Class HierarchicalOptimizationWrapper

Inheritance Relationships

Base Type

Class Documentation

class HierarchicalOptimizationWrapper : public parpe::GradientFunction

The HierarchicalOptimizationWrapper class is a wrapper for hierarchical optimization of scaling parameters.

Parameters with the given indices are hidden by the wrapper and computed analytically internally.

Computes the negative log likelihood for normally distributed measurement (others to be added).

Public Functions

HierarchicalOptimizationWrapper(AmiciSummedGradientFunction *wrapped_function, int numConditions = 0, int numObservables = 0)

For testing.

Parameters:

  • fun –

  • numConditions –

  • numObservables –

  • numTimepoints –

HierarchicalOptimizationWrapper(AmiciSummedGradientFunction *wrapped_function, const H5::H5File &file, const std::string &hdf5RootPath, int numConditions, int numObservables, ErrorModel errorModel)

Get information on analytically computed parameters from HDF5 file.

Parameters:

  • fun –

  • file –

  • hdf5RootPath –

  • numConditions –

  • numObservables –

  • errorModel –

HierarchicalOptimizationWrapper(AmiciSummedGradientFunction *wrapped_function, std::unique_ptr<AnalyticalParameterProvider> scalingReader, std::unique_ptr<AnalyticalParameterProvider> offsetReader, std::unique_ptr<AnalyticalParameterProvider> sigmaReader, int numConditions, int numObservables, ErrorModel errorModel)

Get information on analytically computed parameters from the provided objects.

Parameters:

  • fun –

  • scalingReader –

  • offsetReader –

  • numConditions –

  • numObservables –

  • errorModel –

virtual FunctionEvaluationStatus evaluate(gsl::span<double const> parameters, double &fval, gsl::span<double> gradient, Logger *logger, double *cpuTime) const override
FunctionEvaluationStatus evaluate(gsl::span<double const> reducedParameters, double &fval, gsl::span<double> gradient, std::vector<double> &fullParameters, std::vector<double> &fullGradient, Logger *logger, double *cpuTime) const
std::vector<double> getDefaultScalingFactors() const

Get parameters for initial function evaluation.

Returns:

std::vector<double> getDefaultOffsetParameters() const

Get parameters for initial function evaluation.

Returns:

std::vector<double> getDefaultSigmaParameters() const
std::tuple<std::vector<std::vector<double>>, std::vector<std::vector<double>>> getUnscaledModelOutputsAndSigmas(const gsl::span<double const> reducedParameters, Logger *logger, double *cpuTime) const

Run simulations with scaling parameters set to 1.0 and collect model outputs.

Parameters:

reducedParameters – parameter vector for fun without scaling parameters

Returns:

Vector of double vectors containing AMICI ReturnData::y (nt x ny, column-major)

std::vector<double> computeAnalyticalScalings(std::vector<std::vector<double>> const &measurements, std::vector<std::vector<double>> const &modelOutputsUnscaled) const

Compute proportionality factors.

Parameters:

modelOutputs – Model outputs as provided by getModelOutputs

Returns:

the computed scaling factors

void applyOptimalScalings(std::vector<double> const &proportionalityFactors, std::vector<std::vector<double>> &modelOutputs) const
std::vector<double> computeAnalyticalOffsets(const std::vector<std::vector<double>> &measurements, std::vector<std::vector<double>> &modelOutputsUnscaled) const

Compute offset parameters.

Parameters:

modelOutputs – Model outputs as provided by getModelOutputs

Returns:

the computed offset parameters

std::vector<double> computeAnalyticalSigmas(std::vector<std::vector<double>> const &measurements, const std::vector<std::vector<double>> &modelOutputsScaled) const
void applyOptimalOffsets(std::vector<double> const &offsetParameters, std::vector<std::vector<double>> &modelOutputs) const
void fillInAnalyticalSigmas(std::vector<std::vector<double>> &allSigmas, const std::vector<double> &analyticalSigmas) const

Create vector with sigma matrix for each condition and timepoints from the given analytically computed sigmas.

Parameters:

sigmas –

Returns:

virtual FunctionEvaluationStatus evaluateWithOptimalParameters(std::vector<double> const &fullParameters, std::vector<double> const &sigmas, std::vector<std::vector<double>> const &measurements, std::vector<std::vector<double>> const &modelOutputsScaled, std::vector<std::vector<double>> &fullSigmaMatrices, double &fval, const gsl::span<double> gradient, std::vector<double> &fullGradient, Logger *logger, double *cpuTime) const

Evaluate fun using the computed optimal scaling and offset parameters.

Parameters:

  • reducedParameters – Parameter vector without scaling and offset parameters

  • scalings – Optimal scaling parameters

  • offsets – Optimal offset parameters

  • modelOutputsUnscaled – Model outputs before applying optimal offset and scaling parameters

  • fval – out: computed function value

  • gradient – out: computed function gradient

Returns:

virtual int numParameters() const override

Get number of parameters the function expects.

Returns:

That

int numProportionalityFactors() const
std::vector<int> const &getProportionalityFactorIndices() const
int numOffsetParameters() const
int numSigmaParameters() const
std::vector<int> const &getOffsetParameterIndices() const
std::vector<int> const &getSigmaParameterIndices() const
std::vector<int> getAnalyticalParameterIndices() const
AmiciSummedGradientFunction *getWrappedFunction() const
virtual std::vector<std::string> getParameterIds() const override
FunctionEvaluationStatus evaluate(gsl::span<double const> parameters, double &fval, gsl::span<double> gradient) const

Evaluate the function f(x)

Parameters:

  • parameters – Point x at which to evaluate f(x). Must be of length numParameters().

  • fval – (output) Will be set to the function value f(x)

  • gradient – (output) If not gsl::span<double>(), will contain the gradient of f(x) at x. Must be of length numParameters().

Returns:

functionEvaluationSuccess on success, functionEvaluationFailure otherwise

FunctionEvaluationStatus evaluate(gsl::span<double const> parameters, double &fval, gsl::span<double> gradient, Logger *logger, double *cpuTime) const = 0