option_all_table¶

View page source

All Node Option Table¶

This table specifies certain at_cascade options

Table Format¶

option_all_id¶

is the Primary Key for this table.

option_name¶

This column of the option table has type text. It specifies a name that is attached to each option value.

option_value¶

This column of the option table has type text. It specifies a value that is attached to each option name.

absolute_covariates¶

This is a space separated list of the names for the covariates that always have reference value zero. If this option does not appear, there are no absolute covariates. This option is only used when cov_reference_table is None in a call to create_all_node_db. In any event, it is overridden by the final values in cov_reference_table .

balance_fit¶

The subsample of the data with size max_fit always attempts to balance child nodes; i.e, get an equal number of data values for each child of the node currently being fit. If this option appears, max_fit must appear. This option specifies additional balancing that dismod_at should do for the randomly selected data subsample. It is a space separated list with the following values in the following order:

cov_name is the name of the covariate that we are balancing.
value_1 is the lower covariate value that we are balancing.
value_2 is the upper covariate values that we are balancing.

If this option appears, the max_fit option must also appear.

freeze_type¶

This options specifies the type of freeze corresponding to the rows of the mulcov_freeze_table . It is either mean or posterior and its default is mean . Each row of the mulcov_freeze table specifies a freeze job and a freeze covariate multiplier.

mean¶

If the freeze_type is mean , the mean (optimal value) for the freeze covariate multiplier determined by the freeze job will be the lower and upper limit for the covariate multiplier in descendants of the freeze job; see parent_job_id in the job table. Note that if the lower and upper limits are equal, the corresponding model variable is treated as if it has no uncertainty.

posterior¶

If the freeze_type is posterior , the freeze covariate multiplier posterior, determined by the freeze job, will be its prior distribution for all the descendants of the freeze job. This enables one to account for the uncertainty of covariate multiplier values.

max_abs_effect¶

If this option appears, it specifies an extra bound on the absolute value of the covariate multipliers, except for measurement noise multipliers. To be specific, the bound on the covariate multiplier is as large as possible under the condition

max_abs_effect <= | mul_bnd * ( cov_value - cov_ref ) |

where mul_bnd is the non-negative covariate multiplier bound, cov_value is a data table value of the covariate, and cov_ref is the reference value for the covariate. It is an extra bound because it is in addition to the priors for a covariate multiplier.

max_fit¶

This is a text representation of a non-negative integer specifying the maximum number of values to fit for each integrand. If more than this number of values are available, at a fit_node and for one integrand, a randomly selected set of the values are held out so that only this number are included in the fit. If this option does not appear, all of the data for each integrand is included (unless held out in the root_node data table or option table).

mulcov_freeze¶

If the split_reference_id and the node_id for a fit appears in the mulcov_freeze_table, the maximum number of values to fit is doubled. This results in a better representation of the covariate multipliers before freezing them for the jobs that will use the value determined by this fit.

max_fit_parent¶

If this option appears, max_fit must appear. This option specifies the maximum number of parent node data values for each integrand and max_fit only applies to the child node. If this option does not appear, max_fit applies to all the data that is include for a fit. Note that data corresponding to the parent node will not be used when fitting any of its descendants.

max_number_cpu¶

This is the maximum number of cpus (processors) that cascade_root_node or continue_cascade can use. If this is one (more than one) the jobs in the job_table will be run in sequentially (in parallel). If running sequentially, the command output is printed to the screen. Otherwise, it is printed to a file called $code trace.out$ in the output directory corresponding to the job being run. If this option does not appear, the value one is used.

no_ode_ignore¶

The is a space separated list of rate and integrand names that should be ignored when during a no_ode_fit . The priors for the following variables will not be changed by no_ode_fit:

The rate names in no_ode_ignore .
The covariate multiplies that affect the rates in no_ode_ignore.
The covariate multiplies that affect measurement values for the integrands in no_ode_ignore .

number_sample¶

This is the number of independent samples of the posterior distribution for the fitted variables to generate (for each fit). These samples are used by the dismod_at predict command to get predictions for the children of the node being fit. When splitting, the samples are used to predict for the same node at the new split covariate values. If this option does not appear, the value 20 is used.

perturb_optimization_scale¶

This is the standard deviation of the log of a random multiplier. The multiplier is used to randomly shift the optimization scaling point from the prior mean. The scaling point is then projected back to the feasible region. This avoids bad scaling when the prior mean is very close to the solution. If this option does not appear, or if it is zero, the scaling point is equal to the prior mean for all variables.

perturb_optimization_start¶

This is similar to perturb_optimization_scale except that the starting point (instead of the scaling point) is shifted.

sample_method¶

This is the dismod_at sampling method used to create posterior samples of the model variables for each fit. It must be one of the following: asymptotic , censor_asymptotic or simulate . These samples are used to create prior for the fit’s child jobs and for dismod_at predictions; e.g., csv.predict . It is an error for sample_method to be simulate and number_sample to be greater than 20. If this option does not appear, the value asymptotic is used.

shared_memory_prefix¶

This is used at the start of name for shared memory for this cascade. No two cascades can run with the same shared_memory_prefix. If this option does not appear, the empty string is used for the shared_memory_prefix.

refit_split¶

If this option appears, it specifies if there should be a fits, at the same node that a split occurs at, after the split by value of the fitting covariate. The possible values for this option are true and false and its default value is false. If this is true, then the fits after the split will be used to get the priors for the child node. If it is false, the fit before the split will be used and there with be no other fits at the same node. If split_reference_table is empty, this option must be false.

result_dir¶

This option must appear and all of the at_cascade output files are placed in this directory.

root_node_name¶

This option_name must appear and the corresponding option_value is the name of the root_node in the node table of the root_database . (The node table is the same in all the dismod_at databases.)

root_database¶

This option_name must appear and the corresponding option_value

root_split_reference_name¶

This is the split_reference_table.split_reference_name corresponding to the root_database. If split_reference_table is empty (is not empty) this option must not (must) appear. is the name of the root_database .

split_covariate_name¶

is the name, in the root_database covariate table, of the splitting covariate. If split_reference_table is empty (is not empty) this option must not (must) appear.

shift_prior_std_factor¶

This factor multiplies the parent fit posterior standard deviation for the Value Priors in the shifted databases (except for the covariate multipliers). If it is greater (less) than one, the child priors are larger (smaller) than the posterior corresponding to the parent node fit. If this option does not appear, the value one is used for the factor.

shift_prior_std_factor_mulcov¶

This factor multiplies the parent fit posterior standard deviation for the covariate multipliers in the shifted databases. If this option does not appear, shift_prior_std_factor is used for the factor.

shift_prior_dage¶

The possible values for this option are true and false and its default value is true. It this is false, no dage priors are created for the child jobs; see dage and dtime Priors .

shift_prior_dtime¶

The possible values for this option are true and false and its default value is true. It this is false, no dtime priors are created for the child jobs. If both shift_prior_dage and shift_prior_dtime are false, only value priors are created for the child jobs.