Iteratively Learning Conditional Statements in Transforming Data by Example

Programming by example (PBE) enables users to transform data formats without coding. As data transformation often involves data with heterogeneous formats, it often requires learning a conditional statement to differentiate these different formats. However, to be practical, the method must learn the correct conditional statement efficiently and accurately with little user input. We present an approach to reduce the conditional statement learning time and the required amount of data. This approach takes advantage of the fact that users interact iteratively with a programming-by-example system. Our approach learns from previous iterations to guide the program generation for the current iteration. The final results show that our method successfully reduces the system running time and the number of examples.