fit_one_job¶

Run One Job¶

Prototype¶

# at_cascade.fit_one_job
def fit_one_job(
    job_table               ,
    run_job_id              ,
    all_node_database       ,
    node_table              ,
    fit_integrand           ,
    fit_type                ,
    first_fit               ,
    trace_file_obj   = None ,
) :
    assert type(job_table) == list
    assert type(run_job_id) == int
    assert type(all_node_database) == str
    assert type(node_table) == list
    assert type(fit_integrand) == set
    assert fit_type in [ 'both', 'fixed' ]
    assert type(first_fit) == bool
    if trace_file_obj is not None :
        assert isinstance(trace_file_obj, io.TextIOBase)

Default Value¶

The only argument that can be None is trace_file_obj.

job_table¶

This is a job_table containing the jobs necessary to fit the fit_goal_set.

run_job_id¶

This is the job_id for the job that is run. If run_job_id is zero, this job has no ancestors. This is a special case for fit_var

all_node_database¶

is a python string specifying the location of the all_node_db relative to the current working directory.

node_table¶

is a list of dict containing the node table for this cascade.

fit_integrand¶

is a set of integrand_id values that occur in the data table; see get_fit_integrand.

fit_type¶

is a str equal to ‘both’ or ‘fixed’ and specifies the type of fit that dismod_at will do.

first_fit¶

If first_fit is True, this is assumed to be an input_node_database. Otherwise, it is assumed that this routine has previously been called with first_fit equal to True.

trace_file_obj¶

If this argument is not None, it is a io.TextIOBase object corresponding to a file that is opened for writing the tracing output for this job.

fit_database¶

The fit_database for this fit is fit_node_dir/dismod.db where fit_node_dir is the database_dir returned by get_database_dir for the fit node and split_reference_id corresponding to run_job_id.

Upon Input¶

On input, fit_database is an input_node_database.

fit_var¶

Upon return, the fit_var table correspond to the posterior mean for the model variables for the fit_node. If run_job_id is not zero and there is no data corresponding to this fit, the fit is not done because an ancestor job can be used to predict for this job. In this case a no data abort message will appear in the log table.

sample¶

Upon return, the sample table contains the corresponding samples from the posterior distribution for the model variables for the fit_node.

log¶

The log table is initialized as empty when fit_one_job starts. Upon return or abort due to an exception, the log table contains a summary of the operations preformed by dismod. In addition, the following entries will be added to the log table if the corresponding event occurs:

:header-rows:1¶
message_type	message	event
at_cascade	no data: abort	abort fit because all the data is held out
at_cascade	fit: OK	the maximum likelihood problem was solved
at_cascade	sample: OK	the posterior samples were computed
at_cascade	children: OK	the child databases with priors were created

Note that the events depend on each other in the following way:

If children: OK is present, then sample: OK is present.
If sample: OK is present, then fit: OK is present.
If fit: OK is present, then no data: abort is not present.

Exception¶

If there is no data from this fit, this routine will raise an exception with a message that starts with: no data: abort ; i.e., the same as the message it puts in the log.