Fixing errors occuring when training, and deprecated warning

[lamipaul]

No description provided.

[JulesFrancoise]

Hi @lamipaul

Thanks for contributing to XMM.

I don't really understand what your PR is fixing in the library. It seems that some edits are specific to your problem, not bugfixes for the library. For instance, forcing bimodal to false, or enforcing sequential training should not be hard coded there.

Could you specify here or in a new issue what errors occur during training, and what your PR is addressing?

Thanks

[lamipaul]

Hi Jules,
Indeed some modifications are specific to my problem and have mistakenly been hard coded there, they are taken out in the "get rid of fixed value" commit.

The general issues I was facing are found in the "fixing bugs" commit :

• Index out of range exception (with every value) when using shared_parameters->dimension.set()
• Coma instead of semicolon in xmmModelSingleClass.cpp
  training_status = TrainingEvent(this, label),
• Error with component assignment in xmmGMMSingleClass.cpp
• need of resizing the arrays in xmmHierarchicalHmm.cpp

[JulesFrancoise]

Hi @lamipaul

Regarding the bugs:

• It is actually not recommended to call shared_parameters->dimension.set(). Instead, you should change the dimensions only on the training set using ts.dimension.set(). I should probably document this somewhere. Could you tell me if you get an error by changing the dimension only for the training set?
• I couldn't reproduce the component assignment error. Can you provide more details about this?

Cheers,
Jules

[lamipaul]

Hi Jules,

shared_parameters->dimension.set() isn't personal code, it is called in the train method of xmmModel.hpp :
https://github.com/Ircam-RnD/xmm/blob/398cee77bdeae515c0140394da1b4302a4a4d322/src/core/model/xmmModel.hpp#L254

I am confused, I don't manage to reproduce the component assignment error myself. I'll let you know if I encounter it again.

Cheers,
Paul