\(\newcommand{\B}[1]{ {\bf #1} }\) \(\newcommand{\R}[1]{ {\rm #1} }\) \(\newcommand{\W}[1]{ \; #1 \; }\)
fit_var_table#
View page sourceThe Optimization Results for Variables#
Discussion#
The fit_var table contains the maximum likelihood estimate
for the model_variables corresponding to
the data table meas_value .
A new fit_var
table is created each time the
fit_command is executed.
Lagrange Multipliers#
Setting good upper and lower limits, not zero or infinite,
helps dismod_at
determine the scale for the constraints
and gives better detection of which constraints are active.
fit_var_id#
This column has type integer
and is the primary key for this table.
Its initial value is zero, and it increments by one for each row.
The fit_var_id column is also a foreign key for the
var_table ; i.e.,
var_id = fit_var_id
In addition, the size of both tables is the same.
fit_var_value#
This column has type real
and contains the
final model variables determined by the fit.
This is an approximations for the
fixed effects \(( \theta )\)
that maximize the Laplace approximate objective \(L( \theta)\),
and the random effects that maximum the joint likelihood
\(\hat{u} ( \theta )\); see the
cppad_mixed
documentation for more details.
residual_value#
This column has type real
and contains the
weighted residual
corresponding to the value_prior_id
for this variable.
If there is no such residual, this column is null
.
For example, if the corresponding density is
uniform.
residual_dage#
This column has type real
and contains the
weighted residual
corresponding to the dage_prior_id
for this variable.
If there is no such residual, this column is null
.
For example, if the corresponding dage_prior_id is null
.
residual_dtime#
This column has type real
and contains the
weighted residual
corresponding to the dtime_prior_id
for this variable.
If there is no such residual, this column is null
.
For example, if the corresponding dtime_prior_id is null
.
lagrange_value#
This column has type real
and contains the Lagrange multipliers for
the lower and upper limits corresponding the
value_prior_id for this variable.
If it is positive (negative) the upper (lower) limit is active.
If neither prior limit is active, this column is zero.
The Lagrange multipliers are in the scaled space
where the optimization takes place.
lagrange_dage#
This column has type real
and contains the Lagrange multipliers for
the lower and upper limits corresponding the
dage_prior_id for this variable.
If it is positive (negative) the upper (lower) limit is active.
If neither prior limit is active, this column is zero.
The Lagrange multipliers are in the scaled space
where the optimization takes place.
lagrange_dtime#
This column has type real
and contains the Lagrange multipliers for
the lower and upper limits corresponding the
dtime_prior_id for this variable.
If it is positive (negative) the upper (lower) limit is active.
If neither prior limit is active, this column is zero.
The Lagrange multipliers are in the scaled space
where the optimization takes place.
Example#
See the fit_command.py example and test.