\(\newcommand{\B}[1]{ {\bf #1} }\) \(\newcommand{\R}[1]{ {\rm #1} }\) \(\newcommand{\W}[1]{ \; #1 \; }\)
bnd_mulcov_command#
View page sourceBound The Covariate Multiplier Absolute Data Effect#
Syntax#
dismod_at
database bnd_mulcov
max_abs_effectdismod_at
database bnd_mulcov
max_abs_effect covariate_namePurpose#
This command is used to set the maximum absolute effect in the model for the data values. This is done by changing the lower and upper bounds for the covariate multipliers (ignoring bounds in the corresponding priors). The meas_noise multipliers and Subgroup Covariate Multipliers are not included. The subgroup covariate multipliers are random effects and bound_random set the absolute bound for all the random effects.
database#
Is an
http://www.sqlite.org/sqlite/ database containing the
dismod_at
input tables which are not modified.
max_abs_effect#
is either inf
(for infinity) or
a non-negative value that bounds absolute covariate effects.
A covariate multiplier is defined by each row of the mulcov_table .
We use the notation mul_value for a value of the multiplier.
We use the notation cov_value for a value of the
covariate in the data table.
We use the notation cov_ref for the
reference for the covariate.
The maximum effect condition is
mul_value * ( cov_value-
cov_ref ) | <= max_abs_effect
Note that the limits on the covariate multiplier in its prior have units and the max_abs_effect does not have units.
covariate_name#
If this argument is present, it is a covariate_name . In this case, the inequality above only refers to covariate multipliers that use this covariate.
bnd_mulcov_table#
The table bnd_mulcov_table is an input and output for this command.
max_cov_diff#
The max_cov_diff column is not changed.
max_mulcov#
The max_mulcov column is set so the inequality above is true for all the data that is modeled using this covariate multiplier and that is included in the fit. To be specific, for each covariate multiplier
max_mulcov = max_abs_effect / max_cov_diff
If covariate_name is present, max_mulcov the bound is only changed for multipliers that use that covariate. The max_mulcov value for meas_noise covariates are not changed.
Infinite Case#
The case where max_abs_effect is inf
or
max_cov_diff is zero,
max_mulcov is set to null (which corresponds to plus infinity).
Example#
The file user_bnd_mulcov.py
contains an example and test
using this command.