wish_list

View page source

Wish List for at_cascade

avgint Table

Document the state of the dismod_at.db avgint table at the end of fit_one_job . (Note that the avgint table in ancestor.db and this.db are different; see avgint Table below.)

csv.fit

Child Job Priors

Add an option to create child job priors using the method described in fit_info_theory .

predict_goal.csv

Currently the predictions can only be done for jobs that are fit given the fit_goal.csv . We could extend the predictions to any jog (node, sex pair) that has an ancestor job that was fit.

avgint Table

Document the state of the avgint table in ancestor.db and this.db at the end of csv.pre_one_job . These are the dismod_at databases used by csv.predict to predict using the prior and posterior distributions respectively.

Measurement Value Covariate

Currently csv.fit automatically creates an absolute covariate called one . If we also automatically created a covariate called meas_value , we could use it with a meas_value covariate multiplier to expand or contract measurements values.

Prediction Grid

It would be good to specify a prediction grid that may be different from the covariate age-time grid.

covariate.csv

The rate_eff_cov table is a dismod_at quick fix that attaches covariate functions to locations, instead of covariate values to data points. This may use a lot of memory and take a while to set up; i.e., slow down each run of dismod_at. On the other hand, there is only one copy of this table for each cascade (in the root_database ) so it should not use a lot of disk space. One partial solution would be to not put all the nodes in the covariate.csv file. We could use the closest ancestor’s covariate and omega values when a node is not in covariate.csv.

Duration

Calculate average duration for incident cases given future remission and mortality trends. See the equation for duration as a function of age and time from his DisMod III book section 8.2 in Flaxman et al. (2015). And the definition of duration given with equations in Barendregt et al. (2003).

max_fit

Add an option to ignore the max_fit option when fitting a leaf node; i.e., a node that has no children. This is actually for leaf jobs not leaf nodes in the special case where the splitting covariate is split at a leaf node. On the other hand, perhaps the node tree should be extended so that leaf nodes with lots of data are split into sub-nodes.

ode_step_size

Want to be able to use a bigger ode step size when there is a shock or other rapid change for a particular age or time.