Now that we have a better training script, we can repeat the steps in the Training chapter. You'll need to make the following changes:
sourcedir.tar.gzfile with the improved training script
We'll leave these changes as an exercise for you!
As a final note, given that our data set is quite small, it's very likely that our model won't meet the performance of tree-based algorithms like XGBoost. XGBoost achieves over 95% accuracy with this data set, while achieving better than 86% accuracy with our TensorFlow model is quite good.