Performance of our algorithm in the explored parameter space. Left panels: density contours of the relative error (i.e., the difference between the predicted value and the true value, which is the value of the test set) divided by the true value, as a function of the predicted quantity, either mean μ_pred (top) or standard deviation σ_pred (bottom). The colour scale denotes the number counts within each bin where the error was computed. Right panels: relative error on mean (top) and standard deviation (bottom) as a function of frequency. The box plot highlights the median enclosed by the 25th and 75th percentiles of the error distribution at each frequency, with the error bars enclosing the full spread of values. The mean and standard deviation are computed in logscale. The blue dashed line marks the zero relative error.