I know similar questions have been asked for time series data. But my question is a little bit different.

Consider that we have input dataset $X \in R^{N \times M}$, where $M$ is the dimension of inputs and $N$ is the number of samples. The output can be more than one dimension. Therefore, the output of our dataset is $Y \in R^{N \times T}$, where $T$ is the dimension of outputs. The problem refers to multitarget regression in the literature.

My question: Is there any criteria to measure the smoothness or the roughness of this dataset?

My definition for smoothness: Small change on inputs does not change outputs significantly.

Also, if there is a method for just $T=1$, I would like to know.


Curvature based penalty terms have been suggested for regularising RBF and MLP neural networks, see e.g. Bishop (equation 8). Note however your mileage may vary (hint: what contsraints apply to the gaps between datapoints?).

  • $\begingroup$ Is curvature-based a method to smooth the dataset or a method to measure the smoothness? $\endgroup$ – mohammad Mar 4 '20 at 8:47
  • $\begingroup$ It is a measure of the smoothness of a fit to the dataset, but that might include a model that exactly interpolates the dataset, which is probably as good as any method of estimating the smoothness of the data itself. $\endgroup$ – Dikran Marsupial Mar 4 '20 at 9:20
  • $\begingroup$ Thanks. Is there any package or function in r, python or Matlab to measure this criterion? $\endgroup$ – mohammad Mar 4 '20 at 9:26
  • $\begingroup$ Not that I know of. If there is a package that implements spline fitting, it may have something similar though. $\endgroup$ – Dikran Marsupial Mar 4 '20 at 9:34

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.