Metrics RSS vs RMS

PGS

v2.20 introduces the new metric RSS (Residual Sum of Squares). My understanding is that RSS and RMS lead to identical ranking of candidate formulae, and thus there is no need for both metrics in an application fot symbolic regression. Can admin please explain if I am wrong, and the reasoning for the new metric?
P.S: The webside does not yet list the RSS metric: https://turingbotsoftware.com/documentation.html#error-metrics

RMS ≡ √(Σ[(y_i−y'i)²]/N)
RSS ≡ Σ[(y_i−y'i)²]
→ Therefore: RMS = √(RSS/N), which means that there is a perfect rank-correlation between RMS and RSS, thus no difference in ranking of candidate equations. I would personally therefore use RSS instead of RMS, because it is presumaly faster, avoiding an unnecessary division and square root.

admin

Hi @PGS, you are correct that RSS is almost identical to RMS, but:

Since the errors are larger in magnitude, the probability of accepting a worse solution during the simulated annealing process is affected. Specifically, the search is made more conservative. This changes the optimization behavior and may be useful in some scenarios.
Some people may just be interested in RSS for whatever reason (as was the case for a particular user who requested this feature).

Overall, this is indeed an incremental addition. The documentation has been updated to explain it.