## S3 method for class 'ProbeLevelModel':
fit(this, units="remaining", ..., force=FALSE, ram=NULL, verbose=FALSE)NULL, all units are considered.
If remaining, only non-fitted units are considered.readUnits().TRUE, already fitted units are re-fitted, and
cached data is re-read.double indicating if more or less units should
be loaded into memory at the same time.Verbose.The non-array specific parameter estimates together with standard deviation estimates and convergence information are stored in one file.
The parameter estimates specific to each array, typically "chip effects", are stored in array specific files.
Data set specific estimates [L = number of probes]: phi [L doubles] (probe affinities), sd(phi) [L doubles], isOutlier(phi) [L logicals]
Algorithm-specific results: iter [1 integer], convergence1 [1 logical], convergence2 [1 logical] dTheta [1 double] sd(eps) - [1 double] estimated standard deviation of the error term
Array-specific estimates [K = nbr of arrays]: theta [K doubles] (chip effects), sd(theta) [K doubles], isOutlier(theta) [K logicals]
For each array and each unit group, we store: 1 theta, 1 sd(theta), 1 isOutlier(theta), i.e. (float, float, bit) => For each array and each unit (with $G_j$ groups), we store: $G_j$ theta, $G_j$ sd(theta), $G_j$ isOutlier(theta), i.e. $G_j$*(float, float, bit). For optimal access we store all thetas first, then all sd(theta) and the all isOutlier(theta). To keep track of the number of groups in each unit, we have to have a (unit, ngroups) map. This can be obtained from getUnitNames() for the AffymetrixCdfFile class.
ProbeLevelModel.