Genetic Hyperparameter Optimization Library Development and Its Application on Plant Disease Detection Problem

Hyperparameters have a direct contribution to the training success in deep learning models. Most of the time, hyperparameters are tried to be determined by trial and error. However, this process requires manuel changing of the hyperparameters and training the models. In this study, Pykopt - a Python library - was developed which tries to find the best combination from the given set of hyperparameters by using the genetic algorithms. Pykopt is a Python library for optimizing the hyperparameters of Keras deep learning models. With this library we optimized hyperparameters of the VGG16 model on the dataset containing 87,000 pictures of diseased and healthy plant leaves. The model was trained with the hyperparameter set obtained by genetic algorithm and achieved 98.67% validation accuracy.