Performance improvement #161

vincentsarago · 2024-10-16T15:53:25Z

while working on #160 I wanted to see what were the bottleneck and found that the TileMatrixSets.matrix(z: int) method was the first method slowing things down

import morecantile
import cProfile
import pstats

tms = morecantile.tms.get("WebMercatorQuad")

def get_tile_bounds():
    for n in range(10000):
        _ = tms.xy_bounds(1, 40, 7)
    return

prof = cProfile.Profile()
retval = prof.runcall(get_tile_bounds)
ps = pstats.Stats(prof)
ps.strip_dirs().sort_stats("tottime", "ncalls").print_stats()

         290002 function calls in 0.143 seconds

   Ordered by: internal time, call count

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    20000    0.047    0.000    0.047    0.000 models.py:766(matrix)
    30000    0.021    0.000    0.034    0.000 utils.py:13(_parse_tile_arg)
    10000    0.016    0.000    0.051    0.000 models.py:1024(_ul)
    10000    0.014    0.000    0.071    0.000 models.py:1052(_lr)
    10000    0.011    0.000    0.147    0.000 models.py:1081(xy_bounds)
    80000    0.010    0.000    0.010    0.000 {built-in method __new__ of type object at 0x102feea38}
    20000    0.008    0.000    0.013    0.000 models.py:817(_matrix_origin)
    30000    0.006    0.000    0.009    0.000 <string>:1(<lambda>)
    60000    0.004    0.000    0.004    0.000 {built-in method builtins.len}
        1    0.003    0.003    0.151    0.151 2108246623.py:1(get_tile_bounds)
    20000    0.001    0.000    0.001    0.000 {built-in method math.floor}
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

In this PR I'm adding a _tile_matrices_idx: Dict[str, int] private attribute, which will be used for faster lookup (using dictionary is faster than looping through a list

The second improvement proposed in this PR is the removal of duplicated calls to the _parse_tile_arg function by changing the input for the _ul/ul/_lr/lr methods which accepted Tuple or Tile objects. They now require Tile object.

results:

         290002 function calls (270002 primitive calls) in 0.097 seconds

   Ordered by: internal time, call count

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    20000    0.013    0.000    0.041    0.000 models.py:771(matrix)
    10000    0.012    0.000    0.041    0.000 models.py:1028(_ul)
    10000    0.012    0.000    0.040    0.000 models.py:1055(_lr)
    20000    0.010    0.000    0.011    0.000 fields.py:895(__getattr__)
    10000    0.009    0.000    0.101    0.000 models.py:1082(xy_bounds)
40000/20000    0.009    0.000    0.019    0.000 {built-in method builtins.hasattr}
    20000    0.008    0.000    0.012    0.000 models.py:821(_matrix_origin)
    20000    0.007    0.000    0.026    0.000 main.py:781(__getattr__)
    60000    0.005    0.000    0.005    0.000 {built-in method __new__ of type object at 0x103986a38}
    10000    0.004    0.000    0.008    0.000 utils.py:13(_parse_tile_arg)
        1    0.004    0.004    0.105    0.105 2108246623.py:1(get_tile_bounds)
    20000    0.002    0.000    0.002    0.000 {method 'get' of 'dict' objects}
    10000    0.002    0.000    0.003    0.000 <string>:1(<lambda>)
    20000    0.001    0.000    0.001    0.000 {built-in method builtins.len}
    20000    0.001    0.000    0.001    0.000 {built-in method math.floor}
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

vincentsarago · 2024-10-16T15:57:31Z

🤔 maybe instead of introducing breaking change I can update the functions called by bounds() and xy_bounds() method and not touch the other methods

morecantile/models.py

vincentsarago · 2024-10-16T21:52:33Z

with last commit

         170009 function calls (160009 primitive calls) in 0.059 seconds

   Ordered by: internal time, call count

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    10000    0.020    0.000    0.059    0.000 models.py:1086(xy_bounds)
    10000    0.006    0.000    0.020    0.000 models.py:772(matrix)
    10000    0.005    0.000    0.006    0.000 fields.py:895(__getattr__)
20000/10000    0.005    0.000    0.010    0.000 {built-in method builtins.hasattr}
    10000    0.004    0.000    0.008    0.000 utils.py:13(_parse_tile_arg)
    10000    0.004    0.000    0.007    0.000 models.py:822(_matrix_origin)
    10000    0.004    0.000    0.014    0.000 main.py:781(__getattr__)
        1    0.003    0.003    0.062    0.062 195867796.py:7(get_tile_bounds)
    30000    0.003    0.000    0.003    0.000 {built-in method __new__ of type object at 0x103232a38}
    10000    0.002    0.000    0.003    0.000 <string>:1(<lambda>)
    20000    0.001    0.000    0.001    0.000 {built-in method builtins.len}
    20000    0.001    0.000    0.001    0.000 {built-in method math.floor}
    10002    0.001    0.000    0.001    0.000 {method 'get' of 'dict' objects}
        1    0.000    0.000    0.000    0.000 functools.py:973(__get__)
        1    0.000    0.000    0.000    0.000 models.py:168(ordered_axis_inverted)
        1    0.000    0.000    0.000    0.000 models.py:580(_invert_axis)
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}
        1    0.000    0.000    0.000    0.000 {method '__exit__' of '_thread.RLock' objects}
        1    0.000    0.000    0.000    0.000 {method 'upper' of 'str' objects}

I've reverted the change to _ul and _lr method to directly do the calculation in xy_bounds and

vincentsarago · 2024-10-16T21:53:54Z

morecantile/models.py

+            origin_x
+            + (math.floor(t.x / cf) + 1) * matrix.cellSize * cf * matrix.tileWidth
+        )
+        bottom = origin_y - (t.y + 1) * matrix.cellSize * matrix.tileHeight


do all the calculation here instead of calling _ul and _lr method which will duplicate some stuff (matrix lookup, origin, coalecense)

cons: we end up having duplicated code

vincentsarago · 2024-10-16T21:55:41Z

morecantile/models.py

-        t = _parse_tile_arg(*tile)
+        _left, _bottom, _right, _top = self.xy_bounds(*tile)
+        left, top = self.lnglat(_left, _top)
+        right, bottom = self.lnglat(_right, _bottom)


use the fastest xy_bounds methods instead of ul and lr methods

morecantile/models.py

vincentsarago · 2024-10-16T21:57:33Z

morecantile/models.py

@@ -765,9 +771,8 @@ def custom(

    def matrix(self, zoom: int) -> TileMatrix:


why not use lru_cache here?

I've read some article that mentioned that using lru cache for class method might result in memory leaks 🤷

vincentsarago · 2024-10-16T22:02:17Z

# main
%timeit tms.xy_bounds(1, 40, 7)
7.44 µs ± 45.8 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

# This PR 
%timeit tms.xy_bounds(1, 40, 7)
3.8 µs ± 63.2 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

…into feature/performance-improvement

vincentsarago · 2024-10-17T21:01:15Z

## Main 

----------------------------------------- benchmark 'morecantile.bounds': 7 tests ------------------------------------------
Name (time in us)                                  Min                 Max                Mean              Median          
----------------------------------------------------------------------------------------------------------------------------
morecantile.bounds-Tile(x=0,y=0,z=0)           35.0760 (1.0)       75.6420 (1.00)      35.8244 (1.0)       35.5670 (1.0)    
morecantile.bounds-Tile(x=1,y=0,z=1)           35.4560 (1.01)     107.6420 (1.43)      36.6891 (1.02)      36.0780 (1.01)   
morecantile.bounds-Tile(x=1,y=1,z=1)           35.7070 (1.02)      75.4010 (1.0)       36.5613 (1.02)      36.3080 (1.02)   
morecantile.bounds-Tile(x=1,y=40,z=7)          36.5790 (1.04)      93.9770 (1.25)      37.4655 (1.05)      37.1600 (1.04)   
morecantile.bounds-Tile(x=486,y=332,z=10)      37.3100 (1.06)      82.6840 (1.10)      38.1708 (1.07)      37.8910 (1.07)   
morecantile.bounds-Tile(x=486,y=332,z=20)      38.5020 (1.10)     117.4410 (1.56)      39.9799 (1.12)      39.4940 (1.11)   
morecantile.bounds-Tile(x=486,y=332,z=30)     126.5080 (3.61)     221.2950 (2.93)     130.1808 (3.63)     128.8810 (3.62)   
----------------------------------------------------------------------------------------------------------------------------

---------------------------------------- benchmark 'morecantile.xy_bounds': 7 tests ----------------------------------------
Name (time in us)                                    Min                 Max               Mean             Median          
----------------------------------------------------------------------------------------------------------------------------
morecantile.xy_bounds-Tile(x=0,y=0,z=0)           6.0110 (1.0)       28.1430 (1.0)       6.2460 (1.0)       6.2110 (1.0)    
morecantile.xy_bounds-Tile(x=1,y=1,z=1)           6.1510 (1.02)      30.9480 (1.10)      6.3539 (1.02)      6.3120 (1.02)   
morecantile.xy_bounds-Tile(x=1,y=0,z=1)           6.2020 (1.03)      28.4640 (1.01)      6.4252 (1.03)      6.3920 (1.03)   
morecantile.xy_bounds-Tile(x=1,y=40,z=7)          7.1630 (1.19)      44.2430 (1.57)      7.3898 (1.18)      7.3440 (1.18)   
morecantile.xy_bounds-Tile(x=486,y=332,z=10)      7.7040 (1.28)      31.9600 (1.14)      7.9613 (1.27)      7.9050 (1.27)   
morecantile.xy_bounds-Tile(x=486,y=332,z=20)      9.3570 (1.56)      45.5250 (1.62)      9.6248 (1.54)      9.5670 (1.54)   
morecantile.xy_bounds-Tile(x=486,y=332,z=30)     89.9480 (14.96)    219.0900 (7.78)     92.0518 (14.74)    91.2410 (14.69)  
----------------------------------------------------------------------------------------------------------------------------

#############################################################################################################################
## THIS PR 

---------------------------------------- benchmark 'morecantile.bounds': 7 tests ----------------------------------------
Name (time in us)                                 Min                 Max               Mean             Median          
-------------------------------------------------------------------------------------------------------------------------
morecantile.bounds-Tile(x=0,y=0,z=0)          38.2120 (1.0)      106.0990 (1.33)     39.4564 (1.00)     38.8320 (1.0)    
morecantile.bounds-Tile(x=486,y=332,z=20)     38.3320 (1.00)     114.1940 (1.43)     39.4760 (1.00)     39.0740 (1.01)   
morecantile.bounds-Tile(x=1,y=40,z=7)         38.4920 (1.01)      90.2390 (1.13)     40.0610 (1.02)     39.0040 (1.00)   
morecantile.bounds-Tile(x=1,y=0,z=1)          38.6720 (1.01)      88.0750 (1.10)     39.4304 (1.0)      39.1730 (1.01)   
morecantile.bounds-Tile(x=486,y=332,z=10)     38.7320 (1.01)      79.8900 (1.0)      39.5147 (1.00)     39.2840 (1.01)   
morecantile.bounds-Tile(x=1,y=1,z=1)          38.8930 (1.02)      93.1950 (1.17)     39.7670 (1.01)     39.5040 (1.02)   
morecantile.bounds-Tile(x=486,y=332,z=30)     83.1960 (2.18)     161.5130 (2.02)     86.2243 (2.19)     85.4500 (2.20)   
-------------------------------------------------------------------------------------------------------------------------

---------------------------------------- benchmark 'morecantile.xy_bounds': 7 tests ----------------------------------------
Name (time in us)                                    Min                 Max               Mean             Median          
----------------------------------------------------------------------------------------------------------------------------
morecantile.xy_bounds-Tile(x=1,y=40,z=7)          3.8170 (1.0)       25.7680 (1.0)       3.9653 (1.00)      3.9270 (1.0)    
morecantile.xy_bounds-Tile(x=0,y=0,z=0)           3.8370 (1.01)      41.6990 (1.62)      3.9729 (1.00)      3.9470 (1.01)   
morecantile.xy_bounds-Tile(x=1,y=1,z=1)           3.8370 (1.01)      42.4190 (1.65)      3.9807 (1.00)      3.9470 (1.01)   
morecantile.xy_bounds-Tile(x=1,y=0,z=1)           3.8470 (1.01)      27.8820 (1.08)      3.9614 (1.0)       3.9370 (1.00)   
morecantile.xy_bounds-Tile(x=486,y=332,z=10)      3.8670 (1.01)      34.8660 (1.35)      4.0187 (1.01)      3.9580 (1.01)   
morecantile.xy_bounds-Tile(x=486,y=332,z=20)      3.8770 (1.02)      41.7390 (1.62)      4.0604 (1.03)      4.0170 (1.02)   
morecantile.xy_bounds-Tile(x=486,y=332,z=30)     44.0920 (11.55)    163.7570 (6.36)     46.0675 (11.63)    45.1150 (11.49)  
----------------------------------------------------------------------------------------------------------------------------

add matrix lookup dict and remove duplicated call to parse_tile_arg

c98b77d

AndrewAnnex reviewed Oct 16, 2024

View reviewed changes

morecantile/models.py Outdated Show resolved Hide resolved

AndrewAnnex reviewed Oct 16, 2024

View reviewed changes

morecantile/models.py Outdated Show resolved Hide resolved

revert and update xy_bounds and bounds methods

3a8bed8

vincentsarago force-pushed the feature/performance-improvement branch from 0dec372 to 3a8bed8 Compare October 16, 2024 21:50

vincentsarago commented Oct 16, 2024

View reviewed changes

morecantile/models.py Show resolved Hide resolved

vincentsarago commented Oct 16, 2024

View reviewed changes

vincentsarago changed the title ~~add matrix lookup dict and remove duplicated call to parse_tile_arg~~ Performance improvement Oct 16, 2024

vincentsarago added 2 commits October 17, 2024 22:54

Merge branch 'main' of https://github.com/developmentseed/morecantile …

3cc3b04

…into feature/performance-improvement

update changelog

57e9b1a

vincentsarago marked this pull request as ready for review October 17, 2024 20:57

vincentsarago merged commit b2704d2 into main Oct 17, 2024
7 checks passed

vincentsarago deleted the feature/performance-improvement branch October 17, 2024 21:04

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Performance improvement #161

Performance improvement #161

vincentsarago commented Oct 16, 2024

vincentsarago commented Oct 16, 2024

vincentsarago commented Oct 16, 2024

vincentsarago Oct 16, 2024 •

edited

Loading

vincentsarago Oct 16, 2024

vincentsarago Oct 16, 2024

vincentsarago Oct 16, 2024

vincentsarago commented Oct 16, 2024

vincentsarago commented Oct 17, 2024

		@@ -765,9 +771,8 @@ def custom(

		def matrix(self, zoom: int) -> TileMatrix:

Performance improvement #161

Performance improvement #161

Conversation

vincentsarago commented Oct 16, 2024

vincentsarago commented Oct 16, 2024

vincentsarago commented Oct 16, 2024

vincentsarago Oct 16, 2024 • edited Loading

Choose a reason for hiding this comment

vincentsarago Oct 16, 2024

Choose a reason for hiding this comment

vincentsarago Oct 16, 2024

Choose a reason for hiding this comment

vincentsarago Oct 16, 2024

Choose a reason for hiding this comment

vincentsarago commented Oct 16, 2024

vincentsarago commented Oct 17, 2024

vincentsarago Oct 16, 2024 •

edited

Loading