Skip to content

Commit

Permalink
Add stream vs. stack benchmarks to README
Browse files Browse the repository at this point in the history
  • Loading branch information
brandonwillard committed Jan 9, 2021
1 parent 9fef36c commit 6a1b175
Show file tree
Hide file tree
Showing 4 changed files with 164 additions and 20 deletions.
82 changes: 81 additions & 1 deletion README.md
Expand Up @@ -120,7 +120,87 @@ See the full example in the [examples directory](https://github.com/pythological

## Performance and Reliability

`unification`'s current design allows for unification and reification of nested structures that break the Python stack recursion limit. This scalability incurs an overhead cost compared to simple stack-based recursive unification/reificiation.
`unification`'s current design allows for unification and reification of nested structures that would otherwise break the Python stack recursion limit. It uses a generator-based design to "stream" the unifications and reifications.

Below are some stack vs. stream benchmarks that demonstrate how well the stream-based approach scales against the stack-based approach in terms of unifying and reifying deeply nested lists containing integers. These benchmarks were generated from the tests in `tests/test_benchmarks.py` using CPython 3.7.3.

<details><summary>Stack vs. stream benchmarks</summary>
<p>

```python
-------------------------------------------------------------------------------- benchmark 'reify_chain size=10': 2 tests -------------------------------------------------------------------------------
Name (time in us) Min Max Mean StdDev Median IQR Outliers OPS (Kops/s) Rounds Iterations
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_reify_chain_stack[10] 41.0790 (1.0) 545.1940 (3.20) 52.9087 (1.07) 9.7964 (1.04) 50.8650 (1.08) 6.4301 (8.37) 11815;10849 18.9005 (0.93) 260164 1
test_reify_chain_stream[10] 42.4410 (1.03) 170.5540 (1.0) 49.3080 (1.0) 9.3993 (1.0) 47.2400 (1.0) 0.7680 (1.0) 14962;102731 20.2807 (1.0) 278113 1
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

------------------------------------------ benchmark 'reify_chain size=1000': 1 tests -----------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
-------------------------------------------------------------------------------------------------------------------------------
test_reify_chain_stream_large[1000] 7.7722 28.2579 10.0723 2.5087 9.4899 0.3106 70;155 99.2820 1528 1
-------------------------------------------------------------------------------------------------------------------------------

------------------------------------------------------------------------- benchmark 'reify_chain size=300': 2 tests --------------------------------------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_reify_chain_stack[300] 1.5183 (1.0) 22.1821 (1.19) 1.9826 (1.0) 1.5511 (1.16) 1.7410 (1.0) 0.0801 (1.0) 144;684 504.3878 (1.0) 7201 1
test_reify_chain_stream[300] 1.7059 (1.12) 18.6020 (1.0) 2.1237 (1.07) 1.3389 (1.0) 1.9260 (1.11) 0.1020 (1.27) 118;585 470.8745 (0.93) 6416 1
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------- benchmark 'reify_chain size=35': 2 tests --------------------------------------------------------------------------------
Name (time in us) Min Max Mean StdDev Median IQR Outliers OPS (Kops/s) Rounds Iterations
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_reify_chain_stream[35] 129.2780 (1.0) 868.1510 (1.02) 190.0433 (1.11) 36.2784 (1.41) 179.5690 (1.08) 21.5360 (2.30) 1535;1455 5.2620 (0.90) 26072 1
test_reify_chain_stack[35] 150.7850 (1.17) 853.7920 (1.0) 170.5166 (1.0) 25.7944 (1.0) 165.8500 (1.0) 9.3530 (1.0) 3724;5480 5.8645 (1.0) 81286 1
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

------------------------------------------- benchmark 'reify_chain size=5000': 1 tests ------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
---------------------------------------------------------------------------------------------------------------------------------
test_reify_chain_stream_large[5000] 46.9073 86.9737 52.9724 6.6919 49.6787 3.9609 68;68 18.8778 292 1
---------------------------------------------------------------------------------------------------------------------------------

------------------------------------------------------------------------------- benchmark 'unify_chain size=10': 2 tests -------------------------------------------------------------------------------
Name (time in us) Min Max Mean StdDev Median IQR Outliers OPS (Kops/s) Rounds Iterations
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_unify_chain_stream[10] 77.6280 (1.0) 307.9130 (1.0) 86.7625 (1.0) 17.5355 (1.20) 82.7525 (1.0) 1.7290 (1.0) 809;1736 11.5257 (1.0) 15524 1
test_unify_chain_stack[10] 92.9890 (1.20) 309.8770 (1.01) 104.2017 (1.20) 14.6694 (1.0) 101.0160 (1.22) 4.2368 (2.45) 3657;6651 9.5968 (0.83) 73379 1
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

------------------------------------------- benchmark 'unify_chain size=1000': 1 tests ------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
---------------------------------------------------------------------------------------------------------------------------------
test_unify_chain_stream_large[1000] 27.3518 65.5924 31.1374 4.2563 29.5148 3.5286 38;35 32.1158 496 1
---------------------------------------------------------------------------------------------------------------------------------

------------------------------------------------------------------------- benchmark 'unify_chain size=300': 2 tests --------------------------------------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_unify_chain_stream[300] 3.6957 (1.0) 13.1876 (1.0) 4.4439 (1.0) 1.0719 (1.42) 4.2080 (1.0) 0.2410 (1.67) 51;95 225.0298 (1.0) 1114 1
test_unify_chain_stack[300] 4.2952 (1.16) 13.4294 (1.02) 4.7732 (1.07) 0.7555 (1.0) 4.6623 (1.11) 0.1446 (1.0) 36;136 209.5024 (0.93) 2911 1
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------- benchmark 'unify_chain size=35': 2 tests ---------------------------------------------------------------------------------
Name (time in us) Min Max Mean StdDev Median IQR Outliers OPS (Kops/s) Rounds Iterations
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_unify_chain_stream[35] 285.6880 (1.0) 934.9690 (1.0) 324.5402 (1.0) 40.8338 (1.0) 319.8520 (1.0) 20.4375 (1.0) 962;1159 3.0813 (1.0) 24331 1
test_unify_chain_stack[35] 345.2770 (1.21) 1,088.3650 (1.16) 407.9067 (1.26) 52.2263 (1.28) 396.6640 (1.24) 20.6560 (1.01) 2054;3027 2.4515 (0.80) 37594 1
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------- benchmark 'unify_chain size=5000': 1 tests ---------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
--------------------------------------------------------------------------------------------------------------------------------------
test_unify_chain_stream_large[5000] 555.2733 754.9897 605.4949 50.6124 591.1251 61.4030 2;2 1.6515 26 1
--------------------------------------------------------------------------------------------------------------------------------------

Legend:
Outliers: 1 Standard Deviation from Mean; 1.5 IQR (InterQuartile Range) from 1st Quartile and 3rd Quartile.
OPS: Operations Per Second, computed as 1 / Mean
```

</p>
</details>

## About

Expand Down
58 changes: 48 additions & 10 deletions tests/test_benchmarks.py
@@ -1,3 +1,6 @@
import platform
import sys

import pytest

from tests.utils import gen_long_chain
Expand Down Expand Up @@ -47,8 +50,8 @@ def reify_stack(u, s):
@pytest.mark.parametrize("size", nesting_sizes)
def test_unify_chain_stream(size, benchmark):
a_lv = var()
form = gen_long_chain(a_lv, size)
term = gen_long_chain("a", size)
form, lvars = gen_long_chain(a_lv, size, use_lvars=True)
term, _ = gen_long_chain("a", size)

res = benchmark(unify, form, term, {})
assert res[a_lv] == "a"
Expand All @@ -58,8 +61,8 @@ def test_unify_chain_stream(size, benchmark):
@pytest.mark.parametrize("size", nesting_sizes)
def test_unify_chain_stack(size, benchmark):
a_lv = var()
form = gen_long_chain(a_lv, size)
term = gen_long_chain("a", size)
form, lvars = gen_long_chain(a_lv, size, use_lvars=True)
term, _ = gen_long_chain("a", size)

res = benchmark(unify_stack, form, term, {})
assert res[a_lv] == "a"
Expand All @@ -69,19 +72,54 @@ def test_unify_chain_stack(size, benchmark):
@pytest.mark.parametrize("size", nesting_sizes)
def test_reify_chain_stream(size, benchmark):
a_lv = var()
form = gen_long_chain(a_lv, size)
term = gen_long_chain("a", size)
form, lvars = gen_long_chain(a_lv, size, use_lvars=True)
term, _ = gen_long_chain("a", size)

res = benchmark(reify, form, {a_lv: "a"})
lvars.update({a_lv: "a"})
res = benchmark(reify_stack, form, lvars)
assert res == term


@pytest.mark.benchmark(group="reify_chain")
@pytest.mark.parametrize("size", nesting_sizes)
def test_reify_chain_stack(size, benchmark):
a_lv = var()
form = gen_long_chain(a_lv, size)
term = gen_long_chain("a", size)
form, lvars = gen_long_chain(a_lv, size, use_lvars=True)
term, _ = gen_long_chain("a", size)

res = benchmark(reify_stack, form, {a_lv: "a"})
lvars.update({a_lv: "a"})
res = benchmark(reify_stack, form, lvars)
assert res == term


@pytest.mark.benchmark(group="unify_chain")
@pytest.mark.parametrize("size", [1000, 5000])
def test_unify_chain_stream_large(size, benchmark):
a_lv = var()
form, lvars = gen_long_chain(a_lv, size, use_lvars=True)
term, _ = gen_long_chain("a", size)

res = benchmark(unify, form, term, {})
assert res[a_lv] == "a"


@pytest.mark.skipif(
platform.python_implementation() == "PyPy",
reason="PyPy's sys.getrecursionlimit changes",
)
@pytest.mark.benchmark(group="reify_chain")
@pytest.mark.parametrize("size", [sys.getrecursionlimit(), sys.getrecursionlimit() * 5])
def test_reify_chain_stream_large(size, benchmark):
a_lv = var()
form, lvars = gen_long_chain(a_lv, size, use_lvars=True)
term, _ = gen_long_chain("a", size)

lvars.update({a_lv: "a"})

res = benchmark(reify, form, lvars)

if size < sys.getrecursionlimit():
assert res == term
else:
with pytest.raises(RecursionError):
assert res == term
12 changes: 6 additions & 6 deletions tests/test_core.py
Expand Up @@ -218,18 +218,18 @@ def test_reify_recursion_limit():

a_lv = var()

b = gen_long_chain(a_lv, 10)
b, _ = gen_long_chain(a_lv, 10)
res = reify(b, {a_lv: "a"})
assert res == gen_long_chain("a", 10)
assert res == gen_long_chain("a", 10)[0]

r_limit = sys.getrecursionlimit()

try:
sys.setrecursionlimit(100)

b = gen_long_chain(a_lv, 200)
b, _ = gen_long_chain(a_lv, 200)
res = reify(b, {a_lv: "a"})
exp_res = gen_long_chain("a", 200)
exp_res, _ = gen_long_chain("a", 200)

if platform.python_implementation().lower() != "pypy":
# CPython has stack limit issues when comparing nested lists, but
Expand All @@ -248,8 +248,8 @@ def test_reify_recursion_limit():
def test_unify_recursion_limit():
a_lv = var()

b = gen_long_chain("a")
b_var = gen_long_chain(a_lv)
b, _ = gen_long_chain("a")
b_var, _ = gen_long_chain(a_lv)

s = unify(b, b_var, {})

Expand Down
32 changes: 29 additions & 3 deletions tests/utils.py
@@ -1,10 +1,36 @@
import sys

from unification.variable import var

def gen_long_chain(last_elem=None, N=None):

def gen_long_chain(last_elem=None, N=None, use_lvars=False):
"""Generate a nested list of length `N` with the last element set to `last_elm`.
Parameters
----------
last_elem: object
The element to be placed in the inner-most nested list.
N: int
The number of nested lists.
use_lvars: bool
Whether or not to add `var`s to the first elements of each nested list
or simply integers. If ``True``, each `var` is passed the nesting
level integer (i.e. ``var(i)``).
Returns
-------
list, dict
The generated nested list and a ``dict`` containing the generated
`var`s and their nesting level integers, if any.
"""
b_struct = None
if N is None:
N = sys.getrecursionlimit()
lvars = {}
for i in range(N - 1, 0, -1):
b_struct = [i, last_elem if i == N - 1 else b_struct]
return b_struct
i_el = var(i) if use_lvars else i
if use_lvars:
lvars[i_el] = i
b_struct = [i_el, last_elem if i == N - 1 else b_struct]
return b_struct, lvars

0 comments on commit 6a1b175

Please sign in to comment.