Re: [UAI] A perplexing problem - Version 2
On Mon, 23 Feb 2009, Francisco Javier Diez wrote: Konrad Scheffler wrote: I agree this is problematic - the notion of calibration (i.e. that you can say P(S|70%) = .7) does not really make sense in the subjective Bayesian framework where different individuals are working with different priors, because different individuals will have different posteriors and they can't all be equal to 0.7. I apologize if I have missed your point, but I think it does make sense. If different people have different posteriors, it means that some people will agree that the TWC reports are calibrated, while others will disagree. I think this is another way of saying the same thing - if you define the concept of calibration such that people will, depending on their priors, disagree over whether the reports are calibrated then it is still problematic to prescribe calibration in the problem formulation - because this will mean different things to different people. Unless you take TWC is calibrated to mean everyone has the same prior as TWC, which I don't think was the intention in the original question. In my opinion the source of confusion here is the use of a subjective Bayesian framework (i.e. one where the prior is not explicitly stated and is assumed to be different for different people). If instead we use an objective Bayesian framework where all priors are stated explicitly, the difficulties disappear. Who is right? In the case of unrepeatable events, this question would not make sense, because it is not possible to determine the true probability, and therefore whether a person or a model is calibrated or not is a subjective opinion (of an external observer). However, in the case of repeatable events--and I acknowledge that repeatability is a fuzzy concept--, it does make sense to speak of an objective probability, which can be identified with the relative frequency. Subjective probabilities that agree with the objective probability (frequency) can be said to be correct and models that give the correct probability for each scenario will be considered to be calibrated. If we accept that snow is a repeatable event, the all the individuals should agree on the same probability. If it is not, P(S|70%) may be different for each individual because having different priors and perhaps different likelihoods or even different structures in their models. I strongly disagree with this. The (true) relative frequency is not the same thing as the correct posterior. One can imagine a situation where the correct posterior (calculated from the available information) is very far from the relative frequency which one would obtain given the opportunity to perform exhaustive experiments. Probabilities (in any variant of the Bayesian framework) do not describe reality directly, they describe what we know about reality (typically in the absence of complete information). Coming back to the main problem, I agree again with Peter Szolovits in making the distinction between likelihood and posterior probability. a) If I take the TWC forecast as the posterior probability returned by a calibrated model (the TWC's model), then I accept that the probability of snow is 70%. b) However, if I take 70% probability of snow as a finding to be introduced in my model, then I should combine my prior with the likelihood ratio associated with this finding, and after some computation I will arrive at P(S|70%) = 0.70. [Otherwise, I would be incoherent with my assumption that the model used by the TWC is calibrated.] Of course, if I think that the TWC's model is calibrated, I do not need to build a model of TWC's reports that will return as an output the same probability estimate that I introduce as an input. Therefore I see no contradiction in the Bayesian framework. But this argument only considers the case where your prior is identical to TWC's prior. If your prior were _different_ from theirs (the more interesting case) then you would not agree that they are calibrated. ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem - Version 2
Konrad Scheffler wrote: I agree this is problematic - the notion of calibration (i.e. that you can say P(S|70%) = .7) does not really make sense in the subjective Bayesian framework where different individuals are working with different priors, because different individuals will have different posteriors and they can't all be equal to 0.7. I apologize if I have missed your point, but I think it does make sense. If different people have different posteriors, it means that some people will agree that the TWC reports are calibrated, while others will disagree. Who is right? In the case of unrepeatable events, this question would not make sense, because it is not possible to determine the true probability, and therefore whether a person or a model is calibrated or not is a subjective opinion (of an external observer). However, in the case of repeatable events--and I acknowledge that repeatability is a fuzzy concept--, it does make sense to speak of an objective probability, which can be identified with the relative frequency. Subjective probabilities that agree with the objective probability (frequency) can be said to be correct and models that give the correct probability for each scenario will be considered to be calibrated. If we accept that snow is a repeatable event, the all the individuals should agree on the same probability. If it is not, P(S|70%) may be different for each individual because having different priors and perhaps different likelihoods or even different structures in their models. --- Coming back to the main problem, I agree again with Peter Szolovits in making the distinction between likelihood and posterior probability. a) If I take the TWC forecast as the posterior probability returned by a calibrated model (the TWC's model), then I accept that the probability of snow is 70%. b) However, if I take 70% probability of snow as a finding to be introduced in my model, then I should combine my prior with the likelihood ratio associated with this finding, and after some computation I will arrive at P(S|70%) = 0.70. [Otherwise, I would be incoherent with my assumption that the model used by the TWC is calibrated.] Of course, if I think that the TWC's model is calibrated, I do not need to build a model of TWC's reports that will return as an output the same probability estimate that I introduce as an input. Therefore I see no contradiction in the Bayesian framework. Best regards, Javier - Francisco Javier Diez Phone: (+34) 91.398.71.61 Dpto. Inteligencia Artificial Fax: (+34) 91.398.88.95 UNED. c/Juan del Rosal, 16 http://www.ia.uned.es/~fjdiez 28040 Madrid. Spainhttp://www.cisiad.uned.es ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem - Last Version
Dear Paul, since I was in the consensus for my last response, I give you again my response to this new problem. The principle of my solution is always the same: to try to build a probabilistic model. a) I first reformulate the problem in more familiar notations for me, with a diagnosis D and 2 signs S1 et S2. The first report is S1, the second one is S2, and the diagnosis is location Y for X. Your data are: P(D/S1) = p1, and P(D/S2) = p2 Your question is: P(D/S1 and S2) = p12 = ? b) the problem is clearly underparametrized, so I would make 2 assumptions: A1: S1 and S2 are independant conditionnally to D and not D (it is possible not to make this assumption, but you have then to give a value for the dependence). A2: P(D)=P(not D) = 0.5 (this is for simplyfing the computations, but the solution is easy to compute also if you give another value for P(D)). c) now the solution is straihtforward: let's denote OR(12) = P(D/S1,S2)/P(not D/S1,S2), and OR(i)=P(D/Si)/P(not D/Si), i=1,2 we have by simple Bayes's formula and with assumptions A1 and A2: OR(1,2)= OR(1) OR(2) and p12 = OR(1) OR(2) / (1 + OR(1) OR(2) ) I don't know if it is your answer, since it is a very simple one... I think it is the solution based on the simplest probabilistic computations. sincerely yours Jean-louis Quoting Lehner, Paul E. pleh...@mitre.org: Austin, Jean-Lous, Konrad, Peter Thank you for your responses. They are very helpful. Your consensus view seems to be that when receiving evidence in the form of a single calibrated judgment, one should not update personal judgments by using Bayes rule. This seems incoherent (from a strict Bayesian perspective) unless perhaps one explicitly represents the overlap of knowledge with the source of the calibrated judgment (which may not be practical.) Unfortunately this is the conclusion I was afraid we would reach, because it leads me to be concerned that I have been giving some bad advice about applying Bayesian reasoning to some very practical problems. Here is a simple example. Analyst A is trying to determine whether X is at location Y. She has two principal evidence items. The first is a report from a spectral analyst that concludes based on the match to the expected spectral signature I conclude with high confidence that X is at location Y. The second evidence is a report from a chemical analyst who asserts, based on the expected chemical composition that is typically associated with X, I conclude with moderate confidence that X is at location Y. How should analyst A approach her analysis? Previously I would have suggested something like this. Consider each evidence item in turn. Assume that X is at location Y. What are the chances that you would receive a 'high confidence' report from the spectral analyst, ... a report of 'moderate confidence' from the chemical analyst. Now assume X is not a location Y, In other words I would have lead the analyst toward some simple instantiation of Bayes inference. But clearly the spectral and chemical analyst are using more than just the sensor data to make their confidence assessments. In part they are using the same background knowledge that Analyst A has. Furthermore both the spectral and chemical analysts are good at their job, their confidence judgments are reasonably calibrated. This is just like the TWC problem only more complex. So if Bayesian inference is inappropriate for the TWC problem, is it also inappropriate here? Is my advice bad? Paul From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Lehner, Paul E. Sent: Monday, February 16, 2009 11:40 AM To: uai@ENGR.ORST.EDU Subject: Re: [UAI] A perplexing problem - Version 2 UAI members Thank you for your many responses. You've provided at least 5 distinct answers which I summarize below. (Answer 5 below is clearly correct, but leads me to a new quandary.) Answer 1: 70% chance of snow is just a label and conceptually should be treated as XYZ. In other words don't be fooled by the semantics inside the quotes. My response: Technically correct, but intuitively unappealing. Although I often council people on how often intuition is misleading, I just couldn't ignore my intuition on this one. Answer 2: The forecast 70% chance of snow is ill-defined My response: I agree, but in this case I was more concerned about the conflict between math and intuition. I would be willing to accept any well-defined forecasting statement. Answer 3: The reference set winter days is the wrong reference set. My response: I was just trying to give some justification to my subjective prior. But this answer does point out a distinction between base rates and subjective priors. This distinction relates to my new quandary below so please read
Re: [UAI] A perplexing problem
Dear All, We may not compare "70%" of TWC prediction with Paul's 34%. Simply beacause, as Paul assumed , TWC is right only at a ratio of 1/10 (their 70% "prediction" happens at 10%, true positive compared to 1% "false positive") ! Given the uncertainty of TWC "70%" predictions, the Paul's 34% would have its own uncertainty (belief) that will be observable from the feedback. One is often surprized for the same reasons as Paul. That was my case in the "Alarm/Earthquake.." problem from many sources, e.g. P. Norvig S. Russel. regards. Alex Le 16/02/09 20:49, Agosta, John M a crit: All - The "Bayes ratio" (or odds ratio) interpretation of Bayes rule is enlightening, since it reveals the strength of evidence in a way not clear from just looking at the probabilities. A 5% prior chance becomes odds of 1:19 against snow. With Paul's assigned sensitivity (probability of snow forecast given it will snow) of 10%, the evidence of a positive forcast has an odds ratio of 10:1 in favor of snow. Expressed, for instance in a scale suggested by Kass Raftery this counts as not particularly strong positive evidence. Not surprisingly the combination of 1:19 prior against and a 10:1 odds for results in less than even odds for snow. ___ John Mark Agosta, Intel Research -Original Message- From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Paul Snow Sent: Monday, February 16, 2009 3:24 AM To: uai@engr.orst.edu Subject: Re: [UAI] A perplexing problem Dear Paul, If the Weather Channel is Bayesian, then say they used that empricial prior that you did (5%), and they observed evidence E to arrive at their 70% for the snow S given E. Their Bayes' ratio is 44.3. Yours, effectively, is 10 (assuming that the event "They say 70%" coincides with "They observe evidence with a Bayes ratio in the forties" - that is, they agree with you about the empirical prior and are Bayesian). So, having effectively disagreed with them about the import of what they observed, you also disagreed with them about the conclusion. Hope that helps, Paul 2009/2/13 Lehner, Paul E. pleh...@mitre.org: I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a "70% chance of snow" on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts "70% chance of snow" about 10% of the time 6. On winter days that it does not snow, TWC forecasts "70% chance of snow" about 1% of the time. So now from P(S)=.05; P("70%"|S)=.10; and P("70%"|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|"70%") = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai -- Alexandre Saidi Maitre de Confrences Ecole Centrale de Lyon-Dp. MI Tl : 0472186530, Fax : 0472186443 begin:vcard fn:A-S Saidi n:A-S Saidi;A-S org:Ecole Centrale de Lyon;LIRIS-UMR 5205 CNRS adr;quoted-printable:;;36 Av. Guy de Collongue;Ecully;Rh=C3=B4ne;69134;France email;internet:alexandre.sa...@liris.cnrs.fr title;quoted-printable:Ma=C3=AEtre de Conf=C3=A9rences tel;work:04 72 18 65 30 tel;fax:04 72 18 64 43 x-mozilla-html:TRUE version:2.1 end:vcard ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem - Version 2
I agree this is problematic - the notion of calibration (i.e. that you can say P(S|70%) = .7) does not really make sense in the subjective Bayesian framework where different individuals are working with different priors, because different individuals will have different posteriors and they can't all be equal to 0.7. Instead, you need a notion of calibration with respect to a particular prior. Hopefully the TWC forecasts are calibrated with respect to their own prior (otherwise they are reporting something other than what they believe). In this case your subjective posterior P(S|70%) will only be equal to .7 if your prior happens to be identical to theirs. Hope this helps, Konrad Consider the following revised version. The TWC problem 1. Question: What is the chance that it will snow next Monday? 2. My subjective prior: 5% 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. Notice that I did not justify by subjective prior with a base rate. From P(S)=.05 and P(S|70%) = .7 I can deduce that P(70%|S)/P(70%|~S) = 44.33. So now I can deduce from my prior and evidence odds that P(S|70%) = .7. But this seems silly. Suppose my subjective prior was 20%. Then P(70%|S)/P(70%|~S) = 9.3 and again I can deduce P(S|70%)=.7. My latest quandary is that it seems odd that my subjective conditional probability of the evidence should depend on my subjective prior. This may be coherent, but is too counter intuitive for me to easily accept. It would also suggest that when receiving a single evidence item in the form of a judgment from a calibrated source, my posterior belief does not depend on my prior belief. In effect, when forecasting snow, one should ignore priors and listen to The Weather Channel. Is this correct? If so, does this bother anyone else? ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem - Version 2
Paul, your restated problem reminds me of one I encountered in medicine in the 1980's. When an internist sends a patient's sample to a pathologist and the pathologist says 90% chance of cancer, how is the internist supposed to interpret that answer in light of his own priors? Empirically, what we discovered is that pathologists don't (or at least didn't) have a clear methodology for addressing such problems. Some tried to be scrupulously untainted by any evidence about the patient other than the submitted sample, whereas others would read the entire chart to understand the context in which they were interpreting the sample. My assumption from this is that the first group were trying to judge something like a likelihood, conditional probability, or conditional odds, whereas the second were giving posteriors. If TWC is giving posteriors, integrating everything known about weather in your area based on their extensive professional knowledge (which presumably includes all the almanac information that goes into your prior judgments), then you should simply accept their answer. This is like the second group of pathologists. If, however, they are giving something like conditional odds (how much more likely would this weather pattern be if it turns out to snow Monday than if it does not), then it's most appropriate to do your Bayesian combination. --Pete Szolovits On Feb 16, 2009, at 11:39 AM, Lehner, Paul E. wrote: ... Consider the following revised version. The TWC problem 1. Question: What is the chance that it will snow next Monday? 2. My subjective prior: 5% 3. Evidence: The Weather Channel (TWC) says there is a “70% chance of snow” on Monday. 4. TWC forecasts of snow are calibrated. Notice that I did not justify by subjective prior with a base rate. From P(S)=.05 and P(S|”70%”) = .7 I can deduce that P(“70%”|S)/ P(“70%”|~S) = 44.33. So now I can “deduce” from my prior and evidence odds that P(S|”70%”) = .7. But this seems silly. Suppose my subjective prior was 20%. Then P(“70%”|S)/P(“70%”|~S) = 9.3 and again I can “deduce” P(S|”70%”)=.7. My latest quandary is that it seems odd that my subjective conditional probability of the evidence should depend on my subjective prior. This may be coherent, but is too counter intuitive for me to easily accept. It would also suggest that when receiving a single evidence item in the form of a judgment from a calibrated source, my posterior belief does not depend on my prior belief. In effect, when forecasting snow, one should ignore priors and listen to The Weather Channel. Is this correct? If so, does this bother anyone else? paull From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Lehner, Paul E. Sent: Friday, February 13, 2009 4:29 PM To: uai@ENGR.ORST.EDU Subject: [UAI] A perplexing problem I was working on a set of instructions to teach simple two- hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can’t be a new problem so I’m hoping someone will clear things up for me. The problem 5. Question: What is the chance that it will snow next Monday? 6. My prior: 5% (because it typically snows about 5% of the days during the winter) 7. Evidence: The Weather Channel (TWC) says there is a “70% chance of snow” on Monday. 8. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 9. On winter days that it snows, TWC forecasts “70% chance of snow” about 10% of the time 10. On winter days that it does not snow, TWC forecasts “70% chance of snow” about 1% of the time. So now from P(S)=.05; P(“70%”|S)=.10; and P(“70%”|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|”70%”) = . 3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem - Last Version
Austin, Jean-Lous, Konrad, Peter Thank you for your responses. They are very helpful. Your consensus view seems to be that when receiving evidence in the form of a single calibrated judgment, one should not update personal judgments by using Bayes rule. This seems incoherent (from a strict Bayesian perspective) unless perhaps one explicitly represents the overlap of knowledge with the source of the calibrated judgment (which may not be practical.) Unfortunately this is the conclusion I was afraid we would reach, because it leads me to be concerned that I have been giving some bad advice about applying Bayesian reasoning to some very practical problems. Here is a simple example. Analyst A is trying to determine whether X is at location Y. She has two principal evidence items. The first is a report from a spectral analyst that concludes based on the match to the expected spectral signature I conclude with high confidence that X is at location Y. The second evidence is a report from a chemical analyst who asserts, based on the expected chemical composition that is typically associated with X, I conclude with moderate confidence that X is at location Y. How should analyst A approach her analysis? Previously I would have suggested something like this. Consider each evidence item in turn. Assume that X is at location Y. What are the chances that you would receive a 'high confidence' report from the spectral analyst, ... a report of 'moderate confidence' from the chemical analyst. Now assume X is not a location Y, In other words I would have lead the analyst toward some simple instantiation of Bayes inference. But clearly the spectral and chemical analyst are using more than just the sensor data to make their confidence assessments. In part they are using the same background knowledge that Analyst A has. Furthermore both the spectral and chemical analysts are good at their job, their confidence judgments are reasonably calibrated. This is just like the TWC problem only more complex. So if Bayesian inference is inappropriate for the TWC problem, is it also inappropriate here? Is my advice bad? Paul From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Lehner, Paul E. Sent: Monday, February 16, 2009 11:40 AM To: uai@ENGR.ORST.EDU Subject: Re: [UAI] A perplexing problem - Version 2 UAI members Thank you for your many responses. You've provided at least 5 distinct answers which I summarize below. (Answer 5 below is clearly correct, but leads me to a new quandary.) Answer 1: 70% chance of snow is just a label and conceptually should be treated as XYZ. In other words don't be fooled by the semantics inside the quotes. My response: Technically correct, but intuitively unappealing. Although I often council people on how often intuition is misleading, I just couldn't ignore my intuition on this one. Answer 2: The forecast 70% chance of snow is ill-defined My response: I agree, but in this case I was more concerned about the conflict between math and intuition. I would be willing to accept any well-defined forecasting statement. Answer 3: The reference set winter days is the wrong reference set. My response: I was just trying to give some justification to my subjective prior. But this answer does point out a distinction between base rates and subjective priors. This distinction relates to my new quandary below so please read on. Answer 4: The problem inherently requires more variables and cannot be treated as a simple single evidence with two hypotheses problem. My response: Actually I was concerned that this was the answer. As it may have implied that using Bayes to evaluate a single evidence item was impractical for the community of analysts I'm working with. Fortunately ... Answer 5: The problem statement was inherently incoherent. Many of you pointed out that if TWC predicts 70% snow on 10% of the days that it snows and on 1% of days that it does not snow, and a 5% base rate for snow, then the P(70% snow snow) is .005 and P(70% snow ~snow) = .0095. So for the days that TWC says 70% snow it actually snows on a little over 34% of the days. Clearly my assertion that TWC is calibrated is incoherent relative to the rest of the problem statement. The problem was not underspecified, it was over specified. (I hope I did the math correctly.) My response: Thanks for pointing this out. I'm embarrassed that I didn't notice this myself. Though this clearly solves my initial concern it leads me to an entirely new quandary. Consider the following revised version. The TWC problem 1. Question: What is the chance that it will snow next Monday? 2. My subjective prior: 5% 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. Notice that I did not justify by subjective
Re: [UAI] A perplexing problem - Version 2
This time, the probabilistic model is underspecified, since it has 2 probabilities, but it is not important for answering the question since the answer to question 1 is is propositions 3 et 4: if TWC forecasts are calibrated then P(S/70%) = 70%, and prior 2 plays no role. You find this since with 2 different values, you find always 0.7. I think that you should not try to make computations with the prior since you have aleadythe answer in the problem formulation. Sincerely yours Jean-Louis Quoting Lehner, Paul E. pleh...@mitre.org: UAI members Thank you for your many responses. You've provided at least 5 distinct answers which I summarize below. (Answer 5 below is clearly correct, but leads me to a new quandary.) Answer 1: 70% chance of snow is just a label and conceptually should be treated as XYZ. In other words don't be fooled by the semantics inside the quotes. My response: Technically correct, but intuitively unappealing. Although I often council people on how often intuition is misleading, I just couldn't ignore my intuition on this one. Answer 2: The forecast 70% chance of snow is ill-defined My response: I agree, but in this case I was more concerned about the conflict between math and intuition. I would be willing to accept any well-defined forecasting statement. Answer 3: The reference set winter days is the wrong reference set. My response: I was just trying to give some justification to my subjective prior. But this answer does point out a distinction between base rates and subjective priors. This distinction relates to my new quandary below so please read on. Answer 4: The problem inherently requires more variables and cannot be treated as a simple single evidence with two hypotheses problem. My response: Actually I was concerned that this was the answer. As it may have implied that using Bayes to evaluate a single evidence item was impractical for the community of analysts I'm working with. Fortunately ... Answer 5: The problem statement was inherently incoherent. Many of you pointed out that if TWC predicts 70% snow on 10% of the days that it snows and on 1% of days that it does not snow, and a 5% base rate for snow, then the P(70% snow snow) is .005 and P(70% snow ~snow) = .0095. So for the days that TWC says 70% snow it actually snows on a little over 34% of the days. Clearly my assertion that TWC is calibrated is incoherent relative to the rest of the problem statement. The problem was not underspecified, it was over specified. (I hope I did the math correctly.) My response: Thanks for pointing this out. I'm embarrassed that I didn't notice this myself. Though this clearly solves my initial concern it leads me to an entirely new quandary. Consider the following revised version. The TWC problem 1. Question: What is the chance that it will snow next Monday? 2. My subjective prior: 5% 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. Notice that I did not justify by subjective prior with a base rate. From P(S)=.05 and P(S|70%) = .7 I can deduce that P(70%|S)/P(70%|~S) = 44.33. So now I can deduce from my prior and evidence odds that P(S|70%) = .7. But this seems silly. Suppose my subjective prior was 20%. Then P(70%|S)/P(70%|~S) = 9.3 and again I can deduce P(S|70%)=.7. My latest quandary is that it seems odd that my subjective conditional probability of the evidence should depend on my subjective prior. This may be coherent, but is too counter intuitive for me to easily accept. It would also suggest that when receiving a single evidence item in the form of a judgment from a calibrated source, my posterior belief does not depend on my prior belief. In effect, when forecasting snow, one should ignore priors and listen to The Weather Channel. Is this correct? If so, does this bother anyone else? paull From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Lehner, Paul E. Sent: Friday, February 13, 2009 4:29 PM To: uai@ENGR.ORST.EDU Subject: [UAI] A perplexing problem I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 5. Question: What is the chance that it will snow next Monday? 6. My prior: 5% (because it typically snows about 5% of the days during the winter) 7. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 8. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 9. On winter days that it snows, TWC
Re: [UAI] A perplexing problem
Dear Paul, if you consider TWC prediction as a part of the probabilistic model, you get 4 probabilities for modelling a model which needs 3 probabilities to be specified. (the model is given by the 2-way table given by (Snow/not snow and snow prediction of 70%/not snow prediction of 70%). The problem is that in this model the 4 numbers you give are inconsistent, so, when you accept the probabilities 2,5, and 6, you find that P(S/prediction of snow is 70%) = 0.34, which is not consistent with propositions 3 and 4. If you accept probabilities 2,3 and 5, for example, you find that Pr(prediction of snow = 70% / not snow) = 0.002, and not 0.01 as given in the problem. Hope that helps also, Jean-Louis 2009/2/13 Lehner, Paul E. pleh...@mitre.org: I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time 6. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time. So now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem - Version 2
UAI members Thank you for your many responses. You've provided at least 5 distinct answers which I summarize below. (Answer 5 below is clearly correct, but leads me to a new quandary.) Answer 1: 70% chance of snow is just a label and conceptually should be treated as XYZ. In other words don't be fooled by the semantics inside the quotes. My response: Technically correct, but intuitively unappealing. Although I often council people on how often intuition is misleading, I just couldn't ignore my intuition on this one. Answer 2: The forecast 70% chance of snow is ill-defined My response: I agree, but in this case I was more concerned about the conflict between math and intuition. I would be willing to accept any well-defined forecasting statement. Answer 3: The reference set winter days is the wrong reference set. My response: I was just trying to give some justification to my subjective prior. But this answer does point out a distinction between base rates and subjective priors. This distinction relates to my new quandary below so please read on. Answer 4: The problem inherently requires more variables and cannot be treated as a simple single evidence with two hypotheses problem. My response: Actually I was concerned that this was the answer. As it may have implied that using Bayes to evaluate a single evidence item was impractical for the community of analysts I'm working with. Fortunately ... Answer 5: The problem statement was inherently incoherent. Many of you pointed out that if TWC predicts 70% snow on 10% of the days that it snows and on 1% of days that it does not snow, and a 5% base rate for snow, then the P(70% snow snow) is .005 and P(70% snow ~snow) = .0095. So for the days that TWC says 70% snow it actually snows on a little over 34% of the days. Clearly my assertion that TWC is calibrated is incoherent relative to the rest of the problem statement. The problem was not underspecified, it was over specified. (I hope I did the math correctly.) My response: Thanks for pointing this out. I'm embarrassed that I didn't notice this myself. Though this clearly solves my initial concern it leads me to an entirely new quandary. Consider the following revised version. The TWC problem 1. Question: What is the chance that it will snow next Monday? 2. My subjective prior: 5% 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. Notice that I did not justify by subjective prior with a base rate. From P(S)=.05 and P(S|70%) = .7 I can deduce that P(70%|S)/P(70%|~S) = 44.33. So now I can deduce from my prior and evidence odds that P(S|70%) = .7. But this seems silly. Suppose my subjective prior was 20%. Then P(70%|S)/P(70%|~S) = 9.3 and again I can deduce P(S|70%)=.7. My latest quandary is that it seems odd that my subjective conditional probability of the evidence should depend on my subjective prior. This may be coherent, but is too counter intuitive for me to easily accept. It would also suggest that when receiving a single evidence item in the form of a judgment from a calibrated source, my posterior belief does not depend on my prior belief. In effect, when forecasting snow, one should ignore priors and listen to The Weather Channel. Is this correct? If so, does this bother anyone else? paull From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Lehner, Paul E. Sent: Friday, February 13, 2009 4:29 PM To: uai@ENGR.ORST.EDU Subject: [UAI] A perplexing problem I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 5. Question: What is the chance that it will snow next Monday? 6. My prior: 5% (because it typically snows about 5% of the days during the winter) 7. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 8. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 9. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time 10. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time. So now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.orgmailto:pleh...@mitre.org ___
Re: [UAI] A perplexing problem
All - The Bayes ratio (or odds ratio) interpretation of Bayes rule is enlightening, since it reveals the strength of evidence in a way not clear from just looking at the probabilities. A 5% prior chance becomes odds of 1:19 against snow. With Paul's assigned sensitivity (probability of snow forecast given it will snow) of 10%, the evidence of a positive forcast has an odds ratio of 10:1 in favor of snow. Expressed, for instance in a scale suggested by Kass Raftery this counts as not particularly strong positive evidence. Not surprisingly the combination of 1:19 prior against and a 10:1 odds for results in less than even odds for snow. ___ John Mark Agosta, Intel Research -Original Message- From: uai-boun...@engr.orst.edu [mailto:uai-boun...@engr.orst.edu] On Behalf Of Paul Snow Sent: Monday, February 16, 2009 3:24 AM To: uai@engr.orst.edu Subject: Re: [UAI] A perplexing problem Dear Paul, If the Weather Channel is Bayesian, then say they used that empricial prior that you did (5%), and they observed evidence E to arrive at their 70% for the snow S given E. Their Bayes' ratio is 44.3. Yours, effectively, is 10 (assuming that the event They say 70% coincides with They observe evidence with a Bayes ratio in the forties - that is, they agree with you about the empirical prior and are Bayesian). So, having effectively disagreed with them about the import of what they observed, you also disagreed with them about the conclusion. Hope that helps, Paul 2009/2/13 Lehner, Paul E. pleh...@mitre.org: I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time 6. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time. So now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem
Peter Szolovits wrote: If TWC is really calibrated, then your conditions 5 and 6 are false, no? I agree with Peter's solution. If I build a model for this problem, it must contain at least two variables: Snow and TWC_report. According with my model, the TWC forecasts are calibrated if and only if P(Snow=yes|TWC_report=x) = x, by definition of calibration. Regards, Javier - Francisco Javier Diez Phone: (+34) 91.398.71.61 Dpto. Inteligencia Artificial Fax: (+34) 91.398.88.95 UNED. c/Juan del Rosal, 16 http://www.ia.uned.es/~fjdiez 28040 Madrid. Spainhttp://www.cisiad.uned.es ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem
1. Note that you haven't really used the 70% at all. You could restate the problem with any other statement you liked in there. 2. Your basic reasoning is correct. However, your modelling choice seems poor. I would try replacing TWC forecasts 70% chance of snow with TWC forecasts 70% OR MORE chance of snow. With this replacement, the math is correct, but if TWC only forecasts 70% or more chance of snow 10% of the time when it's gona snow, TWC isn't actually good at weather forecasting. Cheers, rif I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time 6. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time. So now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.orgmailto:pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem
Paul, I'm not aware of this being discussed anywhere but my observation is that the information given makes TWC quite lousy -- the probability of the forecast 70% chance of snow is much too high when there is no snow. It is a very specific piece of forecast and I would expect this probability to be very small given that there is actually going to be no snow. When you reduce this conditional probability, the forecast is going to be more along the lines that you would expect. I'm attaching a GeNIe model capturing your problem. To open it, download GeNIe from http://genie.sis.pitt.edu/. Cheers, Marek -- Marek J. Druzdzelhttp://www.pitt.edu/~druzdzel Lehner, Paul E. wrote: I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can’t be a new problem so I’m hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a “70% chance of snow” on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts “70% chance of snow” about 10% of the time 6. On winter days that it does not snow, TWC forecasts “70% chance of snow” about 1% of the time. So now from P(S)=.05; P(“70%”|S)=.10; and P(“70%”|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|”70%”) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.org mailto:pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai ?xml version=1.0 encoding=ISO-8859-1? smile version=1.0 id=SnowForecast numsamples=1000 discsamples=1 nodes cpt id=It_Snows_on_Monday state id=Snows / state id=DoesNotSnow / probabilities0.05 0.95/probabilities /cpt cpt id=Forecast70 state id=Snow70 / state id=Other / parentsIt_Snows_on_Monday/parents probabilities0.1 0.9 0.01 0.99/probabilities /cpt /nodes extensions genie version=1.0 app=GeNIe 2.0.3306.0 name=Paul Lehnerapos;s problem faultnameformat=nodestate node id=It_Snows_on_Monday nameIt Snows on Monday/name interior color=e5f6f7 / outline color=80 / font color=00 name=Arial size=10 bold=true / position142 14 250 81/position barchart active=true width=368 height=64 / /node node id=Forecast70 nameThe Weather Channel Forecasts 70% Chance of Snow/name interior color=e5f6f7 / outline color=80 / font color=00 name=Arial size=10 bold=true / position147 213 249 276/position barchart active=true width=368 height=64 / /node textbox captionPaul Lehnerapos;s problem lt;pleh...@mitre.orggt;\n\nThe problem:\n\n1. Question: What is the chance that it will snow next Monday?\n2. My prior: 5% (because it typically snows about 5% of the days during the winter)\n3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday.\n4. TWC forecasts of snow are calibrated.\n\nMy initial answer is to claim that this problem is underspecified. So I add\n\n5. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time\n6. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time.\n\nSo now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448.\n\nNow it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do./caption font color=00 name=Arial size=10 bold=true / position406 16 1033 320/position /textbox /genie /extensions /smile ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem
Hi Paul, Your calculation is correct, but the numbers in the example are odd. If TWC really only manage to predict snow 10% of the time (90% false negative rate), you would be right not to assign much value to their predictions (you do assign _some_, hence the seven-fold increase from your prior to your posterior, but with prediction performance like that TWC cannot possibly think there is really a 70% chance of snow). Change the 10% true positives to 90%, and your posterior goes up to 82.6% - much more believable. Also, it's important not to think the figure of 70% has any bearing on the problem. I appreciate that you put it in as a red herring to challenge the students, but be aware that it may also lead to confusion. Konrad On Fri, 13 Feb 2009, Lehner, Paul E. wrote: I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time 6. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time. So now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.orgmailto:pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
Re: [UAI] A perplexing problem
Hi Paul, Your calculations are correct (although I note you really mean P(70%|not S) = 0.01 in the calc below). ^^^ Sometimes it helps to think about what the numbers actually mean. First 0.05 prob of snow is quite a low prior. You need to have quite certain evidence to move that up higher. A posterior of 0.35 means that snow is now *7 times* more likely given the evidence than it was before you knew anything, which is still quite a large shift up. It *sounds* like you have strong evidence with TWC 70% chance of snow. However, you also have a conditional probability that even when there is snow, TWC only says 70% chance of snow one in 10 times. That means that 9 in ten times it doesn't say that. So when you entered such evidence it gets discounted (because it is so often wrong!). Another side point about the way you have modelled this problem is your second variable is TWC70%ChanceOfSnow, a true/false variable. So TWC's confidence isn't really being modelled in the Bayesian updating, only in the way you've structured your variables. It might be better instead to have the second variable be TWCPredictsSnow (True/False) and then incorporate their 70% confidence as virtual (uncertain) evidence on that variable. But then you'd need to know P(TWCPredictsSnow|Snow) and P(TWCPredictsSnow|notSnow)... Hope this helps. regards, Ann On Fri, Feb 13, 2009 at 04:28:41PM -0500, Lehner, Paul E. wrote: I was working on a set of instructions to teach simple two-hypothesis/one-evidence Bayesian updating. I came across a problem that perplexed me. This can't be a new problem so I'm hoping someone will clear things up for me. The problem 1. Question: What is the chance that it will snow next Monday? 2. My prior: 5% (because it typically snows about 5% of the days during the winter) 3. Evidence: The Weather Channel (TWC) says there is a 70% chance of snow on Monday. 4. TWC forecasts of snow are calibrated. My initial answer is to claim that this problem is underspecified. So I add 5. On winter days that it snows, TWC forecasts 70% chance of snow about 10% of the time 6. On winter days that it does not snow, TWC forecasts 70% chance of snow about 1% of the time. So now from P(S)=.05; P(70%|S)=.10; and P(70%|S)=.01 I apply Bayes rule and deduce my posterior probability to be P(S|70%) = .3448. Now it seems particularly odd that I would conclude there is only a 34% chance of snow when TWC says there is a 70% chance. TWC knows so much more about weather forecasting than I do. What am I doing wrong? Paul E. Lehner, Ph.D. Consulting Scientist The MITRE Corporation (703) 983-7968 pleh...@mitre.orgmailto:pleh...@mitre.org ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai -- A/Prof. Ann Nicholson Clayton School_--_|\ www.csse.monash.edu.au/~annn/ of Information Technology, / \ phone: +61 3 9905 5211 Monash University, VIC 3800 \_.--.*/ fax: +61 3 9905 5146 Australia v ann.nichol...@infotech.monash.edu.au ___ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai