DSTO-GAN: Balancing Data with GAN

DSTO-GAN is a Python library that uses a Generative Adversarial Network (GAN) to generate synthetic samples and balance imbalanced datasets. It is beneficial for classification problems where classes are disproportionate.

Features

1. Generating synthetic samples for class balancing.
2. Training a custom GAN for tabular data.
3. Saving the balanced dataset to a .csv file.

Prerequisites

• Python 3.7 or higher.
• pip package manager.

Installation

You can install the library directly via pip:

pip install dsto-gan

Dependencies

The dependencies will be installed automatically during the installation. If you prefer to install manually, run:

pip install numpy torch pandas scikit-learn xgboost scikit-optimize

How to Use

1. Import and Initialization

First, import the DSTO_GAN class and initialize the object:

from dsto_gan import DSTO_GAN

Initialize DSTO-GAN

dsto_gan = DSTO_GAN(dim_h=64, n_z=10, lr=0.0002, epochs=100, batch_size=64)

2. Balancing Data

Use the fit_resample method to balance the data:

# Imbalanced data
X = ... # Features (numpy array or pandas DataFrame)
y = ... # Labels (numpy array or pandas Series)

# Balance the data
X_resampled, y_resampled = dsto_gan.fit_resample(X, y)

print(f"Shape of the balanced data: {X_resampled.shape}, {y_resampled.shape}")

3. Integration with Scikit-Learn

DSTO_GAN is compatible with Scikit-Learn pipelines. You can use it as part of a preprocessing pipeline:

from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

# Split the data into training and testing
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Create a pipeline with DSTO-GAN and a classifier
pipeline = Pipeline([
('dsto_gan', DSTO_GAN()), # Balancing with DSTO-GAN
('classifier', RandomForestClassifier()) # Classifier
])

# Train the model
pipeline.fit(X_train, y_train)

# Evaluate the model
accuracy = pipeline.score(X_test, y_test)
print(f"Model accuracy: {accuracy:.2f}")

---

## Usage Example

Here is a complete usage example of DSTO-GAN:

```bash
import pandas as pd
from sklearn.model_selection import train_test_split
from dsto_gan import DSTO_GAN

# 1. Load imbalanced data
file_path = "path/to/imbalanced.csv"
df = pd.read_csv(file_path)

# 2. Separate features (X) and labels (y)
X = df.iloc[:, :-1].values ​​# All columns except the last one
y = df.iloc[:, -1].values ​​# Last column is the class

# 3. Split data into training and testing
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# 4. Balance training data with DSTO-GAN
dsto_gan = DSTO_GAN()
X_train_resampled, y_train_resampled = dsto_gan.fit_resample(X_train, y_train)

print(f"Shape of balanced data: {X_train_resampled.shape}, {y_train_resampled.shape}")

Project Structure

dsto_gan/
│
├── dsto_gan/ # Main package
│ ├── __init__.py # Package initialization
│ ├── dsto_gan.py # Core code for data balancing
├── setup.py # Package configuration
├── README.md # Project documentation
└── LICENSE # Project license

Contribution

Contributions are welcome! If you encounter problems or have suggestions for improvements, feel free to open an issue or submit a pull request.

License

This project is licensed under the MIT License. See the [LICENSE] file for more details.

Contact

• Author: Erika Assis
• Email: dudabh@gmail.com
• Repository: GitHub

Acknowledgements

This project was developed as part of research on data balancing using GANs. The open-source community provided the libraries used.