HP Forum Archive 19

[ Return to Index | Top of Index ]

OT: Tutorials for R statistical language Message #1 Posted by Namir on 23 Oct 2009, 11:46 a.m. Hi All, During the HHC2009 the subject of the open source statistical language R came up. I was surprised that a few attendees were as gong ho about R as I was. I have posted several tutorials for R on my website (namirshammas[[[dot]]]com). Click on the bottom most link in the main page and you will see the R Master Page with the list of tutorials. All but one link will let you download pdf files. One of the tutorial deals with automatic best linearized regression model selection for two variables and multiple variables. In the tutorial I mention that the concept of best-model search is inspired by vintage programmable HP calculators. If you tinker in R, I recommend that you check that one and of course the other tutorials too. Namir Edited: 23 Oct 2009, 4:11 p.m.

Re: OT: Tutorials for R statistical language Message #2 Posted by Les Bell on 23 Oct 2009, 9:43 p.m., in response to message #1 by Namir You are a gentleman and a scholar, Sir! I'm a moderate R user and have been having some problems with plotting, so I'll be starting straight away with your R 102 Plotting Tutorial. Thanks for the link! Best, --- Les [http://www.lesbell.com.au]

Re: OT: Tutorials for R statistical language Message #3 Posted by Namir on 24 Oct 2009, 12:29 a.m., in response to message #2 by Les Bell You are most welcome Les! Namir

New Approximations for Inverse PDF Message #4 Posted by Namir on 24 Oct 2009, 12:37 a.m., in response to message #1 by Namir I also posted a short PDF file on my web site that has new approximations I was able to obtain for the inverse Normal, Student-t, Chi-square, and F probability distributions. You can find the pdf file when you click here to get to the New Algorithms page and then click on the related link. The approximations for the inverse probability distributions are shorter than the legacy ones you find in the Handbook of Mathematical Functions and can easily be programmed on your calculators. Namir

Re: New Approximations for Inverse PDF Message #5 Posted by Paul Dale on 24 Oct 2009, 1:04 a.m., in response to message #4 by Namir Thanks for this. Very interesting. No error estimate though :-( I only recently did the 20b scientific probability distributions and the inverses were a major pain. I ended up using the solver code and the cumulative distributions for all the hard cases. Since these functions are relatively well behaved, convergence seems to be fairly quick but I might look into your estimates as better starters... - Pauli

Re: New Approximations for Inverse PDF Message #6 Posted by Namir on 24 Oct 2009, 8:55 a.m., in response to message #5 by Paul Dale I remember I was demanding on these equations. I would be very curious as to what your test comes up with. Cheers, Namir

Re: New Approximations for Inverse PDF Message #7 Posted by Tim Wessman on 24 Oct 2009, 11:20 p.m., in response to message #6 by Namir You don't happen to have any good, simple approximations for binomial and Poisson cumulative do you? I am aware of the incomplete beta method, which is actually the method used to calculate most of the stat distros and inverses in the 20b, but nothing that is really simple and quick for general usage. I've seen several that do a good job over specific ranges, but I haven't seen one the works decently over the whole range. TW

Re: New Approximations for Inverse PDF Message #8 Posted by Paul Dale on 24 Oct 2009, 11:54 p.m., in response to message #7 by Tim Wessman For Binomial and Poisson you could also sum the individual probabilities directly. However, I'd go with the regularised incomplete beta and gamma functions for most of the common probability distributions. There are relatively short and quickly converging iterative methods for both. The inverses aren't so nice :-( - Pauli

Re: New Approximations for Inverse PDF Message #9 Posted by Tim Wessman on 29 Oct 2009, 1:37 p.m., in response to message #8 by Paul Dale Yeah, summing directly works great until you start getting large Ns. Of course you can just use an approximated normal at that point. . . TW

[ Return to Index | Top of Index ]

Go back to the main exhibit hall