Knowledge engineering with GeNIe

The front end.
Post Reply
charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

Knowledge engineering with GeNIe

Post by charlie » Tue Jul 02, 2019 2:26 am

I guess the majority of the forum users are using GeNIe/Smile for machine learning and wonder whether there is anyone using GeNIe for knowledge engineering like me. I'm also not sure whether it's appropriate to share bit generic experience in knowledge engineering in this forum instead of more technical aspects of GeNIe.

I found GeNIe is a powerful tool for business modelling through knowledge engineering. I recently built a hybrid model in GeNIe to model a complex task involving multiple actors behaving in technology, finance and social spaces. I attached a screenshot but am not in the position to share the model. It took three rounds of knowledge elicitation sessions and a deep dive into relevant documentation and datasets to get there. I was very inexperienced in knowledge elicitation at the beginning and made a huge (but not very efficient) effort to just get relevant experts in a room to talk meaningfully.

As you may see the struct of model has two parts - the upper part represents a process of factorization to breakdown the task into variables that we were able to handle. It was guided by an established social-behavioral theory so that expert's otherwise fragmented knowledge can be held together in perspective. The lower part is a hybrid model to support the factorization with qual and quan data.

To me the model quite well represented a key problem facing the task, but here is my challenge - I hope the task team could own it as a business support tool so that they could test different scenarios and keep refining it with new knowledge learnt, or just communicate their task to their stakeholders. It proves hard - GeNIe is not designed for this purpose though potentially fully capable of. It's quite scary to the non-tech audience and they don't have a licence to even view it on their PCs.

I think it could have been of great help even if I could just make some cosmetic edits such as to curve the links and make them connect to different points of a node box (like in Excel).

Also I found influence and sensitivity analysis are great features but unfortunately they don't work on hybrid model.

If anyone is doing knowledge engineering with GeNIe it'd be much appreciated if you could share bit your experience here e.g. how you promoted BN or GeNIe in particular for business enabling or decision support.

Cheers

Charlie
Attachments
model1.jpg
model1.jpg (229.85 KiB) Viewed 60 times
Last edited by charlie on Tue Jul 02, 2019 4:56 am, edited 2 times in total.

charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

Re: Knowledge engineering with GeNIe

Post by charlie » Tue Jul 02, 2019 4:46 am

Why can't I attache a JPG file? Got it - the previous size was too big.

marek [BayesFusion]
Site Admin
Posts: 270
Joined: Tue Dec 11, 2007 4:24 pm

Re: Knowledge engineering with GeNIe

Post by marek [BayesFusion] » Tue Jul 09, 2019 10:28 am

Hi Charlie,

Thanks again for sharing your experiences! In early days, Bayesian networks were almost exclusively based on knowledge engineering rather than being learned from data. Even though learning is powerful, it is precisely this subjective aspect of Bayesian network modeling and the ability to use subjective expert knowledge that makes this approach practical and useful. There are plenty of domains and situations in which there are very few or no data. Even if we have data, it is useful to refine learned models with expert knowledge, as data may be limited, biased, partly out of date, and will rarely tell the whole story. I have certainly constructed quite a number of models based on expert knowledge. You can see when working with GeNIe that while it has a comprehensive learning functionality, it stresses the interactive character of model construction and also the idea that a model is a document (e.g., you can put comments and yellow stick-it notes everywhere). Even if there are data available, building a model is never a one-shot process. Knowledge engineering for BNs is an important bottleneck of this business, so you are putting your efforts in the right directions. Good, intuitive GUI is an absolute must: It makes a big difference to spend 5 hours rather than 20 hours on building a model!

As far as sharing and testing your model with other people within your organization goes, we developed recently BayesBox, which is an interactive model repository that can be installed in the cloud or on the organization’s internal computers. Please check this out on our web sites:

http://repo.bayesfusion.com/

and a demo showing customization capabilities at

http://demo.bayesfusion.com/

BayesBox allows for uploading your models to the (local) repository and manage users (including password protection, if desired so. Other users cannot modify the models but they can work with them and test them for possible feedback.

We will eventually extend sensitivity analysis and link strengths to hybrid models. At the moment, one way to work around it is to discretize the hybrid model (Network/Discretize) and then use both functionalities.

If you find it useful, I will be happy to meet with you online and brainstorm about possible strategies to get other people in your organization on-board.

Thanks again for an insightful post -- it seems you are right on track as far as application of this methodology goes.
Cheers,

Marek

charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

Re: Knowledge engineering with GeNIe

Post by charlie » Wed Jul 10, 2019 2:12 am

Hi Marek

Thank you so much for your encouraging feedback, expert advice and kind offer of mentoring. Yes I'll defensively try out BayesBox. I'll also try network discretization for applying sensitivity analysis - sounds not very straightforward.

Yes it would be highly appreciated if I could have a online meeting with you at an appropriate time. How could we do that? My internal clients (actually front-line colleagues) are happy with the modelling we co-built so far and I encourage them to present it to the senior management and other stakeholders after we settle down in a new mega agency. I just couldn't help sharing the limited success to date with a bigger audience - indeed a powerful tool for business enabling and decision support. But I'm also conscious enough of barriers when people are confronted with a disruptive technology so I need to be patient.

Cheers

Charles

marek [BayesFusion]
Site Admin
Posts: 270
Joined: Tue Dec 11, 2007 4:24 pm

Re: Knowledge engineering with GeNIe

Post by marek [BayesFusion] » Wed Jul 10, 2019 3:53 pm

Hi Charlie,

I agree with you about the need to be patient when introducing this technology -- I have been teaching it for over 25 years :-). It is not easy at first but people love it once they understand it and it solves problems for them. By the way, BayesFusion offers free 30-day evaluation of BayesBox, so please do get in touch with us once you are ready to try it within your organization. We host the evaluation versions, so no pain at your end except for some hints about customizing the software to your needs. Access to BayesBox can be password controlled (this is one of the options).

I will send you my proposed times through a private message.
Cheers,

Marek

Post Reply