incognito124 / knarrow Goto Github PK

My own analytical method of finding the knee. More details in the future

https://github.com/charliermarsh/ruff

Smooth the input data using smoothing cubic splines

https://data.princeton.edu/eco572/smoothing.pdf page 6 section 2.3

https://pre-commit.com/

mypy complains about a missing decorator typings so I should really spend some time to fix that

Split dataset in two parts, do OLS and measure R2

Python 3.6 dropped

Consider speeding it up with raw numpy and not function invocations

It would be awesome to write a cicd pipeline to run tests and coverage I've successfully setup at my local computer

I dislike pip-tools and I belive poetry may be a better choice

Add kneedle from https://doi.org/10.1109/ICDCSW.2011.20

This issue lists Renovate updates and detected dependencies. Read the Dependency Dashboard docs to learn more.

Open

These updates have all been created already. Click a checkbox below to force a retry/rebase of any.

• chore(deps): update dependency cli/typer to v0.12.3

Detected dependencies

• Check this box to trigger a request for Renovate to run again on this repository

By using various internet resources, I'm going to automatize the pypi release process with tbump

workflow:

1. augment the version with tbump
2. it does some magic and tags a commit
3. push of the tag triggers a release job
4. release job builds knarrow stuff and pushes it to pypi
   4a. I might even add the documentation stuff update as well

It seems no tests touch these lines: https://github.com/InCogNiTo124/knarrow/blob/master/src/knarrow/kneedle.py#L26,L52

this should be improved

There are 4 main types of trends in the data:

1. decreasing, convex (think 1/x)

• f'(x) < 0, f''(x) > 0

2. increasing, cconcave (think 1-exp(-x))

• f'(x) > 0, f''(x) < 0

3. decreasing, concave (think ln(-x))

• f'(x) < 0, f''(x) < 0

4. increasing, convex (think exp(x))

• f'(x) > 0, f''(x) > 0

a simple infographics which help understand how the knees are numbered:

______________
| 2  ____  3 |
|  /      \  |
| |        | |
| |        | |
| |        | |
|  \ ____ /  |
| 1        4 |
______________

after detection there should be a conversion to a standard knee type (im in favour of 2)

The knee is to be defined as the point which is furthest from the y=x line on the y axis

this implies that x's and y's are scaled into a [0, 1] x [0, 1] domain, so such scaling should be performed in the method

also closes this
https://github.com/InCogNiTo124/knarrow/projects/1#card-70433054