Tuple Sum Filter
A library to play with filtering numeric sequences by sums of their pairs, triplets, etc.
Comes with a bonus CLI to demo the functionality.
Requires (and CI tests on) python 3.8 to 3.10. If you need to use python 3.7 then try replacing math.prod(some_iterable) with functools.reduce(lambda x, y: x * y, some_iterable)
Approach
We're thinking of this mostly as a library with the CLI as only for demo purposes. Ways you can see this in the code:
- logging should really handled by the consumer,
- our
get_loggershould be something that is passed into the lib
- our
- the CLI is pretty light on automated tests
- we use pretty loose production dependency pinning
- rather than
pip freeze > requirements.txtof a deployed app - we want to keep things loose so that consumers can keep installing us alongside other things
- we should probably set up
tox/noxtest runs against v.latest of our dependencies
- rather than
Running the demo
in a fresh virtualenv (python>=3.8)
# install project and deps
pip install git+https://github.com/lbillingham/tuple-sum-filter.git
# create a suitable input file
echo "1721\n979\n366\n299\n675\n1456\n" > example.txt
# run the demo
filter_demo --input_file=example.txt --sum_target=2020 --dimension=2
you should see output like
checking for pairs of numbers that sum to 2020 in example.txt
Results pair (1721, 299) match: sum to 2020 and multiply to 514579
Consuming the library
The main filtering functions are pairs_that_sum_to and triplets_that_sum_to. They both have signatures (numbers: Sequence[int|float], sum_target: int|float) -> things_that_passed_the_filter list[tuple].
There is also a file-reading helper numbers_in_file exported at the top level.
Developing
Run the following to install the project (and dev dependencies) into your active virtualenv:
make dev_install
day-to-day development tasks can be orchestrated via make
- dependency management
- test/lint/typecheck running
- coverage reporting
- run
makewithout any arguments to see a list
There is a CI suite which runs lint and test on several python versions. We don't run typechecking as a gate in CI because we think that turns a sometimes-useful tool into a Goodhart target.
Performance
We have not been optimizing for performance and it kind of shows.
When we run the benchmarking suite we see ~0.4 seconds fairly consistently for the triplet/3D problem.
We have at least 3 ideas of how to speed things up: several of them include dropping floating-point support.
$ make benchmark
tests/performance_check.py .. [100%]
------------------------------------------------------------------------------------- benchmark: 2 tests ------------------------------------------------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_input1_pairs 5.4665 (1.0) 6.2297 (1.0) 5.6687 (1.0) 0.1018 (1.0) 5.6575 (1.0) 0.1289 (1.0) 47;3 176.4077 (1.0) 172 1
test_input1_triplets 384.6154 (70.36) 386.5000 (62.04) 385.4776 (68.00) 0.8287 (8.14) 385.4333 (68.13) 1.5047 (11.67) 2;0 2.5942 (0.01) 5 1
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
--
1 Jan 23, 2022
2 Nov 18, 2022
3 Jan 15, 2022
2 Jun 27, 2022
37 Oct 1, 2022
5 May 13, 2022
134 Jul 29, 2022
6 Nov 10, 2021
5 May 1, 2022
1 Nov 14, 2021
6 Jul 09, 2022
16 Jun 17, 2022
11 Aug 28, 2022
806 Jan 04, 2023
2 Jan 06, 2022
5 Jun 24, 2022
1 Jan 05, 2022
42 Dec 11, 2022
12 May 17, 2022
2 Dec 05, 2021
2 Sep 08, 2022
1 Feb 10, 2021
0 Feb 13, 2022
2.2k Jan 04, 2023
66 Dec 19, 2022
8 Oct 02, 2022
18 Jan 06, 2023
1 Feb 03, 2022
5 Aug 20, 2022