Exploit stoichiometric matrix in pysb import #1761

FFroehlich · 2022-04-04T22:49:06Z

Exploits stoichiometric matrix to compute dxdotdx_explicit, dxdotdp_explicit and dxdotdw as sparse matrix multiplication with derivatives of the flux vector that can be computed efficiently.

codecov · 2022-04-04T23:00:14Z

Codecov Report

Merging #1761 (eecfa99) into develop (707ec6c) will increase coverage by 0.23%.
The diff coverage is 98.41%.

@@             Coverage Diff             @@
##           develop    #1761      +/-   ##
===========================================
+ Coverage    77.77%   78.00%   +0.23%     
===========================================
  Files           74       74              
  Lines        12033    12096      +63     
===========================================
+ Hits          9359     9436      +77     
+ Misses        2674     2660      -14

Flag	Coverage Δ
cpp	`75.05% <ø> (+0.17%)`	⬆️
petab	`59.61% <87.30%> (+0.43%)`	⬆️
python	`68.69% <96.82%> (+0.45%)`	⬆️
sbmlsuite	`86.36% <11.11%> (-0.45%)`	⬇️

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

Impacted Files	Coverage Δ
python/amici/pysb_import.py	`94.59% <98.14%> (+0.68%)`	⬆️
python/amici/ode_export.py	`92.51% <100.00%> (+0.20%)`	⬆️
src/sundials_matrix_wrapper.cpp	`80.08% <0.00%> (-0.45%)`	⬇️
src/solver_cvodes.cpp	`70.43% <0.00%> (+0.16%)`	⬆️
src/amici.cpp	`73.13% <0.00%> (+5.22%)`	⬆️
src/exception.cpp	`81.08% <0.00%> (+5.40%)`	⬆️
src/spline.cpp	`71.57% <0.00%> (+6.31%)`	⬆️

FFroehlich · 2022-04-04T23:04:06Z

Ran two quick tests with smaller models. Looks like its faster even for small models and improvement increases with model size.

nx = 50 (1.8 fold improvement)

new:

2022-04-04 18:52:33.544 - amici.ode_export - INFO - Finished generating cpp code                    (1.59E+01s)

old:

2022-04-04 18:54:37.827 - amici.ode_export - INFO - Finished generating cpp code                    (2.94E+01s)

nx = 93 (2.1 fold improvement)
new:

2022-04-04 19:00:46.940 - amici.ode_export - INFO - Finished generating cpp code                    (5.59E+01s)

old:

2022-04-04 18:57:21.643 - amici.ode_export - INFO - Finished generating cpp code                    (1.17E+02s)

FFroehlich · 2022-04-05T00:31:00Z

For model with ~8k species, processing via stoichiometric matrix completes in ~7min where dxdotdw alone needs >20min and dxdotdx computation has been running for >18h.

* 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 <[email protected]> * 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 <[email protected]>