How to know the slope and intercept of a straight line in a log log plot and how to connect the data in log log scale by a straight line?

Question

DARSHAN KUMAR BISWAS 2022-7-20

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1764690-how-to-know-the-slope-and-intercept-of-a-straight-line-in-a-log-log-plot-and-how-to-connect-the-dat

回答： Star Strider 2022-7-21

for this two graph, I want to do a linear regression, and find out the slope and intercept of that straight line.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

dpb 2022-7-20

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1764690-how-to-know-the-slope-and-intercept-of-a-straight-line-in-a-log-log-plot-and-how-to-connect-the-dat#answer_1012055

在 MATLAB Online 中打开

If one is not concerned about stastical estimation but simply the coefficients, then just use

b=polyfit(log10(x),log10(y),1);     % coefficients in log-log space
yhat=10.^polyval(b,log10(xNew));

2 个评论
显示无隐藏无

DARSHAN KUMAR BISWAS 2022-7-21

Can you please explain the 2nd line.

dpb 2022-7-21

编辑：dpb 2022-7-21

在 MATLAB Online 中打开

How to use coefficients (b) in log space to predict new values...

Try

x=logspace(3,5);y=logspace(2,5);
b=polyfit(log10(x),log10(y),1);
xh=(linspace(1000,100000));
yh=10.^polyval(b,xh);
hold on
plot(xh,yh,'or-')

请先登录，再进行评论。

Answer 2

Star Strider 2022-7-21

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1764690-how-to-know-the-slope-and-intercept-of-a-straight-line-in-a-log-log-plot-and-how-to-connect-the-dat#answer_1012335

The problem with doing regressions on logarithmic transformed variables is that they transform additive errors (that parameter estimation techniques assume) into multiplicative errors (that they do not). The result is that the parameter estimates on transformed variables are not correct.

A better approach would be to use the fminsearch (or fitnlm if you want statistics) function to estimate the parameters directly since:

coding it as:

pwrfcn = @(p,x) exp(p(1)).*x.^p(2); % Logarithmic Intercept = p(1), Logarithmic Slope = p(2)

This can be used as written in fitnlm, however the code for fminsearch requires:

P0 = rand(2,1); % Initial Parameter Estimates (Choose Appropriate Values), Necessary For ‘fminsearch’ & ‘¹fitnlm’

P = fminsearch(@(p) norm(y = pwrfun(p,x)), P0) % Estimate Parameters

yfit = pwrfcn(P,x); % Evaluate Regression

Then plot the data and ‘yfit’ on a loglog plot, as functions of ‘x’.

.