What type of search starts by selecting an input for partitioning available training data?

The type of search that begins by selecting an input for partitioning available training data is known as a split search. In this context, splitting refers to dividing the dataset into segments based on the values of a particular input variable or predictor. This approach is fundamental in many machine learning tasks, especially when building decision trees or other algorithms that require a clear delineation of data points based on specific criteria.

When a split is made, it allows the model to learn how different partitions of data behave and to predict outcomes more effectively. The goal is to maximize the distinction between classes or minimize prediction error within each segment, thereby improving the overall model performance. The splitting process is essential for feature selection and understanding the relationships and interactions among variables.

The other types do not involve this method of partitioning data at the outset. Explanatory searches focus on understanding relationships without necessarily breaking data into segments. Data searches can pertain to querying or retrieving relevant datasets but do not specifically highlight the partitioning process as a starting point. Model searches are more about evaluating and refining existing models rather than actively partitioning data for training.