markmelnic / scalg

List scoring algorithm. Analyse data using a range based procentual proximity algorithm.

Scoring Algorithm (SCALG)

This algorithm works based on a percentual range proximity principle. Initially it was developed for a personal project, however later I found out it is a form of Newton's method used in statistics to solve maximum likelihood equations.

Usage

pip install scalg

As of 15 september 2020 it contains two methods (score and score_columns) which will be described and demonstrated in the examples below.

import scalg

Examples

This will be the sample dataset used as source_data withing the examples with the corresponding indexes and column weights.

Columns ->  0     1      2      3
Weights ->  1     0      0      1
        1[[2016 ,21999 ,62000  ,181],
Sets -> 2 [2013 ,21540 ,89000  ,223],
        3 [2015 ,18900 ,100000 ,223],
        4 [2013 ,24200 ,115527 ,223],
        5 [2016 ,24990 ,47300  ,223]]

Score Method

The output if you pass in source_data and weights:

scalg.score(source_data, [1, 0, 0, 1])

[[2016, 21999, 62000,  181, 2.2756757812463335],
 [2013, 21540, 89000,  223, 1.9553074815952338],
 [2015, 18900, 100000, 223, 2.894245191297678],
 [2013, 24200, 115527, 223, 1.1297208538587848],
 [2016, 24990, 47300,  223, 3.0]]

The output if you pass in source_data, weights and get_scores=True:

scalg.score(source_data, [1, 0, 0, 1], get_scores=True)

[2.2756757812463335, 1.9553074815952338, 2. 894245191297678, 1.1297208538587848, 3.0]

The output if you pass in source_data, weights and get_score_lists=True:

scalg.score(source_data, [1, 0, 0, 1], get_score_lists=True)

[[1.0                 ,0.0, 0.6666666666666666 ,0.0                      ,1.0]
 [0.49113300492610834 ,0.5665024630541872      ,1.0, 0.12972085385878485 ,0.0]
 [0.7845427763202251  ,0.38880501854104677     ,0.22757852463101114      ,0.0]
 [0.0                 ,1.0                     ,1.0                      ,1.0]]

This gives out the score of each element in the list compared to other elements, keeping it's position.

Score Columns Method

Here you may use the same weights which you would use in scalg.score, or you may specify the weights of each column in the corresponding order. In this example using the weights argument [1, 0, 0, 1] or [0, 1] would make no difference.

The output if you pass in source_data, columns and weights:

scalg.score_columns(source_data, [0, 1], [1, 0, 0, 1])

 Scored columns            Scores for corresponding columns
   0|    1|                  |
[[2016 ,21999 ,62000  ,181 ,1.4911330049261085],
 [2013 ,21540 ,89000  ,223 ,0.5665024630541872],
 [2015 ,18900 ,100000 ,223 ,1.6666666666666665],
 [2013 ,24200 ,115527 ,223 ,0.12972085385878485],
 [2016 ,24990 ,47300  ,223 ,1.0]]

The score was computed only based on columns 0 and 1.