Differences

This shows you the differences between two versions of the page.

--- readme.reml [2012/05/28 14:11] – created shogo
+++ readme.reml [2014/11/25 11:16] – shogo
@@ Line 1: / Line 1: @@
-REMLF90 - Additions to PDF documentation
+====== REMLF90 ======
-IM 02/17/2003
-Two new random types were added. Type:
+===== Summary =====
-user_file
+Variance component estimation program with EM-REML algorithm. The final results will be saved in "REMLF90.log".
-accepts a matrix from a file specified under keyword FILE in the following
+\\
-format:
+See PREGSF90 with genotypes (SNP) for options.
-row col value
-row col value
-.....
-Values can be lower- or upper-stored but not both.
+===== Options =====
+<file>
+OPTION conv_crit1d-12
+</file>
+Convergence criterion (default 1d-10).
+<file>
+OPTION maxrounds 10000
+</file>
+Maximum rounds (default 5000).
+<file>
+OPTION sol se
+</file>
+Store solutions and se.
+<file>
+OPTION residual
+</file>
+y-hat and residuals will be included in "yhat_residual".
+<file>
+OPTION missing -999
+</file>
+Specify missing observations (default 0).
+<file>
+OPTION use_yams
+</file>
+Run the program with YAMS (modified FSPAK). The computing time can be dramatically improved.
+<file>
+OPTION constant_var 5 1 2
+</file>
+: effect number\\
+: first trait number\\
+: second trait number\\
+implying the covariance between traits 1 and 2 for effect 5.
+<file>
+OPTION SNP_file snp
+</file>
+Specify the SNP file name to use genotype data.
-Second type:
+===== Does remlf90 always converge? =====
-user_file_inv
+When the expected variance is very small or the covariance matrix is close to non-positive definite, try with two different starting values (e.g., much smaller = 0.00001 and much bigger = 1000 when the expected variance is between 0 and 10).
-works as befoer except that the given matrix is inverted before being
-used in mixed model equations.
+The remlf90 program will always estimate a positive variance (within the parameter space) even if there is no variance (using wrong data or format and/or a completely wrong model).
+If airemlf90 does not converge but remlf90 converges with the same data set and the same model, rerun remlf90 with a small starting value to check the estimate because it could be artifact.