Re: [cellml-discussion] New draft secondary specification: Uncertainty
Hi Andrew, I don't know for the others, but in my case it would certainly help me understand your document better and, hopefully, provide you with more useful feedback, if you were to give us some concrete examples of the type of models that your specification document targets. I appreciate that your document is 'only' a specification document and this is why I am not suggesting to have those examples as part of the specifications, just as a side document to help the readers/reviewers. Alan -Original Message- From: cellml-discussion-boun...@cellml.org [mailto:cellml-discussion- boun...@cellml.org] On Behalf Of Andrew Miller Sent: 29 May 2012 01:28 To: cellml-discussion@cellml.org Subject: [cellml-discussion] New draft secondary specification: Uncertainty Hi all, I've just put up a draft secondary specification on how we could represent parameter uncertainty within the context of my CellML 1.2 draft. The secondary specification can be viewed here: http://www.cellml.org/Members/miller/draft-secondary-spec-uncertainty/ The DocBook it is generated from can be checked out here: https://github.com/A1kmm/cellml-uncertainty-secondary It is written against this draft CellML 1.2 specification: http://www.cellml.org/Members/miller/draft-normative-spec-andrews- preferred/ It builds upon, and refers to, the draft secondary specification for models of differential-algebraic equations with optional events, available here: http://www.cellml.org/Members/miller/draft-secondary-spec-dae-events Best wishes, Andrew ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion
Re: [cellml-discussion] New draft secondary specification: Uncertainty
Dear all, I'm glad to see uncertainty being taken into acount in cell models. Perhaps it would be worthwhile to add the scenario where no assumptions are made about the uncertainty and a parameter value is simply given by its upper and lower bounds. kind regards, Michael On 05/29/2012 01:28 AM, Andrew Miller wrote: Hi all, I've just put up a draft secondary specification on how we could represent parameter uncertainty within the context of my CellML 1.2 draft. The secondary specification can be viewed here: http://www.cellml.org/Members/miller/draft-secondary-spec-uncertainty/ The DocBook it is generated from can be checked out here: https://github.com/A1kmm/cellml-uncertainty-secondary It is written against this draft CellML 1.2 specification: http://www.cellml.org/Members/miller/draft-normative-spec-andrews-preferred/ It builds upon, and refers to, the draft secondary specification for models of differential-algebraic equations with optional events, available here: http://www.cellml.org/Members/miller/draft-secondary-spec-dae-events Best wishes, Andrew ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion
Re: [cellml-discussion] New draft secondary specification: Uncertainty
On 29/05/12 20:17, Michael Clerx wrote: Dear all, I'm glad to see uncertainty being taken into acount in cell models. Perhaps it would be worthwhile to add the scenario where no assumptions are made about the uncertainty and a parameter value is simply given by its upper and lower bounds. I agree that there needs to be some way to handle that scenario. However, it might be that in most cases where people do that, a uniform distribution with upper and lower bounds will suffice - it seems to me that saying a random variable is distributed uniformly between an upper and lower bound is an explicit statement that the modeller knows that the value falls in certain bounds, but has no reason to believe it is any more likely to fall in one place within those bounds than anywhere else. Such distributions already have a precedent of being used in Bayesian statistics as uninformative prior distributions. Best wishes, Andrew kind regards, Michael On 05/29/2012 01:28 AM, Andrew Miller wrote: Hi all, I've just put up a draft secondary specification on how we could represent parameter uncertainty within the context of my CellML 1.2 draft. The secondary specification can be viewed here: http://www.cellml.org/Members/miller/draft-secondary-spec-uncertainty/ The DocBook it is generated from can be checked out here: https://github.com/A1kmm/cellml-uncertainty-secondary It is written against this draft CellML 1.2 specification: http://www.cellml.org/Members/miller/draft-normative-spec-andrews-preferred/ It builds upon, and refers to, the draft secondary specification for models of differential-algebraic equations with optional events, available here: http://www.cellml.org/Members/miller/draft-secondary-spec-dae-events Best wishes, Andrew ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion
Re: [cellml-discussion] [sbml-distrib] New draft secondary specification: Uncertainty
* Andrew Miller ak.mil...@auckland.ac.nz [2012-05-29 10:52] writes: On 29/05/12 20:17, Michael Clerx wrote: Dear all, I'm glad to see uncertainty being taken into acount in cell models. Perhaps it would be worthwhile to add the scenario where no assumptions are made about the uncertainty and a parameter value is simply given by its upper and lower bounds. I agree that there needs to be some way to handle that scenario. However, it might be that in most cases where people do that, a uniform distribution with upper and lower bounds will suffice - it seems to me that saying a random variable is distributed uniformly between an upper and lower bound is an explicit statement that the modeller knows that the value falls in certain bounds, but has no reason to believe it is any more likely to fall in one place within those bounds than anywhere else. A uniform distribution is a huge assumption about a given range; I believe it would be not at all justified to make such an assumption by default. If the user knows that the value is distributed in a uniform linear fashion, they should be able to say that; likewise a uniform log or a normal or an Epanechnikov kernel. But telling them they have to pick, or, worse, assuming you know when they didn't tell you is dangerous. -Lucian ___ cellml-discussion mailing list cellml-discussion@cellml.org http://lists.cellml.org/mailman/listinfo/cellml-discussion