so4gp 0.9.3

A Python library for gradual pattern mining algorithms.

Python implementation of Gradual Pattern (GP) mining algorithms

SO4GP is a high-performance Python library designed to optimize the extraction of gradual patterns from large-scale datasets. By integrating advanced computation techniques and data management strategies, the library significantly reduces processing time and memory overhead during knowledge discovery.

Implemented Extraction Algorithms

The library provides native Python implementations for the following core and meta-heuristic gradual pattern mining algorithms:

• GRAANK: The foundational classical approach for mining gradual patterns.
• Ant Colony Optimization (AntGRAANK): Meta-heuristic swarm optimization for search-space pruning.
• Genetic Algorithm (GeneticGRAANK): Evolutionary search strategy for optimized pattern extraction.
• Particle Swarm Optimization (ParticleGRAANK): Swarm intelligence framework for fast convergence.
• Random Search (HillClimbingGRAANK): Baseline stochastic search variant.
• Clustering-based Mining (ClusterGP): Data partitioning to accelerate pattern discovery.

What are Gradual Patterns?

A Gradual Pattern (GP) is a co-occurring set of gradual items (GI) that captures covariations between attributes. A pattern's quality is measured quantitatively by its computed support value.

Example

Consider a dataset containing 10 objects with 3 attributes: age, salary, and cars. An extracted GP might look like:

$${\text{age}^+, \text{salary}^-} \quad [\text{Support} = 0.8]$$

This output explicitly reveals that in 80% of the dataset (8 out of 10 objects), an increase in age ($^+$) strongly correlates with a simultaneous decrease in salary ($^-$).

Installation

pip install so4gp

Usage

To use any algorithm to mine GPs, follow the instructions that follow.

First and foremost, import the so4gp python package via:

import so4gp as sgp
# OR 
from so4gp.algorithms import GRAANK

GRAdual rANKing Algorithm for GPs (GRAANK)

This is the classical approach (initially proposed by Anne Laurent) for mining gradual patterns. All the remaining algorithms are variants of this algorithm.

import pandas as pd
from so4gp.algorithms import GRAANK

dummy_data = [["2021-03", 30, 3, 1, 10], ["2021-04", 35, 2, 2, 8], ["2021-05", 40, 4, 2, 7], ["2021-06", 50, 1, 1, 6], ["2021-07", 52, 7, 1, 2]]
dummy_df = pd.DataFrame(dummy_data, columns=['Date', 'Age', 'Salary', 'Cars', 'Expenses'])
    
mine_obj = GRAANK(data_source=dummy_df, min_sup=0.5, eq=False)
gp_json = mine_obj.discover()
print(gp_json)

where you specify the parameters as follows:

• data_source - [required] data source {either a file in csv format or a Pandas DataFrame}
• min_sup - [optional] minimum support default = 0.5
• eq - [optional] encode equal values as gradual default = False

Sample Output

The default output is the format of JSON:

{
	"Algorithm": "GRAANK",
	"Best Patterns": [
            [["Age+", "Salary+"], 0.6], 
            [["Expenses-", "Age+", "Salary+"], 0.6]
	],
	"Iterations": 20
}

Contributors ✨

Thanks go to these incredible people:

Made with contrib.rocks.

References

• Owuor, D., Runkler T., Laurent A., Menya E., Orero J (2021), Ant Colony Optimization for Mining Gradual Patterns. International Journal of Machine Learning and Cybernetics. https://doi.org/10.1007/s13042-021-01390-w
• Dickson Owuor, Anne Laurent, and Joseph Orero (2019). Mining Fuzzy-temporal Gradual Patterns. In the proceedings of the 2019 IEEE International Conference on Fuzzy Systems (FuzzIEEE). IEEE. https://doi.org/10.1109/FUZZ-IEEE.2019.8858883.
• Laurent A., Lesot MJ., Rifqi M. (2009) GRAANK: Exploiting Rank Correlations for Extracting Gradual Itemsets. In: Andreasen T., Yager R.R., Bulskov H., Christiansen H., Larsen H.L. (eds) Flexible Query Answering Systems. FQAS 2009. Lecture Notes in Computer Science, vol 5822. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-04957-6_33

See Docs for more details