Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Documentation on training of knowledge sources #51

Open
hagenw opened this issue Dec 2, 2016 · 8 comments
Open

Documentation on training of knowledge sources #51

hagenw opened this issue Dec 2, 2016 · 8 comments
Assignees
Labels

Comments

@hagenw
Copy link
Member

hagenw commented Dec 2, 2016

This issue is a high-level question on how to handle the documentation in general. The seperate training procedures will be handled independently afterwards.

From the list of learned models for our database I identified the following knowledge sources that use trained data:

DnnLocationKS
GenderRecognitionKS
GmmLocationKS
GmtkLocationKS
IdentityKS
NumSrcsKS
IdLocKS
ItdLocationKS
LocationKS
SegmentationKS
StreamSegregationKS

Looking at https://github.com/TWOEARS/blackboard-system/tree/master/src/knowledge_sources I assume this list might be wrong or is not complete. Could you please correct the list.

In addition, we have to discuss how and where we want to document the generation of this training data.
At the moment we have this section http://docs.twoears.eu/en/latest/blackboard/training/ where we have one subsection for training of the localisation knowledge sources and one for the identification knowledge sources, but both are not filled with text yet.
Should we go with this structure and place all the information under this section?

Another question is if we should handle https://github.com/TWOEARS/Auditory-Machine-Learning-Training-and-Testing-Pipeline different then the other as it is a very poweful and hugh tool by itself?

/CC: @ivo--t, @kashefy, @ningma97, @chrschy

@hagenw hagenw added the question label Dec 2, 2016
@hagenw hagenw self-assigned this Dec 2, 2016
@kashefy
Copy link
Member

kashefy commented Dec 2, 2016

@hagenw ,

IdentityKS is correct and should remain in this list
NumSrcsKS should be NumberOfSourcesKS
IdentityLocationKS is missing, these too are trained models.

AMLTTP is used for the training of IdentityKS, ``NumberOfSourcesKS, IdentityLocationKS`. We were thinking of putting the documentation of AMLTTP on the same level as binsim, AFE and blackboard. Is this ok?
The AMLTTP documentation would cover the details of training the above models. For http://docs.twoears.eu/en/latest/blackboard/training/, we would add very brief text for training each KS with a link to the more elaborate documentation in the AMLTTP docs. We think this allows the AMLTTP docs to be self contained, reduces duplication of text, but still allows the reader to find the necessary doc regardless of where they start from.

Does this make sense?

@ningma97
Copy link
Contributor

ningma97 commented Dec 2, 2016

@hagenw The list looks OK to me

@hagenw
Copy link
Member Author

hagenw commented Dec 2, 2016

Here is an update of the list, I incorporated @kashefy proposal and renamed NumSrcsKS and IdLocKS accordingly, in addition I removed GmtkLocationKS and LocationKS as those are deprecated:

DnnLocationKS     => ningma97
GenderRecognitionKS    => chrschy
GmmLocationKS    => ningma97
IdentityKS    => kashefy
NumberOfSourcesKS    => kashefy
IdentityLocationKS    => kashefy
ItdLocationKS    => hagenw
SegmentationKS (outdated?)   => chrschy
StreamSegregationKS    => chrschy

I think it is a good idea to create an entry for the AMLTTP on the main level as it is such a big tool now.
Regarding the brief texts on the training of the single knowledge sources, we could also think of including it directly to the documentation of the single knowledge sources. But on the other hand, you don't need to train them yourself in order to use them, so maybe we stick with http://docs.twoears.eu/en/latest/blackboard/training/ for the short texts.

So, I would propose to create a separate pull request for AMLTTP and then single ones for the texts under http://docs.twoears.eu/en/latest/blackboard/training/. I added names to the corresponding knowledge sources in the list above, those persons should create the corresponding pull requests and write the text. Please complain if you think I assigned you wrongly for some of the knowledge sources.

@ivo--t
Copy link
Member

ivo--t commented Dec 13, 2016

@hagenw , SegmentationKS is deprecated as far as I know (@chrschy to confirm), and StreamSegregationKS was coded by @chrschy hence I suggest he writes documentation about it.

@hagenw
Copy link
Member Author

hagenw commented Dec 14, 2016

Ok, I adjusted the list accordingly and will contact @chrschy.

@chrschy
Copy link
Member

chrschy commented Dec 14, 2016

SegmentationKS is indeed deprecated but the interfaces are more or less the same for StreamSegregationKS. I will update the corresponding documentation accordingly.

@ningma97
Copy link
Contributor

GenderRecognitionKS should be assigned to @chrschy

@hagenw
Copy link
Member Author

hagenw commented Dec 15, 2016

I updated the list accordingly.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

5 participants