no_ode_fit

View page source

Do A No Ode Fit For One Node

Prototype

# at_cascade.no_ode_fit
def no_ode_fit(
   all_node_database   ,
   root_database  ,
   option_all_dict     ,
   fit_type            ,
) :
   assert type(all_node_database) == str
   assert type(root_database) == str
   assert type(option_all_dict) == dict
   assert fit_type == 'fixed' or fit_type == 'both'
   # ...
   assert type(root_fit_database) == str
   return root_fit_database

all_node_database

is a python string containing the name of the all_node_db.

root_database

is a python string specifying the location of the root_database.

option_all_dict

is a dict containing the values in the option_all table. This dictionary has a key for each option_name and the corresponding value is option_value. If an option does not appear in the table, the corresponding key does not appear in option_all_dict.

fit_type

is a str equal to both or fixed (fit both fixed and random effect or just fixed effects).

no_ode_database

An intermediate database is stored in the file

result_dir/root_node_name/no_ode/dismod.db

see result_dir and root_node_name. This contains the results of fitting with only the integrand that do not require solving the ODE; i.e. Sincidence, remission, and mtexcess which measure iota, rho, and chi respectively. It also include relrisk which measures

( omega + chi ) / omega

  1. These integrands are included even if they are held out in the root_fit_database using the hold_out_integrand option.

  2. The integrand mtother is excluded because omega is constrained using omega_constraint.

  3. The data likelihoods are fit as Gaussian using the dismod_at data_density command.

  4. The results of fitting this data base can be converted to csv files and plotted using the dismod_at db2csv and plotting routines.

root_fit_database

The return value root_fit_database is equal to

result_dir/root_node_name/dismod.db

which can’t be the same file name as root_database. This is an input_node_database similar to root_database. The difference is that the mean value in the priors for the fixed effects have been replace by the optimal estimate for fitting with out the integrands that use the ODE. The last operation on this database is a dismod_at init command.