amici.gradient_check

Finite Difference Check

This module provides functions to automatically check correctness of amici computed sensitivities using finite difference approximations

Functions Summary

assert_fun(x)
check_close(result, expected, assert_fun, …)	Compares computed values against expected values and provides rich output information.
check_derivatives(model, solver[, edata, …])	Finite differences check for likelihood gradient.
check_finite_difference(x0, model, solver, …)	Checks the computed sensitivity based derivatives against a finite difference approximation.
check_results(rdata, field, expected, …)	Checks whether rdata[field] agrees with expected according to provided tolerances.

Functions

amici.gradient_check.assert_fun(x)

amici.gradient_check.check_close(result, expected, assert_fun, atol, rtol, field, ip=None)

Compares computed values against expected values and provides rich output information.

Parameters

• result (numpy.array) – computed values

• expected (numpy.array) – expected values

• field (str) – rdata field for which the gradient is checked, only for error reporting

• assert_fun (typing.Callable) – function that asserts the return values of comparison, enables passing of custom assert function from testing frameworks

• atol (float) – absolute tolerance for comparison

• rtol (float) – relative tolerance for comparison

• ip (typing.Optional[int]) – parameter index

Return type

None

amici.gradient_check.check_derivatives(model, solver, edata=None, assert_fun=<function assert_fun>, atol=0.0001, rtol=0.0001, epsilon=0.001, check_least_squares=True, skip_zero_pars=False)

Finite differences check for likelihood gradient.

Parameters

• model (amici.amici.Model) – amici model

• solver (amici.amici.Solver) – amici solver

• edata (typing.Optional[amici.ExpData]) – exp data

• assert_fun (typing.Optional[typing.Callable]) – function that asserts the return values of comparison, enables passing of custom assert function from testing frameworks

• atol (typing.Optional[float]) – absolute tolerance for comparison

• rtol (typing.Optional[float]) – relative tolerance for comparison

• epsilon (typing.Optional[float]) – finite difference step-size

• check_least_squares (bool) – whether to check least squares related values.

• skip_zero_pars (bool) – whether to perform FD checks for parameters that are zero

Return type

None

amici.gradient_check.check_finite_difference(x0, model, solver, edata, ip, fields, assert_fun, atol=0.0001, rtol=0.0001, epsilon=0.001)

Checks the computed sensitivity based derivatives against a finite difference approximation.

Parameters

• x0 (typing.Sequence[float]) – parameter value at which to check finite difference approximation

• model (amici.amici.Model) – amici model

• solver (amici.amici.Solver) – amici solver

• edata (amici.ExpData) – exp data

• ip (int) – parameter index

• fields (typing.List[str]) – rdata fields for which to check the gradient

• assert_fun (typing.Callable) – function that asserts the return values of comparison, enables passing of custom assert function from testing frameworks

• atol (typing.Optional[float]) – absolute tolerance for comparison

• rtol (typing.Optional[float]) – relative tolerance for comparison

• epsilon (typing.Optional[float]) – finite difference step-size

Return type

None

amici.gradient_check.check_results(rdata, field, expected, assert_fun, atol, rtol)

Checks whether rdata[field] agrees with expected according to provided tolerances.

Parameters

• rdata (amici.amici.ReturnData) – simulation results as returned by amici.amici.runAmiciSimulation()

• field (str) – name of the field to check

• expected (numpy.array) – expected values

• assert_fun (typing.Callable) – function that asserts the return values of comparison, enables passing of custom assert function from testing frameworks

• atol (float) – absolute tolerance for comparison

• rtol (float) – relative tolerance for comparison

Return type

None