Carl Boettiger

Algo­rithms group today took a brake from dis­cussing algo­rithms for an aside on cod­ing prac­tices.  Today we focused pri­mar­ily on lit­er­ate pro­gram­ming tools for mak­ing bet­ter doc­u­mented, more read­able code.   We cov­ered: Sweave Google style guide, also see R core-team devel­oper Hadley Wickam’s style guide (mod­i­fied from Google) R cod­ing prac­tices grab-bag: using func­tion­al­ized …

Alisa reviews HMM, EM, and shows us how to com­bine them. The EM ver­sion treats the tran­si­tion prob­a­bil­i­ties (exchang­ing coins) and emis­sion prob­a­bil­i­ties (chance of heads/tails under each coin) as unknowns.  Make a guess, cal­cu­late the blue line from before.  This can be used to cal­cu­late a new guess that will hill climb to the locally …

Great group dis­cus­sion on Hid­den Markov Mod­els tag-teamed by Yaniv and Alisa.  Walked through the coin flip exam­ple, cre­at­ing the for­ward algo­rithm, and the pos­te­rior decod­ing.  In this case it is assumed model and para­me­ters are known: two coins, known ratios, known rate of swap­ping; goal is to guess which coin is being used in …

Fin­ished up the EM algo­rithm today.  Key was get­ting the right func­tion to max­i­mize.  Turns out wikipedia has a very nice write up of this very exam­ple, but in our nota­tion: \[  E_p \log(p f_1) + (1-E_p) \log( (1-p) f_2 )\] Where \(E_p\) comes from the expec­ta­tion step. Jamie has joined us and has set up a …