Comparing the Efficiency of Random Search and Tree-Structured Parzen Estimator Algorithms to Optimize Convolutional Neural Networks for Image Recognition