Why do I keep getting different 'time to test' results for the same dataset, algorithm and test parameters in WEKA?
I need to observe the test/training times of running different algorithms. But when I run an ML such as NB for example, at one point it gives me 1.7s as the test time and at other times 2.3s or 0.8s (all parameters are kept same).
Similarly, using a dataset with 60 features and the same number of flows but with 20 features for example, the results could show me that the smaller dataset took a longer time.
I would be grateful for an explanation or advice please. Thank you
The time that Weka is displaying is the so-called wall time, which is merely the time that elapsed between starting the evaluation and finishing it. This does not represent the actual time (as in number of CPU cycles) that your machine was performing the evaluation. Depending on other processes on your machine, other Java threads requiring CPU time, this can vary easily.
The Experimenter, in contrast to command-line execution or evaluations in the Explorer, also generates time measures based on actual CPU cycles (UserCPU_Time_...).