Let's suppose we want to compare for a given problem several supervised learning algorithms, in terms of accuracy and speed.
If we don't have a lot of data, is it acceptable to duplicate instances to assess the speed performance?
I know it's not recommended when assessing the accuracy, but what about the speed?
Thanks in advance.
https://ieeexplore.ieee.org/iel7/6287639/6514899/08684947.pdf
https://machinelearningmastery.com/a-tour-of-machine-learning-algorithms/
https://machinelearningmastery.com/evaluate-performance-machine-learning-algorithms-python-using-resampling/
https://www.nature.com/articles/s41598-019-40927-6
If you’re trying to assess the speed performance of the algorithms, I think it’s ok to duplicate. However, duplicating the data will affect the accuracy of the algorithms in a negative way!
Yes you can do it by dimensional reduction or convolution techniques. It will reduce the size of your over all data base and accordingly your computation will be faster.
Thanks all for you answer, but as mentioned in the question, the duplicated data are used to asses the speed only, not the accuracy. The idea behind that is the following: "We don't have a lot of data to asses the speed, but if we had a lot of data, the speed results could be similar to the results obtained when we duplicated data".
- Badges
- Science topic