Improving the Reproducibility of Deep Learning Software: An Initial Investigation through a Case Study Analysis