At the time of writing, the available pre-made Estimators are DNNClassifier, DNNRegressor, LinearClassifier, and LinearRegressor. Here, DNN stands for deep neural network. Combined Estimators based on both architectures are also available—DNNLinearCombinedClassifier and DNNLinearCombinedRegressor.
In machine learning, classification is the process of predicting a discrete category, while regression is the process of predicting a continuous number.
Combined Estimators, also called deep-n-wide models, make use of a linear model (for memorization) and a deep model (for generalization). They are mostly used for recommendation or ranking models.
Pre-made Estimators are suitable for some machine learning problems. However, they are not suitable for computer vision problems, as there are no pre-made Estimators with convolutions, a powerful type of layer described in the next chapter.
