XyGen: ML Features Generator

XyGen is a machine learning library that assists researchers in generating synthetic datasets for evaluating feature selection algorithms. The X refers to the commonly known dependent features and the y refers to the dependent target attribute. The library comprises a single, self-contained module and currently includes 5 different methods for generating artificial datasets: ORAND, ANDOR, ADDER, LED, and PRC. These methods are primarily based on concepts in computer science and electronics. Additionally, XyGen is flexible and can easily be extended to include other custom generation methods.

To use the XyGen module, you simply import the XyGen class from the XyGen module and instantiate a generator object from that class. Using this generator object you can generate features that can be used for benchmarking a suite of features selection algorithms. The synthesized datasets can be saved and loaded from CSV.

Installation

To use the XyGen module, you need to:

1. Ensure that numpy and pandas are installed.
2. Download the file named "XyGen.py" to their local machine.
3. In the Python script where the library will be used, add the line "import XyGen" at the top of the script.

The library is now ready to be used. Some examples are below.

Usage/Examples

Basic Dataset Generation using the XyGen generator

# --- Basic usage: Generating X (features) and y (target) attributes
import numpy as np
from XyGen import XyGen

'''
seed: random seed for generating irrelevant variables 
n_obs: number of instances
n_I: number of irrelevant features
csv_file: name of the csv file that stores the (X,y) data (optional)
We recommend to keep the default values for reproducibility.
'''

data_generator = XyGen(seed=0)
X1, y1 = data_generator.gen_ORAND(n_obs=50, n_I=92, csv_file='orand.csv')
X2, y2 = data_generator.gen_LED(csv_file='led.csv')

Trying XyGen-generated datasets with some Feature Selection algorithms

# --- Trying XyGen-generated datasets with some Feature Selection algorithms

import matplotlib.pyplot as plt
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import chi2

# Generate ORAND with different seeds which changes the irrelevant variables but keeps the relevant variables fixed
# Calculate the total chi2-score for each feature

best_scores = []
for i in range(10):
    data_generator = XyGen(seed=i)
    X, y = data_generator.gen_ORAND(n_obs=30, csv_file='orand.csv')
    sel_kbest = SelectKBest(chi2, k=1)
    sel_kbest = sel_kbest.fit(X, y)
    best_scores.append(sel_kbest.scores_)

#Plot the results
plt.figure(figsize=(20,5))
mean_scores = np.array(best_scores).mean(axis=0)
sd_scores = np.array(best_scores).std(axis=0)
plt.bar(range(1, X.shape[1]+1),mean_scores, yerr=sd_scores, ecolor='orangered', 
        color=4*["blue"]+4*["royalblue"]+2*["forestgreen"]+90*["slategray"],
        error_kw=dict(lw=1))
plt.xticks([1, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100], size=14)
plt.yticks(size=14)
plt.xlabel('features', size=15)
plt.ylabel(r'$\chi^2$', size=15)
plt.title('Univariate')

colors = {'rlvnt':'blue', 'rdnt':'royalblue', 'corr':'forestgreen', 'irrlvnt':'slategray'}         
labels = list(colors.keys())
handles = [plt.Rectangle((0,0),1,1, color=colors[label]) for label in labels]
plt.legend(handles, labels)
plt.show()