GAForecast 0.2.0

Prediction library for tabular data powered by neural network.

GAForecast

This Python library is designed for making predictions on tabular data using a neural network based on ResNet50. The library's strength lies in its ability to excel in predicting time series data presented in a tabular format, where each row contains multiple historical time points.

Requirements

Prior to utilizing this library, please ensure that you have installed TensorFlow on your system.

Additionally, having access to a GPU is highly recommended.

Memory requirements

Table with 70 features will require cca 7gb of GPU memory.

Installation

You can install the GAForecast library from pypi: https://pypi.org/project/GAForecast/

Usage

from gaforecast.models.binary_classifier import GAForecastBinaryClassifier

# Create an instance of the GAForecastBinaryClassifier
clf = GAForecastBinaryClassifier()

# Fit the classifier to your tabular data
clf.fit(X_train, y_train)

# Make predictions on new data
y_pred = clf.predict(X_test)

Internals

Transforms tabular data into images using the Gramian Angular Field technique. Reshapes these images to fit into the ResNet50 architecture. Builds a classifier on top of TensorFlow's ResNet50. Offers a common scikit-learn interface for ease of use.

Future Updates

In future updates, the library will include:

1) A regressor for continuous prediction tasks.

2) Support for multi-class classification.

Please stay tuned for these upcoming enhancements.

Memory Usage

When working with tables containing 70 features, approximately 7GB of GPU memory may be required during training. Training on a GPU is recommended for optimal performance.

License

This library is provided under the MIT License.

Contact

For any questions or issues, please contact [[e-mail]].

Acknowledgments

We would like to especially acknowledge the contributions of the pandas, numpy, scikit-learn, pytz, open-cv and TensorFlow communities for their invaluable libraries and resources.