0
$\begingroup$

I'm not certain how to phrase this question:

I have a dataset of ~45000 execution times of two sets of data. Approximately 35000 of these execution times is ran in one environment, and the remaining ~10000 are ran in a very different environment. The goal of the execution is the same,

The first dataset has execution times ranging from 1-300 seconds, and the second set has a theoretically unbounded, but practical range of 1-7000 seconds.

Both datasets perform the same task, but in slightly different ways, over entirely different inputs.

I have a predictive model for the first dataset, which is accurate on new data which belongs to the same set, e.g., I am able to use the model from set 1 to predict the execution time of new executions running in the same environment.

However, when I try to predict the execution times of data from set B, on the environment from set A - I get suspicious results. e.g., the predicted times are bound between 0 and 300.

I've tried both a linear regression and random forest regression. The random forest regression resulted in predictions from 0-300 and the linear regression resulted in non-sensical predictions (e.g., ranging from -4000 to 4000).

What are some other approaches I can try - I cannot get a better range of response variables from the first set - as that environment is bound to a 300 second execution time.

$\endgroup$
  • $\begingroup$ It seems data from your set A and set B follow the different probability laws. So you need to establish the models for A and B separately. $\endgroup$ – user158565 Oct 20 '18 at 17:50
  • $\begingroup$ @a_statistician I'm not concerned with the "accuracy" of predictions on set B - as I don't know the correct answer anyway - what I do know is the scale, e.g., something from set B that took 5000 seconds should take > 5000 seconds if ran on the same hardware as set A - so getting response values of < 300 seconds for ALL predictions of set B is suspicious. $\endgroup$ – Zack Newsham Oct 20 '18 at 21:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.