Steady state sensitivity modes #1727

plakrisenko · 2022-03-16T12:42:47Z

Settings for specifying sensitivities computation method at steady state have been implemented/reworked.
Three SteadyStateSensitivityMode are available for both SensitivityMethods (forward and adjoint):

newtonOnly: only Newton's method is used for sensitivities computation
integrationOnly: only integration is used for sensitivities computation, which in forward sensitivity analysis case means that numerical integration has to be used to find the steady state
integrateIfNewtonFails: more flexible option, where simulation is used if Newton's method fails

Note: Previously available SteadyStateSensitivityMode simulationFSA has been removed.

codecov · 2022-03-21T16:36:24Z

Codecov Report

Merging #1727 (8a45940) into develop (742c464) will increase coverage by 0.06%.
The diff coverage is 100.00%.

@@             Coverage Diff             @@
##           develop    #1727      +/-   ##
===========================================
+ Coverage    77.82%   77.89%   +0.06%     
===========================================
  Files           74       74              
  Lines        12018    12033      +15     
===========================================
+ Hits          9353     9373      +20     
+ Misses        2665     2660       -5

Flag	Coverage Δ
cpp	`75.05% <100.00%> (+0.11%)`	⬆️
petab	`59.18% <ø> (ø)`
python	`68.24% <ø> (ø)`

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/steadystateproblem.cpp	`85.57% <100.00%> (+1.03%)`	⬆️
src/solver_cvodes.cpp	`70.43% <0.00%> (+0.16%)`	⬆️
src/exception.cpp	`81.08% <0.00%> (+5.40%)`	⬆️

tests/petab_test_suite/test_petab_suite.py

src/steadystateproblem.cpp

Co-authored-by: Fabian Fröhlich <fabian_froehlich@hms.harvard.edu>

…to stst_problem_settings

FFroehlich · 2022-03-22T14:47:10Z

I would guess that the expected preeq_status and posteq_status fields in some of the python test will have to be updated.

plakrisenko · 2022-03-23T15:52:15Z

I would guess that the expected preeq_status and posteq_status fields in some of the python test will have to be updated.

I should have set different setSteadyStateSensitivityMode in that test

FFroehlich

great work 👍 , not sure why the last remaining test fails, don't see anything that could lead to test failure.

FFroehlich · 2022-03-24T01:31:28Z

src/steadystateproblem.cpp

@@ -90,16 +90,21 @@ void SteadystateProblem::workSteadyStateBackwardProblem(
 void SteadystateProblem::findSteadyState(const Solver *solver, Model *model,
                                         int it) {
    steady_state_status_.resize(3, SteadyStateStatus::not_run);
+    bool turnOffNewton = solver->getSensitivityMethod() ==


Don't we generally wan't to turn off the newton solver if model->getSteadyStateSensitivityMode() == SteadyStateSensitivityMode::integrationOnly? Why also require forward sensitivities here?

Because in adjoint case steady-state computation part is independent of sensitivities computation.
If model->getSteadyStateSensitivityMode() == SteadyStateSensitivityMode::integrationOnly in adjoint case, I think model steady-state could still be found by Newton method.

Ah, think I am starting to understand. But that ignores solver->getSensitivityMethodPreequilibration(), right?
I think the best way is to go via SteadystateProblem::getSensitivityFlag here.

FFroehlich

Sorry, noticed a couple more things that need to be adressed :(

FFroehlich · 2022-03-24T22:59:15Z

src/steadystateproblem.cpp

+        (model->getSteadyStateSensitivityMode() == 
+         SteadyStateSensitivityMode::integrationOnly || 
+         model->getSteadyStateSensitivityMode() == 
+         SteadyStateSensitivityMode::integrateIfNewtonFails);

    bool simulationStartedInSteadystate =


I think this code requires some updates as the newton solver can't check if we started in a steadystate if it is turned off.

When is this an issue?

For models that start in steady-state (e.g., computed outside of amici or extracted from other simulations), we don't want to be forced to deactivate the newton solver. See #1422.

But thinking about it more carefully, we will only have misleading values if the newton solver is deactivated. In that case we will simply rextract sensitvities from the solver, which should be fine (but we probably perform a single, unnecessary simulaton step with sensitivities, which might be expensive).

FFroehlich · 2022-03-24T23:01:48Z

include/amici/defines.h

@@ -180,7 +180,8 @@ enum class NonlinearSolverIteration {
 /** Sensitivity computation mode in steadyStateProblem */
 enum class SteadyStateSensitivityMode {


I think we need some documentation of the behavior for the different values, for example in the function Model::setSteadyStateSensitivityMode(). https://github.com/AMICI-dev/AMICI/blob/master/python/examples/example_constant_species/ExampleEquilibrationLogic.ipynb also needs to be updated.

Yeah, this notebook definitely needs to be updated.

FFroehlich · 2022-03-24T23:03:53Z

src/steadystateproblem.cpp

@@ -90,16 +90,21 @@ void SteadystateProblem::workSteadyStateBackwardProblem(
 void SteadystateProblem::findSteadyState(const Solver *solver, Model *model,
                                         int it) {
    steady_state_status_.resize(3, SteadyStateStatus::not_run);
+    bool turnOffNewton = solver->getSensitivityMethod() ==


Ah, think I am starting to understand. But that ignores solver->getSensitivityMethodPreequilibration(), right?
I think the best way is to go via SteadystateProblem::getSensitivityFlag here.

plakrisenko · 2022-03-28T14:22:48Z

Sorry, noticed a couple more things that need to be adressed :(

You are right about the preequilibration.
I guess the logic should be rather
bool turnOffNewton = model->getSteadyStateSensitivityMode() == SteadyStateSensitivityMode::integrationOnly && (it == -1 && solver->getSensitivityMethodPreequilibration() == SensitivityMethod::forward || solver->getSensitivityMethod() == SensitivityMethod::forward);
Right?

FFroehlich · 2022-04-04T14:49:17Z

Sorry, noticed a couple more things that need to be adressed :(

You are right about the preequilibration. I guess the logic should be rather bool turnOffNewton = model->getSteadyStateSensitivityMode() == SteadyStateSensitivityMode::integrationOnly && (it == -1 && solver->getSensitivityMethodPreequilibration() == SensitivityMethod::forward || solver->getSensitivityMethod() == SensitivityMethod::forward); Right?

Yes, that looks correct.

FFroehlich

👍

…to stst_problem_settings

sonarqubecloud · 2022-04-05T00:38:19Z

SonarCloud Quality Gate failed.

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

0.0% Coverage
0.0% Duplication

Settings for specifying sensitivities computation method at steady state have been implemented/reworked. Three `SteadyStateSensitivityMode` are available for both `SensitivityMethod`s (`forward` and `adjoint`): - `newtonOnly`: only Newton's method is used for sensitivities computation - `integrationOnly`: only integration is used for sensitivities computation, which in forward sensitivity analysis case means that numerical integration has to be used to find the steady state - `integrateIfNewtonFails`: more flexible option, where simulation is used if Newton's method fails Note: Previously available `SteadyStateSensitivityMode` `simulationFSA` has been removed. Co-authored-by: Fabian Fröhlich <fabian_froehlich@hms.harvard.edu>

* exploit stoichiometric matrix * add solver mapping * fixup * fixups * fixup * fixup conservation laws * Steady state sensitivity modes (#1727) * Steady State Sensitivity modes * integrationOnly and integrateIfNewtonFails in forward sensitivity case * turn off newton method in case integrationOnly and forward sensitivity * python tests * Apply suggestions from code review Co-authored-by: Fabian Fröhlich <fabian_froehlich@hms.harvard.edu> * turn off newtin in findSteadyState * fix test_compare_conservation_laws_sbml * take preequilibration into account * fix test_compare_conservation_laws_sbml * quick notebook fix Co-authored-by: Fabian Fröhlich <fabian_froehlich@hms.harvard.edu> * update according to reviews Co-authored-by: Polina Lakrisenko <p.lakrisenko@gmail.com>

plakrisenko added 7 commits March 15, 2022 19:33

Steady State Sensitivity modes

4d37031

Merge branch 'develop' into stst_problem_settings

ffec862

integrationOnly and integrateIfNewtonFails in forward sensitivity case

3e9df5b

Merge branch 'develop' into stst_problem_settings

9d31820

turn off newton method in case integrationOnly and forward sensitivity

12eede5

python tests

2709d51

fix

e4d9222

plakrisenko marked this pull request as ready for review March 21, 2022 16:46

plakrisenko requested a review from FFroehlich March 21, 2022 16:46

FFroehlich requested changes Mar 21, 2022

View reviewed changes

tests/petab_test_suite/test_petab_suite.py Outdated Show resolved Hide resolved

tests/petab_test_suite/test_petab_suite.py Outdated Show resolved Hide resolved

src/steadystateproblem.cpp Outdated Show resolved Hide resolved

src/steadystateproblem.cpp Outdated Show resolved Hide resolved

plakrisenko and others added 3 commits March 21, 2022 18:24

Apply suggestions from code review

a515db3

Co-authored-by: Fabian Fröhlich <fabian_froehlich@hms.harvard.edu>

turn off newtin in findSteadyState

f974f7f

Merge branch 'stst_problem_settings' of github.com:AMICI-dev/AMICI in…

2c2a793

…to stst_problem_settings

fix test_compare_conservation_laws_sbml

f5d5e08

plakrisenko requested a review from FFroehlich March 23, 2022 16:07

FFroehlich approved these changes Mar 24, 2022

View reviewed changes

FFroehlich requested changes Mar 24, 2022

View reviewed changes

Merge branch 'develop' into stst_problem_settings

df6b96b

plakrisenko added 5 commits March 29, 2022 15:25

Merge branch 'develop' into stst_problem_settings

af8e0d6

fix

ee0ba61

fix

57f7fc4

take preequilibration into account

5f8dce1

add extra parentheses

2c197a4

plakrisenko requested a review from FFroehlich April 3, 2022 23:20

Merge branch 'develop' into stst_problem_settings

d32d64c

plakrisenko added 3 commits April 4, 2022 18:15

fix test_compare_conservation_laws_sbml

c6001d3

fix test_compare_conservation_laws_sbml

249f1a7

Merge branch 'develop' into stst_problem_settings

9fdcaf0

FFroehlich approved these changes Apr 4, 2022

View reviewed changes

plakrisenko mentioned this pull request Apr 4, 2022

AMICI: Update to renamed steady state sensitivity modes ICB-DCM/pyPESTO#843

Merged

plakrisenko added 3 commits April 4, 2022 22:03

Merge branch 'develop' into stst_problem_settings

93e0431

quick notebook fix

b39b925

Merge branch 'stst_problem_settings' of github.com:AMICI-dev/AMICI in…

8a45940

…to stst_problem_settings

plakrisenko merged commit 707ec6c into develop Apr 5, 2022

FFroehlich deleted the stst_problem_settings branch April 5, 2022 12:08

plakrisenko linked an issue Apr 5, 2022 that may be closed by this pull request

Add SteadyStateSensitivityMode(s) for Adjoint sensitivities computation in steadyStateProblem #1670

Closed

plakrisenko mentioned this pull request May 10, 2023

State sensitivities computation at steady state #2080

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Steady state sensitivity modes #1727

Steady state sensitivity modes #1727

plakrisenko commented Mar 16, 2022 •

edited

Loading

codecov bot commented Mar 21, 2022 •

edited

Loading

FFroehlich commented Mar 22, 2022

plakrisenko commented Mar 23, 2022

FFroehlich left a comment •

edited

Loading

FFroehlich Mar 24, 2022

plakrisenko Mar 24, 2022

FFroehlich Mar 24, 2022 •

edited

Loading

FFroehlich left a comment

FFroehlich Mar 24, 2022

plakrisenko Mar 28, 2022

FFroehlich Mar 28, 2022

FFroehlich Mar 28, 2022

FFroehlich Mar 24, 2022

plakrisenko Mar 25, 2022

FFroehlich Mar 24, 2022 •

edited

Loading

plakrisenko commented Mar 28, 2022

FFroehlich commented Apr 4, 2022

FFroehlich left a comment

sonarqubecloud bot commented Apr 5, 2022

		@@ -180,7 +180,8 @@ enum class NonlinearSolverIteration {
		/** Sensitivity computation mode in steadyStateProblem */
		enum class SteadyStateSensitivityMode {

Steady state sensitivity modes #1727

Steady state sensitivity modes #1727

Conversation

plakrisenko commented Mar 16, 2022 • edited Loading

codecov bot commented Mar 21, 2022 • edited Loading

Codecov Report

FFroehlich commented Mar 22, 2022

plakrisenko commented Mar 23, 2022

FFroehlich left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

FFroehlich Mar 24, 2022 • edited Loading

Choose a reason for hiding this comment

FFroehlich left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

FFroehlich Mar 24, 2022 • edited Loading

Choose a reason for hiding this comment

plakrisenko commented Mar 28, 2022

FFroehlich commented Apr 4, 2022

FFroehlich left a comment

Choose a reason for hiding this comment

sonarqubecloud bot commented Apr 5, 2022

plakrisenko commented Mar 16, 2022 •

edited

Loading

codecov bot commented Mar 21, 2022 •

edited

Loading

FFroehlich left a comment •

edited

Loading

FFroehlich Mar 24, 2022 •

edited

Loading

FFroehlich Mar 24, 2022 •

edited

Loading