The official package to compute the Triangulated Maximally Filtered Graph (TMFG).
Project description
Fast-TMFG
Fast_TMFG is an ultra-fast implementation of the Triangulated Maximally Fileterd Graph (TMFG). It is based on the work by G. P. Massara and is fully implemented by A. Briola.
The interface is fully scikit-learn compatible. Consequently, it has three main methods:
fit(c_matrix, output)
: Fits the model to the input matrixc_matrix
(a correlation matrix). This method computes the Triangulated Maximal Filtered Graph (TMFG) based on the input matrix. Theoutput
parameter specifies what is the nature of the desired output:- sparse inverse covariance matrix (
output = 'logo'
) - sparse unweighted correlation matrix (
output = 'unweighted_inverse_correlation'
) - sparse weighted correlation matrix (
output = 'weighted_inverse_correlation'
)
- sparse inverse covariance matrix (
transform()
: Returns the computed cliques and separators set of the model. The method also returns the TMFG adjacency matrix.fit_transform(c_matrix, output)
: Fits the model to the input matrixc_matrix
(a correlation matrix) and returns the computed cliques and separators set and the TMFG adjacency matrix. Theoutput
parameter specifies what is the nature of the desired output:- sparse inverse covariance matrix (
output = 'logo'
) - sparse unweighted correlation matrix (
output = 'unweighted_inverse_correlation'
) - sparse weighted correlation matrix (
output = 'weighted_inverse_correlation'
)
- sparse inverse covariance matrix (
We provide a detailed explanation of each function/method. Such an explanation is entirely generated through ChatGPT.
For a fully understanding of the TMFG, we refer the interested reader to the follwing papers:
- Parsimonious modeling with information filtering networks
- Network filtering for big data: Triangulated maximally filtered graph
- Dependency structures in cryptocurrency market from high to low frequency
Usage Example
import numpy as np
import pandas as pd
from fast_tmfg import *
def generate_random_df(num_rows, num_columns):
data = np.random.randint(0, 100, size=(num_rows, num_columns))
df = pd.DataFrame(data, columns=['col_{}'.format(i) for i in range(num_columns)])
return df
df = generate_random_df(100, 50)
corr = df.corr()
model = TMFG()
cliques, seps, adj_matrix = model.fit_transform(corr, output='unweighted_inverse_correlation')
How to cite us
If you use TMFG in a scientific publication, we would appreciate citations to the following paper:
@article{briola2022dependency,
title={Dependency structures in cryptocurrency market from high to low frequency},
author={Briola, Antonio and Aste, Tomaso},
journal={arXiv preprint arXiv:2206.03386},
year={2022}
}
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.