Description: GANify is an algorithm based on Generative Adversarial Learning to generate synthetic non-tensor data. The name GANify is an adaptation of acronym GAN (generative adversarial network) and AmplIFY, meaning you can amplify the amount of data available with GANs .
Installation: One can easily install GANify using the PIP:
pip install ganify==1.0.10
How to use: Once installed, simply import the library and instatiate the model as described below:
Other info: The package also enables the creation of synthetic data using both GANs (Bengio et al., 2014) and WGANs (Chintala et al. 2017), by simply changing the argument "type" on "fit_data". Additionally, you can view the model overall loss performance by calling the "plot_performance()" method after fit.
Further improvements: Early stopping to optimize training interruption
References: Generative Adversarial Nets (Bengio et al. 2014) - https://arxiv.org/pdf/1406.2661.pdf
Wasserstein GAN (Chintala et al. 2017) - https://arxiv.org/abs/1701.07875
Stabilizing Training of Generative Adversarial Networks through Regularization (Hofmann, 2017) - https://papers.nips.cc/paper/6797-stabilizing-training-of-generative-adversarial-networks-through-regularization.pdf
Improved Techniques for Training GANs (Chen et al., 2017) - https://arxiv.org/abs/1606.03498
Big thanks to Jason Brownlee from Machine Learning Mastery. If you don't know his blog, you centainly must: https://machinelearningmastery.com/