fit.petab_v2.sciml¶

The neural networks of a PEtab SciML problem.

PEtab SciML is the extension sciml of PEtab v2 for hybrid problems, in which a model is combined with neural networks. This module translates the extension into the objects of sbmlsim.sciml, which know nothing of PEtab:

PEtab SciML	`sbmlsim`
a network of the block `neural_networks` with its array files	`Network`
`pre_initialization`, the hybridization table and the observables	`Hybridization` and its `NetworkPattern`
the rows of the mapping table	the ids of the inputs, outputs and arrays
the rows of the parameter table of a network	`FitParameter` per element
an array of a condition	an array of the simulation of an experiment

The module imports sbmlsim.sciml and with it petab_sciml, which is the extra sciml. sbmlsim.fit.petab_v2.reader imports it only for a problem with the extension.

petab.v2.Problem.from_yaml of petab 0.9.0 reads the networks of a problem through PyTorch. torch is no dependency of sbmlsim, so the tables of a problem are read by petab without the block of the extension, and the networks, the hybridization tables and the array files are read here.

SciMLProblemError ¶

SciMLProblemError(message, gap=None)

Bases: ValueError

A PEtab SciML problem which cannot be read.

Attributes:

Name	Type	Description
`gap`		id of the gap the problem runs into, see `sbmlsim.fit.petab_v2.gaps`, `None` for a problem which is not valid.

Initialize the error.

Parameters:

Name	Type	Description	Default
`message`	`str`	what cannot be read.	required
`gap`	`str \| None`	id of the gap the problem runs into.	`None`

NetworkEntity `dataclass` ¶

NetworkEntity(
    petab_id, network, kind, k=None, index=None, key=None
)

A part of a network a row of the mapping table names.

Attributes:

Name	Type	Description
`petab_id`	`str`	the `petabEntityId` of the row.
`network`	`str`	id of the network.
`kind`	`str`	`inputs`, `outputs` or `parameters`.
`k`	`int \| None`	position of the input or output, `None` for parameters.
`index`	`tuple[int, ...] \| None`	index of the element, `None` for an input which is an array and for parameters.
`key`	`str \| None`	the key of the parameters, e.g. `net1.layer1.weight`, see `sbmlsim.sciml.parameters`, `None` for an input or output.

SciMLReader ¶

SciMLReader(
    petab_problem,
    config,
    base_path,
    simulations,
    later_periods=None,
    fallbacks=None,
)

Read the networks of a PEtab SciML problem.

Attributes:

Name	Type	Description
`networks`	`dict[str, Network]`	the networks by their id, with the nominal values of the problem.
`entities`	`dict[str, NetworkEntity]`	the parts of the networks the mapping table names, by their `petabEntityId`.

Read the networks, the hybridization tables and the array files.

Parameters:

Name	Type	Description	Default
`petab_problem`	`Problem`	the problem, read without the block of the extension.	required
`config`	`SciMLConfig`	the block of the extension.	required
`base_path`	`Path`	directory the files of the problem are relative to.	required
`simulations`	`Mapping[str, list[str]]`	id of the simulation of every experiment -> the ids of the conditions the simulation starts with, which set the inputs of the networks.	required
`later_periods`	`Mapping[str, list[tuple[float, list[str]]]] \| None`	id of the simulation of every experiment -> the time and the ids of the conditions of every period after the first, which must not set an input of a network.	`None`
`fallbacks`	`Mapping[str, str] \| None`	id of an input -> its formula for the simulations without a formula of their own, in the math of PEtab, which the conditions repeat: the block `sbmlsim` of a problem which `sbmlsim` wrote, see `sbmlsim.fit.petab_v2.sciml_export`.	`None`

Raises:

Type Description

SciMLProblemError

if the problem has not one model, if a network is not in the format YAML, if a row of the mapping table names a network the problem does not have, if a row of the hybridization table cannot be read or assigns something else than an input of a network or an output of a network to an entity of the model, if a period after the first of an experiment sets an input of a network, if an input has no value, or if a parameter of a network has a prior.

NetworkImportError

if a network or its arrays cannot be read.

parameter_ids `property` ¶

parameter_ids

Get the ids of the parameter table which are arrays of networks.

input_ids `property` ¶

input_ids

Get the ids which are inputs of networks.

condition_ids `property` ¶

condition_ids

Get the conditions the array files have arrays for.

is_pre_initialization ¶

is_pre_initialization(sid)

Check whether a network runs before the simulation.

is_compiled_symbol ¶

is_compiled_symbol(symbol)

Check whether a symbol is used by a network which is compiled.

Such a symbol is a parameter of the model with the networks, so a parameter of the fit of its id is an entity of the model.

Parameters:

Name	Type	Description	Default
`symbol`	`str`	id of a parameter of the parameter table.	required

hybridizations ¶

hybridizations()

Get the hybridizations of the networks of the problem.

The pattern of a network is PRE_INITIALIZATION when the problem says so. Otherwise an output which an observable uses is OBSERVABLE and an output which the hybridization table assigns to an entity of the model is RHS, and a network with outputs of both kinds has two hybridizations.

The hybridizations are built once, so the model the networks are compiled into and the problem are built from the same objects.

Returns:

Type	Description
`list[Hybridization]`	The hybridizations, in the order of the networks of the problem.

Raises:

Type	Description
`SciMLProblemError`	if an output of the mapping table is not an element of the outputs of its network, if an output is assigned to a target and used by an observable, if a network has no output which is used, or if a network in the model has a layer without MathML of SBML (gap `sciml-layer-sbml`).
`NetworkHybridizationError`	if the inputs and outputs do not fit a network.

network_parameters ¶

network_parameters(sid)

Get the parameters of the fit which are elements of a network.

Parameters:

Name	Type	Description	Default
`sid`	`str`	id of the network.	required

Returns:

Type	Description
`list[FitParameter]`	One parameter per estimated element, see
`list[FitParameter]`	`sbmlsim.sciml.parameters.network_fit_parameters`. The elements of
`list[FitParameter]`	a network which runs before the simulation are not entities of
`list[FitParameter]`	the model. The parameters are resolved once, every call gets
`list[FitParameter]`	copies of them.

fit_parameters ¶

fit_parameters(unit_of)

Get the parameters of the fit which the networks add.

These are the estimated elements of the networks and the estimated parameters of the parameter table which are inputs of networks and not entities of the model.

Parameters:

Name	Type	Description	Default
`unit_of`	`Callable[[str], str \| None]`	gets the unit of an entity of the model.	required

Returns:

Type	Description
`list[FitParameter]`	The parameters, the inputs first and the elements in the order of
`list[FitParameter]`	the networks.

parse_entity ¶

parse_entity(petab_id, model_entity_id)

Read a row of the mapping table which names a part of a network.

Parameters:

Name	Type	Description	Default
`petab_id`	`str`	the `petabEntityId` of the row.	required
`model_entity_id`	`str`	the `modelEntityId`, e.g. `net1.inputs[0][1]`, `net1.inputs[0]`, `net1.outputs[0][0]`, `net1.parameters`, `net1.parameters[layer1]` or `net1.parameters[layer1].weight`.	required

Returns:

Type	Description
`NetworkEntity \| None`	The part of the network, `None` for a row which names an entity of a
`NetworkEntity \| None`	model.

Raises:

Type	Description
`SciMLProblemError`	if the row names a part of a network and is not one of the forms.

read_sciml_config ¶

read_sciml_config(extensions)

Get the block of the extension sciml of a problem.

Parameters:

Name	Type	Description	Default
`extensions`	`Mapping[str, Any] \| None`	the blocks of the extensions of the YAML of the problem.	required

Returns:

Type	Description
`SciMLConfig \| None`	The configuration, `None` for a problem without the extension.

Raises:

Type	Description
`SciMLProblemError`	if the block is not a block of the extension.

parameter_scale ¶

parameter_scale(parameter)

Get the scale of a parameter of the parameter table.

PEtab v2 has no scale of a parameter, it is a property of the optimization. The problems of PEtab SciML carry the column parameterScale of PEtab v1, which is read when it is there.

Parameters:

Name	Type	Description	Default
`parameter`	`Any`	parameter of the parameter table.	required

Returns:

Type	Description
`ParameterScaleType \| None`	The scale, `None` for a parameter without one.

Raises:

Type	Description
`SciMLProblemError`	if the scale is not `lin`, `log` or `log10`.

fit.petab_v2.sciml¶

SciMLProblemError ¶

NetworkEntity dataclass ¶

SciMLReader ¶

parameter_ids property ¶

input_ids property ¶

condition_ids property ¶

is_pre_initialization ¶

is_compiled_symbol ¶

hybridizations ¶

network_parameters ¶

fit_parameters ¶

parse_entity ¶

read_sciml_config ¶

parameter_scale ¶

NetworkEntity `dataclass` ¶

parameter_ids `property` ¶

input_ids `property` ¶

condition_ids `property` ¶