tag:blogger.com,1999:blog-5105311754915271701.post3323367975923547200..comments2010-01-23T21:49:29.859-05:00Comments on Deep Thoughts: MemoizeWalthttp://www.blogger.com/profile/05235362518037728416noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-5105311754915271701.post-81921223390805129802009-08-28T23:14:11.797-04:002009-08-28T23:14:11.797-04:00One of the cool things about techniques that track...One of the cool things about techniques that track the range of a number, or the error bound, is that you can add the error due to the calculation itself.<br /><br />Even if the inputs are known exactly, if the function is an approximation to the true result, and knows its error bounds, that can be reflected in the range of the returned result.<br /><br />It seems pretty slick to handle errors in input value and errors due to approximations all at once. Some techniques even can let you know how much of the error in the output value is due to each error source.Walthttps://www.blogger.com/profile/05235362518037728416noreply@blogger.comtag:blogger.com,1999:blog-5105311754915271701.post-56496355151599093482009-08-26T22:27:29.203-04:002009-08-26T22:27:29.203-04:00With disk space no longer being a premium as it wa...With disk space no longer being a premium as it was in the past, memoization makes perfect sense; processing time is accelerated the next time a value is needed and the trade-off in disk space seems very justifiable!<br /><br />Numbers storing their own error bounds is interesting as statistical results that have error bounds are computed values by definition, right? <br /><br />Statistical results, if saved such as in memoization, should be allowed associated metadata such as confidence intervals. If any of the initial parameters upon which the statistical results is based were changed, one would want both the results and their metadata to be updated.<br /><br />I usually see applications where tests are run and the results and CIs not saved and updated in the same database; rather, researchers save the routines and re-calculate results in their usual workflow, saving results only in output files rather than within the database as new calculated fields. I like the discussion as I see the potential application to health science databases.Catherine Schenck-Yglesias, MHShttps://www.blogger.com/profile/13571296523684083951noreply@blogger.com