Abstract—Conditional Independence (CI) testing is a crucial operation in causal model discovery and validation. Effectively performing this requires a linearly scalable and robust algorithm and its implementation. Previous techniques, such as cross-correlation, a linear method; Kernel Conditional Independence Test (KCIT,) and a kernel-based algorithm, do not scale well with dataset size and pose a bottleneck for CI algorithms. An improved version of kernel-based algorithms which use linear mapping to decrease computational time is the Randomized conditional Correlation Test (RCoT) and Randomized Conditional Independence Test (RCIT). This paper describes their use and implementation in Python. This paper then compares the time complexity of the RCoT algorithm with a previously implemented Discretization-based algorithm Probspace. The results show that the accuracy of the previous and current models is similar, but the time taken to get these results has been reduced by 50%. The implemented algorithm takes about 3s to run the testcases (the data used and testcases generated are described in Section IV-C).
 
Keywords—causal inference, conditional independence testing, Randomized conditional Correlation Test (RCoT) algorithm, Lindsay-Pilla-Basaky approximation, Fourier features


Copyright © 2023 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.