This is an old revision of the document!
Table of Contents
predictf90
Original description
This is program is used to do cross-validations. It reads a blupf90 parameter file, a solutions file
and a data file.
It needs OPTION include_effects
followed by a series of effects.
It computes
- y_hat = sum of estimates of the included effects
- y_star = y corrected by the other (not included) effects
- residual = y - included effects (not a true residual)
More explanation
This program splits the phenotype (y) into 3 pieces.
y = (effects to be corrected) + (effects to be kept) + residual
In a cross validation, you need a quantity “EBV + residual”, where EBV is (effects to be kept)
shown above, and it is y_star
.
The program calculates the following quantities.
- y_hat:
(effects to kept)
- y_star:
y - (effects to be corrected)
- residual:
residual
The user should specify the effects to be kept using OPTION include_effects
.
Example
For instance consider
y = herd + age + animal + e
with OPTION include_effects 3
.
- y_hat = animal_hat
- ystar = y - herd_hat - age_hat
Which makes cor(y_hat,y_star) = cor(ebv, corrected y) which is a measure of accuracy.
It outputs the correlation between y_hat and y_star, for instance cor(ystar,yhat)=cor(u+e, uhat) and outputs these columns into a file, together with animal id (if there is animal in the model) or record number (if not).
In addition, if animal effect is in the model, it produces a file with ebvs from the solutions file.
Output files
yhat_residual
The main file is yhat_residual
, which has corrected phenotypes and predicted residuals.
The number of columns in this file depend on the number of traits (N).
- Column 1: Animal ID (renumbered i.e., same as the 1st column in renaddxx.ped)
- Column 2 to N+1: “y_star” explained above
- Column N+2 to 2N+1: “y_hat” explained above
- Column 2N+2 to 3N+1: “residual” explained above