Mao Q. and Cao Y., 2024. Can a single tree outperform an entire forest? https://doi.org/10.48550/arXiv.2411.17003.

There is no issue with the findings of the paper. I would just like to provide additional clarification regarding the reported prediction time in Table 5, specifically how it is calculated across 10,000 repeated runs for all compared models.

In Table 5, the reported prediction times for GET, GET-Linear, and RF represent the average time required to make predictions on the entire test set over 10,000 runs. This was implemented using the following code snippet:

1
2

import timeit
times = timeit.repeat(stmt=test_code, setup=setup_code, globals=globals(), repeat=repetition, number=numInference)

However, we did not directly compute the average time per individual inference run, that is specified by the number parameter in the timeit.repeat function. If one prefers to measure the exact average time per inference run, this can be obtained by simply dividing the reported time by the number parameter, as shown below with one line of code:

1

times_eachInference = [t/numInference for t in times]

This clarification is provided only to detail how the prediction times are calculated. The findings and discussion of Table 5 remain unaffected, as all models are implemented and evaluated consistently.