Why my fitglm() function ignore the 'Options' input?

% Initialize the predictors
predictor1 = normrnd(0,200,225,40);
predictor2 = normrnd(100,200,225,40);
predictor = [predictor1;predictor2];

% Initialize the response
outcome = [repmat(0,225,1); repmat(1,225,1)];

% Set the options
options = statset('fitglm');
options.MaxIter = 1000;
options.Display = 'iter';

mdl = fitglm(predictor, outcome, "Distribution", "binomial",'Options',options);

However these fitglm completely ignored the options I changed. I am running Matlab in version "23.2.0.2485118 (R2023b) Update 6" and Statistics and Machine learning toolbox in version 23.2.

P.S.: I have seen 'Options' in the function script of fitglm in my matlab, but cannot see it with automatic fullfil when I wrote the fitglm line.

Many thanks for the help!

edit:

I have changed options.Display = 'final', but still nothing was displayed.

[Aquatris]

Documentations says 'Options' so did you try with capital letter O instead of lowercase o :P

[Ziyue]

Yeh, I have tried both :(

[Aquatris]

Why do you think it ignores the options? what is the expected result and what do you observe? Maybe provide a runable code here.

[Ziyue]

Thank you for your response, @Aquatris! I've updated a runnable version here. I suspect the options are being ignored because 'Display', 'iter' should show information at each iteration, but I didn't see any output during fitting.

[Aquatris]

Looks like showing all the iterations is not supported for fitglm function as shown by @Cris LaPierre's answer.

[Ziyue]

Thank you for noticing! I have changed it, but still nothing appeared after I changed it to 'final'. Also, the max iteration is still 100 as default.

Answer 1

[Cris LaPierre]

Note that fitglm does not appear to support the 'iter' option. Looking at the documentation, the options listed for Display are 'off' or 'final'.

[Ziyue]

Hi, thank you for the answer! I have changed it to final, but still nothing was diplayed, so that is not the main issue I guess.

[Aquatris]

Ye looks like it does not work in latest matlab release either. Time to contact technical support!! Nice catch on the bug.

rng('default') % for reproducibility
X = randn(100,7);
mu = exp(X(:,[1 3 6])*[.4;.2;.3] + 1);
y = poissrnd(mu);
% with terminating the line
mdl =  fitglm(X,y,'linear','Distribution','poisson','Options',statset('Display','final','MaxIter',1000));
disp('fitglm Ran! yet no display')

fitglm Ran! yet no display

[Ziyue]

Thank you for checking! I would contact the technical support now!

[Cris LaPierre]

I think you should still contact support, but try removing the semicolon at the end of your fitglm line.

I would expect the display option to superceed the semicolon, so that is what you would want to share with them.

I also noticed that when I tab-complete in MATLAB, there is an 'iter' option. I've pass that discrepancy along internally.

rng('default') % for reproducibility
X = randn(100,7);
mu = exp(X(:,[1 3 6])*[.4;.2;.3] + 1);
y = poissrnd(mu);
% with terminating the line
mdl =  fitglm(X,y,'linear','Distribution','poisson','Options',statset('Display','iter','MaxIter',1000))

mdl = 
Generalized linear regression model:
    log(y) ~ 1 + x1 + x2 + x3 + x4 + x5 + x6 + x7
    Distribution = Poisson

Estimated Coefficients:
                   Estimate        SE        tStat        pValue  
                   _________    ________    ________    __________

    (Intercept)      0.88723    0.070969      12.502    7.3149e-36
    x1               0.44413    0.052337      8.4858    2.1416e-17
    x2             0.0083388    0.056527     0.14752       0.88272
    x3               0.21518    0.063416      3.3932    0.00069087
    x4             -0.058386    0.065503    -0.89135       0.37274
    x5             -0.060824    0.073441     -0.8282       0.40756
    x6               0.34267    0.056778      6.0352    1.5878e-09
    x7               0.04316     0.06146     0.70225       0.48252


100 observations, 92 error degrees of freedom
Dispersion: 1
Chi^2-statistic vs. constant model: 119, p-value = 1.55e-22

[Ziyue]

Hi I opened your script in the online version, and find a very interesting phenomenon that even I changed the code into

mdl =  fitglm(X,y,'linear','Distribution','poisson')%,'Options',statset('Display','iter','MaxIter',1000))

The display was also showed. Also, even if I changed it to

mdl =  fitglm(X,y,'linear','Distribution','poisson','Options',statset('Display','off','MaxIter',1000))

The output was still displayed. I am not sure what is happening in this online matlab version.

Then I copied it into my local matlab and tried again, and found that, if I delete the " ; " at the end of the fitglm() line, the output will be displayed. Otherwise, nothing will be displayed no matter what was input into 'Options'.

[Aquatris]

I dont think the shown things are the iterations. They are just the tuned model parameters. Just to demonstrate, I set display to off and do the same.

rng('default') % for reproducibility
X = randn(100,7);
mu = exp(X(:,[1 3 6])*[.4;.2;.3] + 1);
y = poissrnd(mu);
% with terminating the line
mdl =  fitglm(X,y,'linear','Distribution','poisson','Options',statset('Display','off','MaxIter',1000))

mdl = 
Generalized linear regression model:
    log(y) ~ 1 + x1 + x2 + x3 + x4 + x5 + x6 + x7
    Distribution = Poisson

Estimated Coefficients:
                   Estimate        SE        tStat        pValue  
                   _________    ________    ________    __________

    (Intercept)      0.88723    0.070969      12.502    7.3149e-36
    x1               0.44413    0.052337      8.4858    2.1416e-17
    x2             0.0083388    0.056527     0.14752       0.88272
    x3               0.21518    0.063416      3.3932    0.00069087
    x4             -0.058386    0.065503    -0.89135       0.37274
    x5             -0.060824    0.073441     -0.8282       0.40756
    x6               0.34267    0.056778      6.0352    1.5878e-09
    x7               0.04316     0.06146     0.70225       0.48252


100 observations, 92 error degrees of freedom
Dispersion: 1
Chi^2-statistic vs. constant model: 119, p-value = 1.55e-22

[Cris LaPierre]

i would recommend reporting your observations to Support:

https://www.mathworks.com/support/contact_us.html

[Ziyue]

Sure! I have reported it! Thank you so much!

[Cris LaPierre]

I have learned that the documentation is incorrect. fitglm does not support the 'Display' option at all. The documentation will be updated to remove reference to it.