I would absolutely love to see you model the affect of DOAC blood thinners in prevention of stroke in atrial fibrillation. It’s a huge debate within the community where even with the application of the CHADvasc2 risk assessment some still refuse the risks of blood thinners and other take them to avoid the risk of stroke even when there risk of serious bleeding is higher than stroke. Most of the study’s are Pharma sponsored. Anyways great work as always.
Hello Norman, I very much enjoyed this video and helped me a lot with the comprehension of Bayesian networks. Unfortunately I ended up still with some questions I can't figure out: 1-, what would be the exact function of the green box given this case? The thing is that I cannot seem to find a reason for it's existence in this graph and I'd like to know if it influences any of the other nodes probabilities. 2- So in this graph the nodes could reversibly have an effect on the parent nodes? the way I had investigated is that the edges(arrows) visually demonstrate which "parent nodes" have a direct influence on other nodes, however, looking at this example, I can see that this influence can be done in a reversed order, for example, when checking the Dyspnoea or the Positive X-ray nodes, having an effect on the "diagnosis" nodes. Does this happen in all graphs? is it because of the green rectangle? hahaha Thanks in advance for anyone who takes their time to responding my questions! :)
The green box node is only included to simplify the specification of the probability tables of the nodes X-ray and Dyspnoea. The assumption is that tubercolous or cancer will have the same effect on X-ray and Dyspnoea so there is no need to seprately consider conditional probabilities. This is how the original BN model for this classic problem was presented by Jensen et al. so I just stuck with that. In general we often introduce what we call 'synthetic nodes' like this to avoid very large probability tables - usually it involves making assumptions about conditional indpendence which are not strictly correct but which are a good enough approximation. All of this is explained in our book www.normanfenton.com/bn-book
I also wanted to ask, so in this graph the nodes could reversibly have an effect on the parent nodes? the way I had investigated is that the edges(arrows) visually demonstrate which "parent nodes" have a direct influence on other nodes, however, looking at this example, I can see that this influence can be done in a reversed order, does this happen in all graphs? is it because of the green rectangle? hahah
Perhaps you could do another example of a Bayesian network to show the probability of detecting a new disease, which has pneumonia-like symptoms, in a country with very high rates of pneumonia before anybody died from it (and without any autopsies). And also, how somebody (in another country) without any samples from any "infected" person could develop a test that would be the gold standard test (used throughout the world) to determine (with no other clinical test being used) if somebody was a "case" for this new disease. Also, it may be worth adding into the mix that the main symptoms that people (around the world) were advised were associated with this disease are common with numerous other ailments (majority of which are not life-threatening) and also this new disease can have a high proportion of "cases" where the person has zero symptoms at all.
@@NormanFenton81 Professor, but to run a simulation regarding the issue would involve a considerable amount of data concerning public policies, wouldn't it?
@@NormanFenton81 Thanks for the reply. It's only recently I've discovered your channel and blog. I consider the papers on ONS data, which you and your Co-authors have written, to be excellent as they expose serious anomalies in data from a source that could previously be trusted. I was aware of videos you'd done regarding testing and false positives. I was not aware you'd done anything on the probabilities of detecting the signal of a purported new disease (similar to a long established infection) in a country that has high rates of such infection. I therefore apologise for my error. Keep up the great work you've being doing👍
So well explained and absolutly facinating. Is this used in criminal investigations as well, just to mention one other potential domain that comes to mind?
Obviously I recommend AgenaRisk www.agena.ai/ (note: I'm a Director!). But it is the only BN software that can properly handle relaibility assessments because it is the only one that properly supports continuous as well as discrete variables
Excellent presentation. Very few people, if any, have a good intuitive feel for conditional probabilities
Thank you. I was looking for exactly such a video, and surprisingly it took a few video tries to find it. :)
I would absolutely love to see you model the affect of DOAC blood thinners in prevention of stroke in atrial fibrillation. It’s a huge debate within the community where even with the application of the CHADvasc2 risk assessment some still refuse the risks of blood thinners and other take them to avoid the risk of stroke even when there risk of serious bleeding is higher than stroke. Most of the study’s are Pharma sponsored. Anyways great work as always.
Query: Does AgenaRisk use Gibbs sampling for its bayesian network solve?
Super helpful presentation!
Hello Norman, I very much enjoyed this video and helped me a lot with the comprehension of Bayesian networks. Unfortunately I ended up still with some questions I can't figure out:
1-, what would be the exact function of the green box given this case? The thing is that I cannot seem to find a reason for it's existence in this graph and I'd like to know if it influences any of the other nodes probabilities.
2- So in this graph the nodes could reversibly have an effect on the parent nodes? the way I had investigated is that the edges(arrows) visually demonstrate which "parent nodes" have a direct influence on other nodes, however, looking at this example, I can see that this influence can be done in a reversed order, for example, when checking the Dyspnoea or the Positive X-ray nodes, having an effect on the "diagnosis" nodes. Does this happen in all graphs? is it because of the green rectangle? hahaha
Thanks in advance for anyone who takes their time to responding my questions! :)
The green box node is only included to simplify the specification of the probability tables of the nodes X-ray and Dyspnoea. The assumption is that tubercolous or cancer will have the same effect on X-ray and Dyspnoea so there is no need to seprately consider conditional probabilities. This is how the original BN model for this classic problem was presented by Jensen et al. so I just stuck with that. In general we often introduce what we call 'synthetic nodes' like this to avoid very large probability tables - usually it involves making assumptions about conditional indpendence which are not strictly correct but which are a good enough approximation. All of this is explained in our book www.normanfenton.com/bn-book
Helpful, but what is up with the green box? Why is it there when there are separate boxes with cancer and tuberculosis?
I'm still trying to figure this out, please @NormanFenton, I'm sorry to bother, but, what would be the exact function of this box given this case?
I also wanted to ask, so in this graph the nodes could reversibly have an effect on the parent nodes? the way I had investigated is that the edges(arrows) visually demonstrate which "parent nodes" have a direct influence on other nodes, however, looking at this example, I can see that this influence can be done in a reversed order, does this happen in all graphs? is it because of the green rectangle? hahah
Perhaps you could do another example of a Bayesian network to show the probability of detecting a new disease, which has pneumonia-like symptoms, in a country with very high rates of pneumonia before anybody died from it (and without any autopsies). And also, how somebody (in another country) without any samples from any "infected" person could develop a test that would be the gold standard test (used throughout the world) to determine (with no other clinical test being used) if somebody was a "case" for this new disease.
Also, it may be worth adding into the mix that the main symptoms that people (around the world) were advised were associated with this disease are common with numerous other ailments (majority of which are not life-threatening) and also this new disease can have a high proportion of "cases" where the person has zero symptoms at all.
That's what most of my recent work has been doing!!!!
@@NormanFenton81 Professor, but to run a simulation regarding the issue would involve a considerable amount of data concerning public policies, wouldn't it?
@@NormanFenton81
Thanks for the reply. It's only recently I've discovered your channel and blog. I consider the papers on ONS data, which you and your Co-authors have written, to be excellent as they expose serious anomalies in data from a source that could previously be trusted.
I was aware of videos you'd done regarding testing and false positives. I was not aware you'd done anything on the probabilities of detecting the signal of a purported new disease (similar to a long established infection) in a country that has high rates of such infection. I therefore apologise for my error.
Keep up the great work you've being doing👍
So well explained and absolutly facinating. Is this used in criminal investigations as well, just to mention one other potential domain that comes to mind?
Yes. Quite a few of the papers listed here apply Bayesian networks to legal cases: www.normanfenton.com/law-and-forensics
What software was used in the video?
Which software can I download to run BN ? I really need to analyze human reliability assessment in maritime
Obviously I recommend AgenaRisk www.agena.ai/ (note: I'm a Director!). But it is the only BN software that can properly handle relaibility assessments because it is the only one that properly supports continuous as well as discrete variables
What is the best python package for doing this?
I'll think two times an Asian trip 😅
interesting but not at all informative.