
target = np.result_type(*dtypes)
for n in x:
if isinstance(n, NeuronList):
for i in range(len(n)):
n[i].points = n[i].points.astype(target, copy=False)
elif isinstance(n, Dotprops):
n.points = n.points.astype(target, copy=False)
else:
raise TypeError(f'Unable to process "{type(n)}"')
return x
def get_dtypes(*x):
"""Collect data types of dotprops points."""
dtypes = set()
for n in x:
if isinstance(n, NeuronList):
dtypes = dtypes | get_dtypes(n)
elif isinstance(n, Dotprops):
dtypes.add(n.points.dtype)
else:
raise TypeError(f'Unable to process "{type(n)}"')
return dtypes
def lowest_type(*x):
"""Find the lowest data type."""
dtypes = get_dtypes(x)
if len(dtypes) == 1:
return dtypes[0]
lowest = dtypes[0]
for dt in dtypes[1:]:
lowest = demote_types(lowest, dt)
return lowest
def demote_types(a, b):
"""Determine the lower of two dtypes."""
if isinstance(a, np.ndarray):
a = a.dtype
if isinstance(b, np.ndarray):
b = b.dtype
# No change is same
if a == b:
return a
# First, get the "higher" type
higher = np.promote_types(a, b)
# Now get the one that's not higher
if a != higher:
return a
return b
def sim_to_dist(x):
"""Convert similarity scores to distances.
Parameters
----------
x : (M, M) np.ndarray | pandas.DataFrame
Similarity score matrix to invert.
Returns
-------
distances
"""
if not isinstance(x, (np.ndarray, pd.DataFrame)):
raise TypeError(f'Expected numpy array or pandas DataFrame, got "{type(x)}"')
if isinstance(x, pd.DataFrame):
mx = x.values.max()
else:
mx = x.max()
return (x - mx) * -1
def nblast_preflight(query, target, n_cores, batch_size=None,
req_unique_ids=False, req_dotprops=True, req_points=True,
req_microns=True):
"""Run preflight checks for NBLAST."""
if req_dotprops:
if query.types != (Dotprops, ):
raise TypeError(f'`query` must be Dotprop(s), got "{query.types}". '
'Use `navis.make_dotprops` to convert neurons.')
if target.types != (Dotprops, ):
raise TypeError(f'`target` must be Dotprop(s), got "{target.types}". '
'Use `navis.make_dotprops` to convert neurons.')
if req_points:
no_points = query.n_points == 0
if any(no_points):
raise ValueError('Some query dotprops appear to have no points: '
f'{query.id[no_points]}')
no_points = target.n_points == 0
if any(no_points):
raise ValueError('Some target dotprops appear to have no points: '
f'{target.id[no_points]}')
if req_unique_ids:
# At the moment, neurons need to have a unique ID for things to work
if query.is_degenerated:
raise ValueError('Queries have non-unique IDs.')
if target.is_degenerated:
raise ValueError('Targets have non-unique IDs.')
# Check if query or targets are in microns
# Note this test can return `None` if it can't be determined
if req_microns:
if check_microns(query) is False:
logger.warning('NBLAST is optimized for data in microns and it looks '
'like your queries are not in microns.')
if check_microns(target) is False:
logger.warning('NBLAST is optimized for data in microns and it looks '
'like your targets are not in microns.')
if not isinstance(n_cores, numbers.Number) or n_cores < 1:
raise TypeError('`n_cores` must be an integer > 0')
n_cores = int(n_cores)
if n_cores > 1 and n_cores % 2:
logger.warning('NBLAST is most efficient if `n_cores` is an even number')
elif n_cores < 1:
raise ValueError('`n_cores` must not be smaller than 1')
elif n_cores > os.cpu_count():
logger.warning('`n_cores` should not larger than the number of '
'available cores')
if batch_size is not None:
if batch_size <= 0:
raise ValueError('`batch_size` must be >= 1 or `None`.')
def eval_limit_dist(x):
"""Evaluate `limit_dist` parameter."""
if x == 'auto':
return
if isinstance(x, type(None)):
return
if isinstance(x, numbers.Number):
return
raise ValueError(f'`limit_dist` must be None, "auto" or float, got {x}' )
def check_pykdtree_flag():
"""Check if pykdtree is used and if the OMP_NUM_THREADS flag is set.
The issue is that on Linux pykdtree uses threads by default which causes
issues when we're also using multiple cores (= multiple layers of concurrency).
"""
# This is only relevant for Linux (unless someone compiled pykdtree
# themselves using a compiler that supports openmp)
from sys import platform
if platform not in ("linux", "linux2"):
return
# See if pykdtree is present
try:
import pykdtree
except ImportError:
# If not present, just return
return
import os
if os.environ.get('OMP_NUM_THREADS', None) != "1":
msg = ('`OMP_NUM_THREADS` environment variable not set to 1. This may '
'result in multiple layers of concurrency which in turn will '
'slow down NBLAST when using multiple cores. '
'See also https://github.com/navis-org/navis/issues/49')
logger.warning(msg)
def set_omp_flag(limits=1):
"""Set OMP_NUM_THREADS flag to given value.
This is to avoid pykdtree causing multiple layers of concurrency which
will over-subcribe and slow down NBLAST on multi-core systems.
Use as context manager!
"""
class OMPSetter:
def __init__(self, num_threads):
assert isinstance(num_threads, (int, type(None)))
self.num_threads = num_threads
def __enter__(self):
if self.num_threads is None:
return
# Track old value (if there is one)
self.old_value = os.environ.get('OMP_NUM_THREADS', None)
# Set flag
os.environ['OMP_NUM_THREADS'] = str(self.num_threads)
def __exit__(self, *args, **kwargs):
if self.num_threads is None:
return
# Reset flag
if self.old_value:
os.environ['OMP_NUM_THREADS'] = str(self.old_value)
else:
os.environ.pop('OMP_NUM_THREADS', None)
return OMPSetter(limits)