About Normalized MSE in paper

[titu1994]

First, thank you for providing your implementation, it is quite instructive.

I had a question regarding what normalization was done for MSE during evaluation - was the normalization denominator the variance(y_true) or mean((y_true)^2)?

Because the code for training uses just mse without normalization -

• https://github.com/deep-symbolic-mathematics/LLM-SR/blob/main/specs/specification_bactgrow_numpy.txt#L25
• https://github.com/deep-symbolic-mathematics/LLM-SR/blob/main/specs/specification_oscillator1_numpy.txt#L25
• https://github.com/deep-symbolic-mathematics/LLM-SR/blob/main/specs/specification_oscillator2_numpy.txt#L25
• https://github.com/deep-symbolic-mathematics/LLM-SR/blob/main/specs/specification_oscillator2_torch.txt#L38
• https://github.com/deep-symbolic-mathematics/LLM-SR/blob/main/specs/specification_stressstrain_numpy.txt

I could not find a reference to normalized mse for evaluation in this repo (though I may have missed it)