Exercises

Once again, you are invited to answer all the following questions. You will struggle to find answers when the problems are hard, since this is the only way to master Estimator and Data APIs:

1. What is an ETL process?
2. How is an ETL process related to the tf.data API?
3. Why can't a tf.data.Dataset object can't be manipulated directly, but every non-static method returns a new dataset object that's the result of the transformation applied?
4. Which are the most common optimizations in the context of the tf.data API? Why is prefetching so important?
5. Given the two datasets of the next question, which one loops faster? Explain your response.
6. Given the following two datasets:

data = tf.data.Dataset.range(100)
data2 = tf.data.Dataset.from_generator(lambda: range(100), (tf.int32))

def l1():
    for v in data:
        tf.print(v)
def l2():
    for v in data2:
        tf.print...