Previously parallelization was implemented by open-mp.

However, it is not able to compile with open-mp by cmake on mac and windows platform, and therefore we can't register it to PyPI via Travis CI w/ cibuildwheel with open-mp. So, there is no parallelization support now.

Since std::thread is a cross-platform multi-thread library, it is expected that we can compile it on any platforms (i.e, gcc, clang and visual studio). Plus, std::async looks to be good for parallelization with returning some deferrable values.

Now, there is little difference between batch_query and query methods; this parallelization would make batch_query much faster than single query again.

Hi,
Thank you for this nice work.
I love the fact it support batch query (unlike rtree Toblerity/rtree#178).
It would be great if it could support 3 dimensions (or maybe even n dimensions if that does not impact the speed). Does that seem easily feasible ?
This could then potentially be used in the trimesh library to speed thing up and solve issue mikedh/trimesh#1116

Hi @atksh!
Thanks for this extremely fast solution! Works fine in Python 3.8, but i have some outdated software with integrated Python 2.7. Is it possible to make version for IronPython 2.7? I can succsessfully use some dlls from other projects, but here we have pybind11. Tried to build it with latest pybind for python 2.7, had many errors. Thanks in advance!

How do you get the performance graphs you showed?

E.g. as a an array of concatenated idx's and an array of lengths for each query rectangle.
This would speed up software like trimesh where otherwise there is large performance impact in iterating over this list of lists in the python side.

An interesting package. Thanks for sharing.
When I ran the first example, it does not return as expected.
Please have a look.

import numpy as np
from python_prtree import PRTree2D
idxes = np.array([1, 2])

rects is a list of (xmin, ymin, xmax, ymax)

rects = np.array([[0.0, 0.0, 1.0, 0.5],
[1.0, 1.5, 1.2, 3.0]])
prtree = PRTree2D(idxes, rects)

batch query

q = np.array([[0.5, 0.2, 0.6, 0.3],
[0.8, 0.5, 1.5, 3.5]])
result = prtree.batch_query(q)
print(result)

[[1], [1, 2]]

You can insert an additional rectangle by insert method,

prtree.insert(3, np.array([1.0, 1.0, 2.0, 2.0]))
q = np.array([[0.5, 0.2, 0.6, 0.3],
[0.8, 0.5, 1.5, 3.5]])
result = prtree.batch_query(q)
print(result)

[[1], [1, 2, 3]]

Plus, you can erase by an index.

prtree.erase(2)
result = prtree.batch_query(q)
print(result)

[[1], [1, 3]]

Non-batch query is also supported.

print(prtree.query(0.5, 0.5))

[1]

print(prtree.query((0.5, 0.5)))

[1]

=============================================
The final output look like as follows:
[[1], [2]]
[[1], [2, 3]]
[[1], [3]]
[]
[]

It was described as follows.

• http://hhsprings.pinoko.jp/site-hhs/2016/09/pure-python-な-r-tree-実装が欲しいの/

Also, the error described in the above blog has been fixed in v0.5.0.