sheffieldml / gpmat Goto Github PK

Hello, I downloaded your code in order to implement HopLand (https://github.com/NetLand-NTU/HopLand).

The code isn't running, there appears to be a problem with the L2_distance function being undefined in isomapEmbed (line 11).

Any help would be greatly appreciated, I'm new to this!

anyone got the result of demRobotWireless4.m shows in the main page of GPmat as following?

And this is what I got,

not same with what they are showing, Can someone tell me what's going on?

Hello, I am a graduate student.
I download this Repo code and then I want to run this code like README.md.
But I have some problems....

My used MATLAB is R2021 version.
How can I solve this problem?

Hi,

Is GPmat designed to handle multivariate observation data? Many of the examples show single-variable observations.

Sterling

Hi,

I am trying to use the GPmat tool and check the plots, but when click on Figure1 (Latent Variables plot), I get an error as:

Attempted to access visualiseInfo.latentPos(-0.909202); index must be a positive integer or logical.

Error in lvmClassVisualise (line 25)
visualiseInfo.latentPos(visualiseInfo.dim1) = x;

Error while evaluating Figure WindowButtonDownFcn

Could it be because of the version of Matlab I have (2015a).

Thanks a lot.

Sinan

Dear all,

first of all, thank you very much for providing your code as open source. It's very impressive and educational. I have a quick question concerning the calculation of the first part of the variational lower bound in gpLogLikelihood.m:54:

  ll =  -0.5*(model.d*(-(model.N-model.k)*log(model.beta) ...
                       - model.logDetK_uu +model.logdetA) ...
              - (sum(sum(model.Ainv.*EET)) ...
                 -sum(sum(model.m.*model.m)))*model.beta)   ;

As far as I understand it, this is a reformulation of

 F = log[ N( y| 0, sigma^2 * I + Q_nn )] with Q_nn = K_nm / K_mm K_mn

Do you happen to have a article / paper for me which helps me understanding the mathmatical reformulation that are going on here?

Thank you very much in advance.
Best,
Simon

Hi,

Thank you for the great toolbox. I am new on GP and I have a question. Following the example of demstickfgplvm2.m, I have obtained a latent space with my own data, and I want to make new projections on it (using samples from a different dataset). Is there any function in the toolbox for this task? How can I use the model obtained by fgplvmOptimise.m to get new representations?

Any help is more than welcome.

I ran the demo demOil4.
The errors went from positive to negative and continue to decrease, like following.
Cycle 13 Error 20962.592172 Scale 2.083569e+04
Cycle 14 Error 12966.345081 Scale 1.041785e+04
Cycle 15 Error 5170.066683 Scale 5.208923e+03
Cycle 16 Error 2142.240905 Scale 2.604461e+03
Cycle 17 Error -1369.209243 Scale 1.302231e+03
Cycle 18 Error -2120.706700 Scale 6.511154e+02
Cycle 19 Error -3046.368921 Scale 3.255577e+02
Cycle 20 Error -3528.695085 Scale 1.627788e+02
Cycle 21 Error -3834.332228 Scale 8.138942e+01
Cycle 22 Error -4124.404594 Scale 4.069471e+01
Cycle 23 Error -4296.765671 Scale 2.034735e+01

It seems to go unbounded and won't stop.

I try to run the demo function demOilFgplvm1.m. However it throws the following error:

Error using *
Inner matrix dimensions must agree.

Error in ppcaEmbed (line 50)
X = Ycentre_u(:, 1:dims)_diag(1./sqrt(v(1:dims)));

Error in fgplvmCreate (line 30)
X = initFunc(Y, q);

Error in demOilFgplvm1 (line 21)
model = fgplvmCreate(latentDim, d, Y, options);

The error comes from the function /mltools/ppcaEmbed.m. It shows that Ycentre has dimension 1000x12, u has dim 1000x2 and diag(1./sqrt(v(1:dims))) has dim 2x2; therefore these dimensions do not match. Can you explain this?

My Matlab version is R2015a.

Hi,

I am trying to run your Latent Force Model scripts on my Mac but the mex files in kern/mex/ are not recognised because they were compiled for Windows.

So when I run lfmSample.m i get the following error:
"Undefined function or variable 'lfmComputeUpsilonMatrix'."

I am struggling to recompile the Fortran files myself (I have tried for a while) - are you able to compile them for OS X, or could they be re-written as m files?

Thanks

If I have the mean and covariance functions (via MATLAB's fitrgp() ), how can I sample from the posterior?

If one adds GPmat to MATLAB's path, then this causes the GPmat implementation of fcnchk to replace MATLAB's provided function named fcnchk. This causes errors if one needs to use fminunc, for example.

The following example code:

addpath('/path/to/GPmat');
myfunc = @(x) x^2;
x0 = 1;
fOpt = fminunc(myfunc, x0);

...produces the following output:

Error using fcnchk
Too many output arguments.

Error in optimfcnchk (line 83)
    [funfcn, idandmsg] = fcnchk(funstr,lenVarIn); %#ok<DFCNCHK>

Error in fminunc (line 249)
   funfcn = optimfcnchk(FUN,'fminunc',length(varargin),funValCheck,gradflag,hessflag);

Using MATLAB R2016b.

gsamp.m seems to be a function of interest to me, but I can't seem to find it in the repository.

Hi SheffieldML group,
Thank you very much for releasing such useful toolbox. I have downloaded the code and successfully run the first fgplvm demo examples. However, I fail to execute the demCmu35gplvmReconstruct.m:

Reference to non-existent field 'indexAll'.

Error in gpDynamicsSequenceLogLikeGradient (line 59)
tmp = zeros(size(gX,1),length(model.indexAll));

Error in fgplvmSequenceLogLikeGradient (line 44)
g = g + feval(model.dynamics, X);

Error in fgplvmSequenceGradient (line 26)
g = - fgplvmSequenceLogLikeGradient(model, X, Y, varargin{:});

Error in conjgrad (line 91)
gradnew = feval(gradf, x, varargin{:});

Error in fgplvmOptimiseSequence (line 51)
Xout = optim('fgplvmSequenceObjective', X(:)', options, ...

Error in fgplvmTaylorAngleErrors (line 59)
Xpred = fgplvmOptimiseSequence(model, Xinit, Ytest, 1, 1000);

The gplvm model was created using demCmu35gplvm1.m

Hello,I am a student from China.
I downloaded your code in order to use convolutional kernels in DeepGP（https://github.com/SheffieldML/GPmat).
Is there convolution kernel in the GPmat? If not. How do I add my kernel function?
Any help would be greatly appreciated, I'm new to this!

The 'neighbours' parameter is set 7 in the latest version for this function, while its value is 10 in its previous version. Why did you change this parameter ‘'neighbours" from 10 to 7?

Hello,
I'd like to use the GP-LVM as a dimensionality reduction technique - I wanted to start from the "oil data" demos, but basically running any of them ends up in this error:

demOilFgplvm1
Error using *
Inner matrix dimensions must agree.

Error in ppcaEmbed (line 50)
X = Ycentre_u(:, 1:dims)_diag(1./sqrt(v(1:dims)));

Error in fgplvmCreate (line 30)
X = initFunc(Y, q);

Error in demOilFgplvm1 (line 21)
model = fgplvmCreate(latentDim, d, Y, options);

Hi SheffieldML group,

Thankyou very much for providing such a thorough code which is useful at so many levels.There is a small issue/suggestion that I wanted to put forward though. I was getting errors when I simply downloaded GPmat and ran demos for FGPLVM, but adding the Netlab dependency solves the issue.

So, I wanted to suggest that if we add Netlab as part of GPmat it would just become a one stop shop where we wouldn't have to worry about errors and dependencies. I will mark it as resolved as soon as there is any comment because its basically a suggestion not an issue.

Thanks.

Hello,
the demStickFgplvm2 demo outputs the following on Mac OSX El Capitan, Matlab 2014, nor on Matlab2013-2016 on Linux SuseLeap 42.
After the maximum number of iterations are reached, the visualization shows this

Moving the mouse on top of figure 1, or pressing the radio button Run dynamics results in the following error on the matlab command prompt.

...
Cycle  998  Error -9216.916698  Scale 1.000000e-15
Cycle  999  Error -9217.442359  Scale 1.000000e-15
Cycle 1000  Error -9218.522775  Scale 1.000000e-15
Maximum number of iterations has been exceeded
Subscripted assignment dimension mismatch.

Error in stickVisualise (line 20)
handle(1) = plot3(vals(:, 1), vals(:, 2), vals(:, 3), '.');

Error in lvmVisualise (line 124)
visHandle = visualiseInfo.visualiseFunction(visData(1:dim), varargin{:});

Error in lvmResultsDynamic (line 23)
  lvmVisualise(model, lbls, [dataType 'Visualise'], [dataType 'Modify'], ...

Error in demStickFgplvm2 (line 48)
lvmResultsDynamic(model.type, dataSetName, experimentNo, 'stick', connect)
 
Reference to non-existent field 'visHandle'.

The optimization works but what is it supposed to be shown on figure 2?

Regards,