In the context of the equation Err(x) = bias^2 + var + sigma^2, where bias is defined as E[(f(x) – hat{f}(x))^2] and f(x) represents an unknown true function, how can we calculate the bias value without knowledge of f(x)? Specifically, when using the “bias_variance_decomp” function from the “mlxtend.evaluate” library, how does it estimate the bias component? And Sigma how can be calculated?

Try to ask of most smart guys, except good solutions