Newbie question: linear regression

Fuyuya-san,

Thanks for your pointer. I wasn't trying to derive equations for the
general regression case, but just an expression that I made up quickly.
What I would like to be able to do is get to the point where I can
derive least-squares (and maximum likelihood) estimators for a wide
range of applications.

The application I was thinking of when I made up the equation y = ax +
bw was the linear combination of the output (thinking of this being in a
range between 0 and 1, for a two class problem) of a number of machine
learning algorithms working on a single problem. In this case, there are
weights for each variable (algorithm), but not usually a constant.The
same applies for weights applied to terms extracted from documents.
Usually we're trying to find a winning class by applying (by some
twisted manner) an 'argmax c in Clases', and hence the constant doesn't
add anything.

The paper:

http://www3.oup.co.uk/search97cgi/s97_cgi?action=view&Return_page=http%3A%2F%2Fwww3.oup.co.uk%2Fsearch97cgi%2Fs97_cgi%3Faction%3DFilterSearch%26QueryZip%3Dkoppel%26SourceQueryZip%3D%2B%2528%2Bjournal%253D%2522Literary%2Band%2BLinguistic%2BComputing%2522%2B%2529%2B%26Filter%3Dfilter%252Ehts%26ResultTemplate%3Dresults%252Ehts%26SourceQueryText%3D%2B%2528%2Bjournal%253D%2522Literary%2Band%2BLinguistic%2BComputing%2522%2B%2529%2B%26QueryText%3Dkoppel%26Collection%3Djournal%26SortField%3DScore%26SortOrder%3Ddesc%26ViewTemplate%3Ddocview%252Ehts%26ResultStart%3D1%26ResultCount%3D10%26Collection%3Djournal%26FieldQueryText%3Dkoppel%26FieldName%3DDocumentBody%26Journal%3DLiterary%2Band%2BLinguistic%2BComputing&ViewTemplate=docview.hts&Base_Url=/litlin/hdb/Volume_17/Issue_04/&VdkVgwKey=%2Fvol3%2Fhtdocs%2Fjnls%2Flist%2Flitlin%2Fhdb%2FVolume_17%2FIssue_04%2F170401.sgm.abs.html&QueryZIP=koppel

is an example of the kind of thing I'm thinking of.

Cheers,

Nippon no daigaku de ryuugaku shita Ross-c kara desu.