Matlab mysterious new error with fitglm and fitlm

Hello! What could be going on with this bug? Any help would be appreciated. I used to be able to run fitglm and fitlm in Matlab 2019, but now I get the following errror, including when I use the piece of code from the help documenthttps://www.mathworks.com/help/stats/fitglm.html:

Error using matlab.internal.datatypes.throwInstead (line 7)
The class table has no Constant property or Static method named 'fromScalarStruct'.

Error in struct2table (line 68)
            matlab.internal.datatypes.throwInstead(ME,'MATLAB:table:UnequalFieldLengths',message('MATLAB:struct2table:UnequalFieldLengths'));

Error in dataset2table (line 13)
t = struct2table(s,'AsArray',false);

Error in classreg.regr.TermsRegression/handleDataArgs (line 525)
                X = dataset2table(X);

Error in LinearModel.fit (line 1184)
            [X,y,haveDataset,otherArgs] = LinearModel.handleDataArgs(X,varargin{:});

Error in fitlm (line 121)
model = LinearModel.fit(X,varargin{:});

Error in fitlm_try_debug (line 4)
       lm = fitlm(d,'MPG ~ Year + Weight + Weight^2')

[Ive J]

Pls provide sample/real data to reproduce the same error.

[Walter Roberson]

what is class(d)?

[Andra Mihali]

class(d)

ans =

'table'

[Andra Mihali]

Also I could provide some subset of my data if necessary, but i was wondering why I get the same error even with the standard dataset from the Matlab helper file.

[Image Analyst]

Yes, please provide the data. Also, your link does not work. Please provide your code that was pulled from the help documentation so we can run it, plus the code that you modified to work with your data.

[Andra Mihali]

Link : https://www.mathworks.com/help/stats/fitglm.html

Code:

load carsmall
       d = table(MPG,Weight);
       d.Year = ordinal(Model_Year);
       lm = fitlm(d,'MPG ~ Year + Weight + Weight^2')

Will provide my code as well in a bit, but meanwhile the code from the help yields the error I copy-pasted above:(

[Andra Mihali]

https://www.mathworks.com/help/stats/fitglm.html

[Adam Danz]

What release of Matlab are you using?

In 2020b, the error that appears in your stack trace points to a line that is on line 83 in struct2table but your stack trace says it's on line 68 which is just an 'else' statement in 2020b.

[Andra Mihali]

2019a!

[Image Analyst]

It saves time if you read the posting guidelines on the question submission form and fill out all the fields. I'll fill out the Release for you on the right hand side of this page.

This runs fine for me in R2020b

load carsmall
d = table(MPG,Weight);
d.Year = ordinal(Model_Year);
lm = fitlm(d,'MPG ~ Year + Weight + Weight^2')

I get

lm =

Linear regression model:

MPG ~ 1 + Weight + Year + Weight^2

Estimated Coefficients:

Estimate SE tStat pValue

___________________ ____________________ _________________ ____________________

(Intercept) 54.2062873729581 4.71166738346461 11.5046931290593 2.66480529582129e-19

Weight -0.0164036290463721 0.0031249122854486 -5.24930863587977 1.02828937448822e-06

Year_76 2.08865700045782 0.714914020475329 2.92154992158236 0.00441367253657404

Year_82 8.18639613482828 0.815306703812534 10.0408792133648 2.63644864390097e-16

Weight^2 1.5573239444627e-06 4.94540084493951e-07 3.14903481697801 0.00223027408350944

Number of observations: 94, Error degrees of freedom: 89

Root Mean Squared Error: 2.78

R-squared: 0.885, Adjusted R-Squared: 0.88

F-statistic vs. constant model: 172, p-value = 5.52e-41

Perhaps your d is different. Can you attach it in a .mat file?

What does this say:

whos d

Mine says

Name Size Bytes Class Attributes

d 100x2 2802 table

[Andra Mihali]

Thank you so much! Am downloading 2020b now. The error is the same as before. Indeed my d is different:

    whos d
    

Name Size Bytes Class Attributes

d 100x3 3918 table

[Adam Danz]

Image Analyst and I must have filled in your Product/Release data at the same time!

I just tested the same commands in r17b and there were no errors. Please attach a mat file containing your inputs to fitlm.

[Adam Danz]

It should work in 19a, too.

What's in the table?

Could you show us the results of

head(d)
varfun(@class,d)

[Walter Roberson]

I have the suspicion that you have a function that is overshadowing the Mathworks version, but at the moment I do not know which function that might be. Experiment with

restoredefaultpath
rehash toolboxcache

and see if the code works properly. If it does then something that you have added to your path is interfering.

[Ive J]

Maybe the error comes from

Error in classreg.regr.TermsRegression/handleDataArgs (line 525)
                X = dataset2table(X);

Your MATLAB treats d as being a dataset, while you just showed d is a table (can you set a break point and verify if d indeed is treated as a table?)

Also as Adam pointed out, your struct2table throws an error on line 68, while line 83 takes care of this error in R2020b and also mine, R2019b. I couldn't track down any changes to this function in release notes.

load carsmall
d = table(MPG,Weight);
d.Year = ordinal(Model_Year);
lm = fitlm(d,'MPG ~ Year + Weight + Weight^2'); % no error is thrown in R2019b

whos d
  Name        Size            Bytes  Class    Attributes

  d         100x3              3556  table 

Maybe you can check also this.

 which struct2table -all

[Andra Mihali]

Thank you all so much for all your useful suggestions! Downloading 2020b fixed it, but going back to 2019a was still not running. Walter's suggestion fixed it also for 2019a:)

restoredefaultpath
rehash toolboxcache

[Andra Mihali]

Adam, btw the output to what you suggested was:

head(d)
Undefined function 'head' for input arguments of type 'table'.
 
>> varfun(@class,d)
Undefined function 'varfun' for input arguments of type 'table'.

[Image Analyst]

Andra, head() was added in R2016b and you have R2019a so you should have head(). What does this say:

which -all head

[Andra Mihali]

Post restoring the defaults with

restoredefaultpath
rehash toolboxcache

I get:

which -all head

/Applications/MATLAB_R2019a.app/toolbox/matlab/datatypes/@tabular/head.m % tabular method

/Applications/MATLAB_R2019a.app/toolbox/matlab/bigdata/@tall/head.m % tall method

/Applications/MATLAB_R2019a.app/toolbox/distcomp/parallel/@codistributed/head.m % codistributed method

/Applications/MATLAB_R2019a.app/toolbox/distcomp/gpu/@gpuArray/head.m

Also, to your previous suggestion:

head(d)

ans =

8×3 table

MPG Weight Year

___ ______ ____

18 3504 70

15 3693 70

18 3436 70

16 3433 70

17 3449 70

15 4341 70

14 4354 70

14 4312 70

varfun(@class, d)

ans =

1×3 table

class_MPG class_Weight class_Year

_________ ____________ __________

double double ordinal

[Adam Danz]

...and varfun was introduced in r2013b. Didn't WR's suggestion to restore the path and toolbox cache fix that?

[Andra Mihali]

Indeed, WR's suggestion fixed it (my first answer was from before restoring the path and toolboxcache). Thank you all again, relieved now:)