7 More experiments with the trained model

In chapter 6, we trained the deep learning model and did a series of experiments to measure and improve its performance. In this chapter, we will go through a set of additional experiments to validate two key aspects of the model: removing bad values (a step that we took as part of the data preparation described in chapters 3 and 4) and including embeddings for the categorical columns (as described in chapter 5). Then we will describe an experiment to compare the deep learning solution using the streetcar delay prediction deep learning model with a solution that uses a non-deep-learning approach called XGBoost.

7.1 Code for more experiments with the model

When you have cloned the GitHub repo (http://mng.bz/v95x) associated with this book, you’ll find the code related to the experiments in the notebooks subdirectory. The following listing shows the files used in the experiments described in this chapter.

Listing 7.1 Code in the repo related to the model training experiments

7.2 Validating whether removing bad values improves the model

7.3 Validating whether embeddings for columns improve the performance of the model

7.4 Comparing the deep learning model with XGBoost

7.5 Possible next steps for improving the deep learning model

Summary