Case Reader

The CaseReader object is provided to read case recordings, regardless of which case recorder was used.

Currently, OpenMDAO only implements SqliteCaseRecorder, therefore all the examples will make use of this recorder. Other types of case recorders are expected to be supported in the future.

CaseReader Constructor

The call signature for the CaseReader constructor is:

SqliteCaseReader.__init__(filename, pre_load=False, metadata_filename=None)[source]

Initialize.

Parameters
filenamestr

The path to the filename containing the recorded data.

pre_loadbool

If True, load all the data into memory during initialization.

metadata_filenamestr

The path to the filename containing the recorded metadata, if separate.

Determining What Sources and Variables Were Recorded

The CaseReader object provides methods to determine which objects in the original problem were sources for for the recorded cases and what variables they recorded. Sources can include the problem, driver, components and solvers.

The list_sources method provides a list of the names of objects that are the sources of recorded data in the file.

BaseCaseReader.list_sources(out_stream=DEFAULT_OUT_STREAM)[source]

List of all the different recording sources for which there is recorded data.

Parameters
out_streamfile-like object

Where to send human readable output. Default is sys.stdout. Set to None to suppress.

Returns
list
One or more of: problem, driver, <system hierarchy location>,

<solver hierarchy location>

The complementary list_source_vars method will provide a list of the input and output variables recorded for a given source.

BaseCaseReader.list_source_vars(source, out_stream=DEFAULT_OUT_STREAM)[source]

List of all inputs and outputs recorded by the specified source.

Parameters
source{‘problem’, ‘driver’, <system hierarchy location>, <solver hierarchy location>}

Identifies the source for which to return information.

out_streamfile-like object

Where to send human readable output. Default is sys.stdout. Set to None to suppress.

Returns
dict

{‘inputs’:[key list], ‘outputs’:[key list], ‘residuals’:[key list]}. No recurse.

Here is an example of their usage:

import openmdao.api as om
from openmdao.test_suite.components.sellar_feature import SellarMDA

import numpy as np

# define Sellar MDA problem
prob = om.Problem(model=SellarMDA())

model = prob.model
model.add_design_var('z', lower=np.array([-10.0, 0.0]),
                          upper=np.array([10.0, 10.0]))
model.add_design_var('x', lower=0.0, upper=10.0)
model.add_objective('obj')
model.add_constraint('con1', upper=0.0)
model.add_constraint('con2', upper=0.0)

prob.driver = om.ScipyOptimizeDriver(optimizer='SLSQP', tol=1e-9, disp=False)

# add recorder to the driver, model and solver
recorder = om.SqliteRecorder('cases.sql')

prob.driver.add_recorder(recorder)
model.add_recorder(recorder)
model.nonlinear_solver.add_recorder(recorder)

# run the problem
prob.setup()
prob.set_solver_print(0)
prob.run_driver()
prob.cleanup()

cr = om.CaseReader('cases.sql')
sources = cr.list_sources()
Sources
driver
root.nonlinear_solver
root
driver_vars = cr.list_source_vars('driver')
inputs outputs residuals
con1
con2
obj
x
z
model_vars = cr.list_source_vars('root')
inputs outputs residuals
y1 z z
y2 x x
x con1 con1
z con2 con2
y1 y1
y2 y2
obj obj
solver_vars = cr.list_source_vars('root.nonlinear_solver')
inputs outputs residuals
z z
x x
y2 y1
y1 y2
obj
con1
con2

Case Names

The CaseReader provides access to Case objects, each of which encapsulates a data point recorded by one of the sources.

Case objects are uniquely identified in a case recorder file by their case names. A case name is a string. As an example, here is a case name:

'rank0:ScipyOptimize_SLSQP|1|root._solve_nonlinear|1'

The first part of the case name indicates which rank or process that the case was recorded from. The remainder of the case name shows the hierarchical path to the object that was recorded along with the iteration counts for each object along the path. It follows a pattern of repeated pairs of

- object name ( problem, driver, system, or solver )
- iteration count

These are separated by the | character.

So in the given example, the case is:

- from rank 0
- the first iteration of the driver, `ScipyOptimize_SLSQP`
- the first execution of the `root` system which is the top-level model

Getting Names of the Cases

The list_cases method returns the names of the cases in the order in which the cases were executed. You can optionally request cases only from a specific source.

BaseCaseReader.list_cases(source=None, recurse=True, flat=True, out_stream=DEFAULT_OUT_STREAM)[source]

Iterate over Driver, Solver and System cases in order.

Parameters
source{‘problem’, ‘driver’, <system hierarchy location>, <solver hierarchy location>,

case name} If not None, only cases originating from the specified source or case are returned.

recursebool, optional

If True, will enable iterating over all successors in case hierarchy.

flatbool, optional

If False and there are child cases, then a nested ordered dictionary is returned rather than an iterator.

out_streamfile-like object

Where to send human readable output. Default is sys.stdout. Set to None to suppress.

Returns
iterator or dict

An iterator or a nested dictionary of identified cases.

There are two optional arguments to the list_cases method that affect what is returned.

- recurse: causes the returned value to include child cases.

- flat: works in conjunction with the `recurse` argument to determine if the returned
  results are in the form of a list or nested dict. If recurse=True, flat=False, and there
  are child cases, then the returned value is a nested ordered dict. Otherwise, it is a list.

Accessing Cases

Getting information from the cases is a two-step process. First, you need to get access to the Case object and then you can call a variety of methods on the Case object to get values from it. The second step is described on the Getting Data from a Case page.

There are two methods used to get a specific Case:

- get_cases
- get_case

Accessing Cases Using get_cases Method

The get_cases method provides a quick and easy way to iterate over all the cases.

BaseCaseReader.get_cases(source, recurse=True, flat=False)[source]

Iterate over the cases.

Parameters
source{‘problem’, ‘driver’, <system hierarchy location>, <solver hierarchy location>,

case name} Identifies which cases to return.

recursebool, optional

If True, will enable iterating over all successors in case hierarchy

flatbool, optional

If False and there are child cases, then a nested ordered dictionary is returned rather than an iterator.

Returns
list or dict

The cases identified by source

This method is similar to the list_cases method in that it has the two optional arguments recurse and flat to control what is returned as described above.

Here is an example of its usage:

import openmdao.api as om
from openmdao.test_suite.components.sellar_feature import SellarMDA

import numpy as np

prob = om.Problem(model=SellarMDA())

model = prob.model
model.add_design_var('z', lower=np.array([-10.0, 0.0]),
                          upper=np.array([10.0, 10.0]))
model.add_design_var('x', lower=0.0, upper=10.0)
model.add_objective('obj')
model.add_constraint('con1', upper=0.0)
model.add_constraint('con2', upper=0.0)

driver = prob.driver = om.ScipyOptimizeDriver(optimizer='SLSQP', tol=1e-5)
driver.add_recorder(om.SqliteRecorder('cases.sql'))

prob.setup()
prob.set_solver_print(0)
prob.run_driver()
prob.cleanup()
Optimization terminated successfully.    (Exit mode 0)
            Current function value: 3.18339553462378
            Iterations: 8
            Function evaluations: 9
            Gradient evaluations: 8
Optimization Complete
-----------------------------------
/usr/share/miniconda/envs/test/lib/python3.8/site-packages/openmdao/recorders/sqlite_recorder.py:224: UserWarning:The existing case recorder file, cases.sql, is being overwritten.
cr = om.CaseReader('cases.sql')
cases = cr.get_cases()
for case in cases:
    print(case.name, sorted(case.outputs))
rank0:ScipyOptimize_SLSQP|0 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|1 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|2 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|3 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|4 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|5 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|6 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|7 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|8 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|9 ['con1', 'con2', 'obj', 'x', 'z']

Accessing Cases Using get_case Method

The get_case method returns a Case object given a case name.

.. automethod:: openmdao.recorders.base_case_reader.BaseCaseReader.get_case
    :noindex:

You can use the get_case method to get a specific case from the list of case names returned by list_cases as shown here:

    case_names = cr.list_cases()
driver
rank0:ScipyOptimize_SLSQP|0
rank0:ScipyOptimize_SLSQP|1
rank0:ScipyOptimize_SLSQP|2
rank0:ScipyOptimize_SLSQP|3
rank0:ScipyOptimize_SLSQP|4
rank0:ScipyOptimize_SLSQP|5
rank0:ScipyOptimize_SLSQP|6
rank0:ScipyOptimize_SLSQP|7
rank0:ScipyOptimize_SLSQP|8
rank0:ScipyOptimize_SLSQP|9
# access a Case by name (e.g. first case)
case = cr.get_case("rank0:ScipyOptimize_SLSQP|0")
print(case.name, sorted(case.outputs))
rank0:ScipyOptimize_SLSQP|0 ['con1', 'con2', 'obj', 'x', 'z']
# access a Case by index (e.g. first case)
case = cr.get_case(0)
print(case.name, sorted(case.outputs))
rank0:ScipyOptimize_SLSQP|0 ['con1', 'con2', 'obj', 'x', 'z']
# access a Case by index (e.g. last case)
case = cr.get_case(-1)
print(case.name, sorted(case.outputs))
rank0:ScipyOptimize_SLSQP|9 ['con1', 'con2', 'obj', 'x', 'z']
# get each case by looping over case names
for name in case_names:
    case = cr.get_case(name)
    print(case.name, sorted(case.outputs))
rank0:ScipyOptimize_SLSQP|0 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|1 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|2 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|3 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|4 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|5 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|6 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|7 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|8 ['con1', 'con2', 'obj', 'x', 'z']
rank0:ScipyOptimize_SLSQP|9 ['con1', 'con2', 'obj', 'x', 'z']

Processing a Nested Dictionary of Its Child Cases

The following example demonstrates selecting a case from a case list and processing a nested dictionary of its child cases.

import openmdao.api as om
from openmdao.test_suite.components.sellar_feature import SellarMDA

import numpy as np

# define Sellar MDA problem
prob = om.Problem(model=SellarMDA())

model = prob.model
model.add_design_var('z', lower=np.array([-10.0, 0.0]),
                          upper=np.array([10.0, 10.0]))
model.add_design_var('x', lower=0.0, upper=10.0)
model.add_objective('obj')
model.add_constraint('con1', upper=0.0)
model.add_constraint('con2', upper=0.0)

prob.driver = om.ScipyOptimizeDriver(optimizer='SLSQP', tol=1e-5)

# add recorder to the driver, model and solver
recorder = om.SqliteRecorder('cases.sql')

prob.driver.add_recorder(recorder)
model.add_recorder(recorder)
model.nonlinear_solver.add_recorder(recorder)

# run the problem
prob.setup()
prob.set_solver_print(0)
prob.run_driver()
prob.cleanup()
Optimization terminated successfully.    (Exit mode 0)
            Current function value: 3.18339553462378
            Iterations: 8
            Function evaluations: 9
            Gradient evaluations: 8
Optimization Complete
-----------------------------------
cr = om.CaseReader('cases.sql')

# get the last driver case
driver_cases = cr.list_cases('driver')
solver
rank0:ScipyOptimize_SLSQP|0|root._solve_nonlinear|0|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|0|root._solve_nonlinear|0
driver
rank0:ScipyOptimize_SLSQP|0
solver
rank0:ScipyOptimize_SLSQP|1|root._solve_nonlinear|1|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|1|root._solve_nonlinear|1
driver
rank0:ScipyOptimize_SLSQP|1
solver
rank0:ScipyOptimize_SLSQP|2|root._solve_nonlinear|2|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|2|root._solve_nonlinear|2
driver
rank0:ScipyOptimize_SLSQP|2
solver
rank0:ScipyOptimize_SLSQP|3|root._solve_nonlinear|3|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|3|root._solve_nonlinear|3
driver
rank0:ScipyOptimize_SLSQP|3
solver
rank0:ScipyOptimize_SLSQP|4|root._solve_nonlinear|4|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|4|root._solve_nonlinear|4
driver
rank0:ScipyOptimize_SLSQP|4
solver
rank0:ScipyOptimize_SLSQP|5|root._solve_nonlinear|5|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|5|root._solve_nonlinear|5
driver
rank0:ScipyOptimize_SLSQP|5
solver
rank0:ScipyOptimize_SLSQP|6|root._solve_nonlinear|6|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|6|root._solve_nonlinear|6
driver
rank0:ScipyOptimize_SLSQP|6
solver
rank0:ScipyOptimize_SLSQP|7|root._solve_nonlinear|7|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|7|root._solve_nonlinear|7
driver
rank0:ScipyOptimize_SLSQP|7
solver
rank0:ScipyOptimize_SLSQP|8|root._solve_nonlinear|8|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|8|root._solve_nonlinear|8
driver
rank0:ScipyOptimize_SLSQP|8
solver
rank0:ScipyOptimize_SLSQP|9|root._solve_nonlinear|9|NLRunOnce|0
system
rank0:ScipyOptimize_SLSQP|9|root._solve_nonlinear|9
driver
rank0:ScipyOptimize_SLSQP|9
# get a recursive dict of child cases of the last driver case
last_driver_case = driver_cases[-1]
cases = cr.get_cases(last_driver_case, recurse=True, flat=False)

# display selected information from nested dict of cases
def print_cases(cases, indent=0):
    for case, children in cases.items():
        print(indent*' ', case.source, '-', case.name.split('.')[-1], sorted(case.outputs))
        if children:
            print_cases(children, indent+2)
            
print_cases(cases)
 driver - rank0:ScipyOptimize_SLSQP|9 ['con1', 'con2', 'obj', 'x', 'z']
   root - _solve_nonlinear|9 ['con1', 'con2', 'obj', 'x', 'y1', 'y2', 'z']
     root.nonlinear_solver - _solve_nonlinear|9|NLRunOnce|0 ['con1', 'con2', 'obj', 'x', 'y1', 'y2', 'z']