General Deep Learning collected examples, and important notes throughout the preprocess training and predicting, the most important note is start from the last epochs after the first training and hyperparameters tuning, there's no need to consume time and computational resources for redundant process, also to use any layer from any model and build on top of that. No need to say the comparison of two histories of training data for different epochs is important step see the behavior of data and notice the overfitting, underfitting. Not always use sophisticated model and algorithm gives good results. pay attention to the data's dimension as well.