tag:blogger.com,1999:blog-6894866515532737257.post989232941345561576..comments2024-04-22T21:33:32.590-07:00Comments on Probably Overthinking It: There is still only one testAllen Downeyhttp://www.blogger.com/profile/01633071333405221858noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-6894866515532737257.post-75181598575358914112017-12-12T07:36:37.453-08:002017-12-12T07:36:37.453-08:00Yes, me too! Thanks for the link.Yes, me too! Thanks for the link.Allen Downeyhttps://www.blogger.com/profile/01633071333405221858noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-47616136231365182042017-01-18T10:46:50.416-08:002017-01-18T10:46:50.416-08:00There was disagreement among statisticians about h...There was disagreement among statisticians about how to do hypothesis testing; what we ended up with is actually a weird hybrid that makes less sense than the alternatives. See https://en.wikipedia.org/wiki/Statistical_hypothesis_testing#Origins_and_early_controversy<br /><br />Regarding precision and recall, those are the terms most often used in the context of information retrieval. In the context of stats, it's usually false positives and false negatives (or type I and type II errors). And in the context of machine learning, it's usually a confusion matrix. But they are all representations of the same information.Allen Downeyhttps://www.blogger.com/profile/01633071333405221858noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-45725377773039094592017-01-17T15:02:17.176-08:002017-01-17T15:02:17.176-08:00I think there was some other place where I found a...I think there was some other place where I found an explanation that there are two approaches to hypothesis validation. One used p-values. The other one was something like precision and recall (false positives and negatives) or something similar. I think it mentioned some disagreement between statisticians before p-values became popular. May this be possible?trylkshttps://www.blogger.com/profile/14821108955806430210noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-79677701644913361342016-07-14T10:23:03.802-07:002016-07-14T10:23:03.802-07:00In a somewhat similar vein, I'm quite fond of ...In a somewhat similar vein, I'm quite fond of McElreath's recent Statistical Rethinking book and lecture videos:<br />http://xcelab.net/rm/statistical-rethinking/Anonymoushttps://www.blogger.com/profile/08220616256423969075noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-2056969920698747272016-06-14T09:02:49.510-07:002016-06-14T09:02:49.510-07:00Yes! We take a similar approach here in a class c...Yes! We take a similar approach here in a class called Modeling and Simulation, where we use MATLAB's ode45 function to solve differential equations. That gives students the ability to develop models of physical systems that include realistic factors, like friction and air resistance, that are out of scope in a class that is limited to analytic methods.<br /><br />And I am very interested in the other idea you raise, how to do engineering design on systems that don't lend themselves to mathematical analysis. Do we have to search enormous design spaces? Or can we grow/evolve solutions?Allen Downeyhttps://www.blogger.com/profile/01633071333405221858noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-20879997571517255982016-06-08T00:52:11.800-07:002016-06-08T00:52:11.800-07:00This is amazing stuff. Similar conclusions have be...This is amazing stuff. Similar conclusions have been made about different equations: you just simulate them instead of trying to solve them. And in mechanical engineering, you learn all kinds of methods, but then in the real world, you just learn how to use numerical methods (fundamentally based on D.E.s). Even in the idealized world of electrical engineering, there is finally a push in recent years to go "full spice" (having numerical methods more readily available). Although in mechanical and electrical, generalized concepts in the brain is required to zero in on designs that are most likely to be the most efficient, then switching over to the numerical methods to work out the details. I wonder if there is a similar process needed here.<br /><br />It makes me wonder if physics and math are just compression methods suited for the brain. I mean they actually a priori *are*, but I wonder if they may not have any deeper significance than simply being something like patterns that are "accidentally" more common as a result of deeper, simpler logic rules with less K-complexity like cellular automata (Wolfram). And that the most efficient designs would be discoverable with quantum computing, in some way related to how it can break cryptography.Zawyhttps://www.blogger.com/profile/03727573717462028189noreply@blogger.com