MATRIXLAB·MULTIVERSE

20 packages · AD added

Back to CodeUniverse

LALSuite · /code/lalsuite/

CodeUniverse · package

LALSuite

many-parameter correlated likelihood gradient

AD speed-up on the same differentiation workload (against finite differences) Speed-up: 8.6×

High-dimensional demo

many-parameter correlated likelihood gradient

Supplied demo scenario wording.

FD components requested 1 – 15 · AD components 15

View demo measurements
LALSuite · High-dimensional demo
MeasureValue
AD median (s)0.000018291990272700787
FD median (s)0.0001580000389367342
FD / AD8.637662527764165
Max scaled error4.089917656346624e-9
AD components15
FD components requested15
AD parameters15
Repeats51
Gradient cost by requested FD components
FD components requestedAD median (s)FD median (s)FD / ADMax scaled errorAD components
1150.0000167499529197812080.000010250019840896130.61194320306364291.3652293417980917e-9
2150.000018500024452805520.0000221249647438526151.19594245944347271.3652293417980917e-9
4150.0000185829121619462970.000045541906729340552.45074110733840331.3652293417980917e-9
8150.0000185829121619462970.000087750027887523174.7220816157768794.089917656346624e-9
15150.0000182919902727007870.00015800003893673428.6376625277641654.089917656346624e-9

One-direction performance audit

View audit measurements
Demo scenarioProblem sizeRepeatsAD median (s)FD median (s)FD / ADRelative errorSupplied verdict
TaylorF2_phi_ref1025—0.0000594170.000050250.8457175560.00000266667—
IMRPhenomXHM_distance1025—0.0004866250.0008863751.82147444——

FD/AD > 1 means automatic differentiation is faster; FD/AD < 1 means the finite-difference comparison is faster. One direction only, measured inside the package's own case.