Which function is used to discretize floating point values into categorical bins?

The function used to discretize floating point values into categorical bins is indeed the bucketized_column function found within the TensorFlow features module. This function allows you to convert continuous numerical data into discrete categories by specifying boundaries for the bins. When applied, it transforms continuous features into a format that more effectively captures the relationships in the data, especially when working with machine learning models that might benefit from categorical inputs.

Using bucketized_column helps in situations where the relationship between the numerical input and the target variable may not be linear, or where specific ranges of values have significant implications for the categorical outcome. This makes it a vital tool in preprocessing steps, enhancing the model's ability to learn from the increased structure in the data.

The other options serve different purposes; for instance, categorical_column is meant for handling existing categorical features but does not discretize continuous values. Numeric_column allows you to input continuous values into the model without altering their representation, and indicator_column is used for converting categorical columns into a one-hot encoded representation, which is also distinct from the discretization process intended by bucketized_column.