Models

LUMEBaseModel

Bases: BaseModel, ABC

Abstract base class for models using lume-model variables.

Inheriting classes must define the evaluate method and variable names must be unique (respectively). Models build using this framework will be compatible with the lume-epics EPICS server and associated tools.

Attributes:

Name	Type	Description
input_variables	list[SerializeAsAny[InputVariable]]	List defining the input variables and their order.
output_variables	list[SerializeAsAny[OutputVariable]]	List defining the output variables and their order.

Source code in lume_model/base.py

class LUMEBaseModel(BaseModel, ABC):
    """Abstract base class for models using lume-model variables.

    Inheriting classes must define the evaluate method and variable names must be unique (respectively).
    Models build using this framework will be compatible with the lume-epics EPICS server and associated tools.

    Attributes:
        input_variables: List defining the input variables and their order.
        output_variables: List defining the output variables and their order.
    """
    input_variables: list[SerializeAsAny[InputVariable]]
    output_variables: list[SerializeAsAny[OutputVariable]]

    model_config = ConfigDict(arbitrary_types_allowed=True, validate_assignment=True)

    @field_validator("input_variables", mode="before")
    def validate_input_variables(cls, value):
        new_value = []
        if isinstance(value, dict):
            for name, val in value.items():
                if isinstance(val, dict):
                    if val["variable_type"] == "scalar":
                        new_value.append(ScalarInputVariable(name=name, **val))
                elif isinstance(val, InputVariable):
                    new_value.append(val)
                else:
                    raise TypeError(f"type {type(val)} not supported")
        elif isinstance(value, list):
            new_value = value

        return new_value

    @field_validator("output_variables", mode="before")
    def validate_output_variables(cls, value):
        new_value = []
        if isinstance(value, dict):
            for name, val in value.items():
                if isinstance(val, dict):
                    if val["variable_type"] == "scalar":
                        new_value.append(ScalarOutputVariable(name=name, **val))
                elif isinstance(val, OutputVariable):
                    new_value.append(val)
                else:
                    raise TypeError(f"type {type(val)} not supported")
        elif isinstance(value, list):
            new_value = value

        return new_value

    def __init__(self, *args, **kwargs):
        """Initializes LUMEBaseModel.

        Args:
            *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
              formatted string or file path.
            **kwargs: See class attributes.
        """
        if len(args) == 1:
            if len(kwargs) > 0:
                raise ValueError("Cannot specify YAML string and keyword arguments for LUMEBaseModel init.")
            super().__init__(**parse_config(args[0], self.model_fields))
        elif len(args) > 1:
            raise ValueError(
                "Arguments to LUMEBaseModel must be either a single YAML string "
                "or keyword arguments passed directly to pydantic."
            )
        else:
            super().__init__(**kwargs)

    @field_validator("input_variables", "output_variables")
    def unique_variable_names(cls, value):
        verify_unique_variable_names(value)
        return value

    @property
    def input_names(self) -> list[str]:
        return [var.name for var in self.input_variables]

    @property
    def output_names(self) -> list[str]:
        return [var.name for var in self.output_variables]

    @abstractmethod
    def evaluate(self, input_dict: dict[str, Any]) -> dict[str, Any]:
        pass

    def to_json(self, **kwargs) -> str:
        return json_dumps(self, **kwargs)

    def dict(self, **kwargs) -> dict[str, Any]:
        config = super().model_dump(**kwargs)
        return {"model_class": self.__class__.__name__} | config

    def json(self, **kwargs) -> str:
        result = self.to_json(**kwargs)
        config = json.loads(result)
        config = {"model_class": self.__class__.__name__} | config
        return json.dumps(config)

    def yaml(
            self,
            base_key: str = "",
            file_prefix: str = "",
            save_models: bool = False,
    ) -> str:
        """Serializes the object and returns a YAML formatted string defining the model.

        Args:
            base_key: Base key for serialization.
            file_prefix: Prefix for generated filenames.
            save_models: Determines whether models are saved to file.

        Returns:
            YAML formatted string defining the model.
        """
        output = json.loads(
            self.to_json(
                base_key=base_key,
                file_prefix=file_prefix,
                save_models=save_models,
            )
        )
        s = yaml.dump({"model_class": self.__class__.__name__} | output,
                      default_flow_style=None, sort_keys=False)
        return s

    def dump(
            self,
            file: Union[str, os.PathLike],
            base_key: str = "",
            save_models: bool = True,
    ):
        """Returns and optionally saves YAML formatted string defining the model.

        Args:
            file: File path to which the YAML formatted string and corresponding files are saved.
            base_key: Base key for serialization.
            save_models: Determines whether models are saved to file.
        """
        file_prefix = os.path.splitext(os.path.abspath(file))[0]
        with open(file, "w") as f:
            f.write(
                self.yaml(
                    base_key=base_key,
                    file_prefix=file_prefix,
                    save_models=save_models,
                )
            )

    @classmethod
    def from_file(cls, filename: str):
        if not os.path.exists(filename):
            raise OSError(f"File {filename} is not found.")
        with open(filename, "r") as file:
            return cls.from_yaml(file)

    @classmethod
    def from_yaml(cls, yaml_obj: [str, TextIOWrapper]):
        return cls.model_validate(parse_config(yaml_obj, cls.model_fields))

__init__(*args, **kwargs)

Initializes LUMEBaseModel.

Parameters:

Name	Type	Description	Default
*args		Accepts a single argument which is the model configuration as dictionary, YAML or JSON formatted string or file path.	()
**kwargs		See class attributes.	{}

Source code in lume_model/base.py

def __init__(self, *args, **kwargs):
    """Initializes LUMEBaseModel.

    Args:
        *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
          formatted string or file path.
        **kwargs: See class attributes.
    """
    if len(args) == 1:
        if len(kwargs) > 0:
            raise ValueError("Cannot specify YAML string and keyword arguments for LUMEBaseModel init.")
        super().__init__(**parse_config(args[0], self.model_fields))
    elif len(args) > 1:
        raise ValueError(
            "Arguments to LUMEBaseModel must be either a single YAML string "
            "or keyword arguments passed directly to pydantic."
        )
    else:
        super().__init__(**kwargs)

dump(file, base_key='', save_models=True)

Returns and optionally saves YAML formatted string defining the model.

Parameters:

Name	Type	Description	Default
file	Union[str, PathLike]	File path to which the YAML formatted string and corresponding files are saved.	required
base_key	str	Base key for serialization.	''
save_models	bool	Determines whether models are saved to file.	True

Source code in lume_model/base.py

def dump(
        self,
        file: Union[str, os.PathLike],
        base_key: str = "",
        save_models: bool = True,
):
    """Returns and optionally saves YAML formatted string defining the model.

    Args:
        file: File path to which the YAML formatted string and corresponding files are saved.
        base_key: Base key for serialization.
        save_models: Determines whether models are saved to file.
    """
    file_prefix = os.path.splitext(os.path.abspath(file))[0]
    with open(file, "w") as f:
        f.write(
            self.yaml(
                base_key=base_key,
                file_prefix=file_prefix,
                save_models=save_models,
            )
        )

yaml(base_key='', file_prefix='', save_models=False)

Serializes the object and returns a YAML formatted string defining the model.

Parameters:

Name	Type	Description	Default
base_key	str	Base key for serialization.	''
file_prefix	str	Prefix for generated filenames.	''
save_models	bool	Determines whether models are saved to file.	False

Returns:

Type	Description
str	YAML formatted string defining the model.

Source code in lume_model/base.py

def yaml(
        self,
        base_key: str = "",
        file_prefix: str = "",
        save_models: bool = False,
) -> str:
    """Serializes the object and returns a YAML formatted string defining the model.

    Args:
        base_key: Base key for serialization.
        file_prefix: Prefix for generated filenames.
        save_models: Determines whether models are saved to file.

    Returns:
        YAML formatted string defining the model.
    """
    output = json.loads(
        self.to_json(
            base_key=base_key,
            file_prefix=file_prefix,
            save_models=save_models,
        )
    )
    s = yaml.dump({"model_class": self.__class__.__name__} | output,
                  default_flow_style=None, sort_keys=False)
    return s

TorchModel

Bases: LUMEBaseModel

LUME-model class for torch models.

By default, the models are assumed to be fixed, so all gradient computation is deactivated and the model and transformers are put in evaluation mode.

Attributes:

Name	Type	Description
model	Module	The torch base model.
input_variables	Module	List defining the input variables and their order.
output_variables	Module	List defining the output variables and their order.
input_transformers	list[ReversibleInputTransform]	List of transformer objects to apply to input before passing to model.
output_transformers	list[ReversibleInputTransform]	List of transformer objects to apply to output of model.
output_format	str	Determines format of outputs: "tensor", "variable" or "raw".
device	Union[device, str]	Device on which the model will be evaluated. Defaults to "cpu".
fixed_model	bool	If true, the model and transformers are put in evaluation mode and all gradient computation is deactivated.

Source code in lume_model/models/torch_model.py

class TorchModel(LUMEBaseModel):
    """LUME-model class for torch models.

    By default, the models are assumed to be fixed, so all gradient computation is deactivated and the model and
    transformers are put in evaluation mode.

    Attributes:
        model: The torch base model.
        input_variables: List defining the input variables and their order.
        output_variables: List defining the output variables and their order.
        input_transformers: List of transformer objects to apply to input before passing to model.
        output_transformers: List of transformer objects to apply to output of model.
        output_format: Determines format of outputs: "tensor", "variable" or "raw".
        device: Device on which the model will be evaluated. Defaults to "cpu".
        fixed_model: If true, the model and transformers are put in evaluation mode and all gradient
          computation is deactivated.
    """
    model: torch.nn.Module
    input_transformers: list[ReversibleInputTransform] = []
    output_transformers: list[ReversibleInputTransform] = []
    output_format: str = "tensor"
    device: Union[torch.device, str] = "cpu"
    fixed_model: bool = True

    def __init__(self, *args, **kwargs):
        """Initializes TorchModel.

        Args:
            *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
              formatted string or file path.
            **kwargs: See class attributes.
        """
        super().__init__(*args, **kwargs)

        # set precision
        self.model.to(dtype=self.dtype)
        for t in self.input_transformers + self.output_transformers:
            if isinstance(t, torch.nn.Module):
                t.to(dtype=self.dtype)

        # fixed model: set full model in eval mode and deactivate all gradients
        if self.fixed_model:
            self.model.eval().requires_grad_(False)
            for t in self.input_transformers + self.output_transformers:
                if isinstance(t, torch.nn.Module):
                    t.eval().requires_grad_(False)

        # ensure consistent device
        self.to(self.device)

    @property
    def dtype(self):
        return torch.double

    @property
    def _tkwargs(self):
        return {"device": self.device, "dtype": self.dtype}

    @field_validator("model", mode="before")
    def validate_torch_model(cls, v):
        if isinstance(v, (str, os.PathLike)):
            if os.path.exists(v):
                v = torch.load(v)
            else:
                raise OSError(f"File {v} is not found.")
        return v

    @field_validator("input_transformers", "output_transformers", mode="before")
    def validate_botorch_transformers(cls, v):
        if not isinstance(v, list):
            raise ValueError("Transformers must be passed as list.")
        loaded_transformers = []
        for t in v:
            if isinstance(t, (str, os.PathLike)):
                if os.path.exists(t):
                    t = torch.load(t)
                else:
                    raise OSError(f"File {t} is not found.")
            loaded_transformers.append(t)
        v = loaded_transformers
        return v

    @field_validator("output_format")
    def validate_output_format(cls, v):
        supported_formats = ["tensor", "variable", "raw"]
        if v not in supported_formats:
            raise ValueError(f"Unknown output format {v}, expected one of {supported_formats}.")
        return v

    def evaluate(
            self,
            input_dict: dict[str, Union[InputVariable, float, torch.Tensor]],
    ) -> dict[str, Union[OutputVariable, float, torch.Tensor]]:
        """Evaluates model on the given input dictionary.

        Args:
            input_dict: Input dictionary on which to evaluate the model.

        Returns:
            Dictionary of output variable names to values.
        """
        formatted_inputs = self._format_inputs(input_dict)
        input_tensor = self._arrange_inputs(formatted_inputs)
        input_tensor = self._transform_inputs(input_tensor)
        output_tensor = self.model(input_tensor)
        output_tensor = self._transform_outputs(output_tensor)
        parsed_outputs = self._parse_outputs(output_tensor)
        output_dict = self._prepare_outputs(parsed_outputs)
        return output_dict

    def random_input(self, n_samples: int = 1) -> dict[str, torch.Tensor]:
        """Generates random input(s) for the model.

        Args:
            n_samples: Number of random samples to generate.

        Returns:
            Dictionary of input variable names to tensors.
        """
        input_dict = {}
        for var in self.input_variables:
            if isinstance(var, ScalarInputVariable):
                input_dict[var.name] = var.value_range[0] + torch.rand(size=(n_samples,)) * (
                            var.value_range[1] - var.value_range[0])
            else:
                torch.tensor(var.default, **self._tkwargs).repeat((n_samples, 1))
        return input_dict

    def random_evaluate(self, n_samples: int = 1) -> dict[str, Union[OutputVariable, float, torch.Tensor]]:
        """Returns random evaluation(s) of the model.

        Args:
            n_samples: Number of random samples to evaluate.

        Returns:
            Dictionary of variable names to outputs.
        """
        random_input = self.random_input(n_samples)
        return self.evaluate(random_input)

    def to(self, device: Union[torch.device, str]):
        """Updates the device for the model, transformers and default values.

        Args:
            device: Device on which the model will be evaluated.
        """
        self.model.to(device)
        for t in self.input_transformers + self.output_transformers:
            if isinstance(t, torch.nn.Module):
                t.to(device)
        self.device = device

    def insert_input_transformer(self, new_transformer: ReversibleInputTransform, loc: int):
        """Inserts an additional input transformer at the given location.

        Args:
            new_transformer: New transformer to add.
            loc: Location where the new transformer shall be added to the transformer list.
        """
        self.input_transformers = (self.input_transformers[:loc] + [new_transformer] +
                                   self.input_transformers[loc:])

    def insert_output_transformer(self, new_transformer: ReversibleInputTransform, loc: int):
        """Inserts an additional output transformer at the given location.

        Args:
            new_transformer: New transformer to add.
            loc: Location where the new transformer shall be added to the transformer list.
        """
        self.output_transformers = (self.output_transformers[:loc] + [new_transformer] +
                                    self.output_transformers[loc:])

    def update_input_variables_to_transformer(self, transformer_loc: int) -> list[InputVariable]:
        """Returns input variables updated to the transformer at the given location.

        Updated are the value ranges and default of the input variables. This allows, e.g., to add a
        calibration transformer and to update the input variable specification accordingly.

        Args:
            transformer_loc: The location of the input transformer to adjust for.

        Returns:
            The updated input variables.
        """
        x_old = {
            "min": torch.tensor([var.value_range[0] for var in self.input_variables], dtype=self.dtype),
            "max": torch.tensor([var.value_range[1] for var in self.input_variables], dtype=self.dtype),
            "default": torch.tensor([var.default for var in self.input_variables], dtype=self.dtype),
        }
        x_new = {}
        for key in x_old.keys():
            x = x_old[key]
            # compute previous limits at transformer location
            for i in range(transformer_loc):
                x = self.input_transformers[i].transform(x)
            # untransform of transformer to adjust for
            x = self.input_transformers[transformer_loc].untransform(x)
            # backtrack through transformers
            for transformer in self.input_transformers[:transformer_loc][::-1]:
                x = transformer.untransform(x)
            x_new[key] = x
        updated_variables = deepcopy(self.input_variables)
        for i, var in enumerate(updated_variables):
            var.value_range = [x_new["min"][i].item(), x_new["max"][i].item()]
            var.default = x_new["default"][i].item()
        return updated_variables

    def _format_inputs(
            self,
            input_dict: dict[str, Union[InputVariable, float, torch.Tensor]],
    ) -> dict[str, torch.Tensor]:
        """Formats values of the input dictionary as tensors.

        Args:
            input_dict: Dictionary of input variable names to values.

        Returns:
            Dictionary of input variable names to tensors.
        """
        # NOTE: The input variable is only updated if a singular value is given (ambiguous otherwise)
        formatted_inputs = {}
        for var_name, var in input_dict.items():
            if isinstance(var, InputVariable):
                formatted_inputs[var_name] = torch.tensor(var.value, **self._tkwargs)
                # self.input_variables[self.input_names.index(var_name)].value = var.value
            elif isinstance(var, float):
                formatted_inputs[var_name] = torch.tensor(var, **self._tkwargs)
                # self.input_variables[self.input_names.index(var_name)].value = var
            elif isinstance(var, torch.Tensor):
                var = var.double().squeeze().to(self.device)
                formatted_inputs[var_name] = var
                # if var.dim() == 0:
                #     self.input_variables[self.input_names.index(var_name)].value = var.item()
            else:
                TypeError(
                    f"Unknown type {type(var)} passed to evaluate."
                    f"Should be one of InputVariable, float or torch.Tensor."
                )
        return formatted_inputs

    def _arrange_inputs(self, formatted_inputs: dict[str, torch.Tensor]) -> torch.Tensor:
        """Enforces order of input variables.

        Enforces the order of the input variables to be passed to the transformers and model and updates the
        returned tensor with default values for any inputs that are missing.

        Args:
            formatted_inputs: Dictionary of input variable names to tensors.

        Returns:
            Ordered input tensor to be passed to the transformers.
        """
        default_tensor = torch.tensor(
            [var.default for var in self.input_variables], **self._tkwargs
        )

        # determine input shape
        input_shapes = [formatted_inputs[k].shape for k in formatted_inputs.keys()]
        if not all(ele == input_shapes[0] for ele in input_shapes):
            raise ValueError("Inputs have inconsistent shapes.")

        input_tensor = torch.tile(default_tensor, dims=(*input_shapes[0], 1))
        for key, value in formatted_inputs.items():
            input_tensor[..., self.input_names.index(key)] = value

        if input_tensor.shape[-1] != len(self.input_names):
            raise ValueError(
                f"""
                Last dimension of input tensor doesn't match the expected number of inputs\n
                received: {default_tensor.shape}, expected {len(self.input_names)} as the last dimension
                """
            )
        return input_tensor

    def _transform_inputs(self, input_tensor: torch.Tensor) -> torch.Tensor:
        """Applies transformations to the inputs.

        Args:
            input_tensor: Ordered input tensor to be passed to the transformers.

        Returns:
            Tensor of transformed inputs to be passed to the model.
        """
        for transformer in self.input_transformers:
            input_tensor = transformer.transform(input_tensor)
        return input_tensor

    def _transform_outputs(self, output_tensor: torch.Tensor) -> torch.Tensor:
        """(Un-)Transforms the model output tensor.

        Args:
            output_tensor: Output tensor from the model.

        Returns:
            (Un-)Transformed output tensor.
        """
        for transformer in self.output_transformers:
            output_tensor = transformer.untransform(output_tensor)
        return output_tensor

    def _parse_outputs(self, output_tensor: torch.Tensor) -> dict[str, torch.Tensor]:
        """Constructs dictionary from model output tensor.

        Args:
            output_tensor: (Un-)transformed output tensor from the model.

        Returns:
            Dictionary of output variable names to (un-)transformed tensors.
        """
        parsed_outputs = {}
        if output_tensor.dim() in [0, 1]:
            output_tensor = output_tensor.unsqueeze(0)
        if len(self.output_names) == 1:
            parsed_outputs[self.output_names[0]] = output_tensor.squeeze()
        else:
            for idx, output_name in enumerate(self.output_names):
                parsed_outputs[output_name] = output_tensor[..., idx].squeeze()
        return parsed_outputs

    def _prepare_outputs(
            self,
            parsed_outputs: dict[str, torch.Tensor],
    ) -> dict[str, Union[OutputVariable, torch.Tensor]]:
        """Updates and returns outputs according to output_format.

        Updates the output variables within the model to reflect the new values.

        Args:
            parsed_outputs: Dictionary of output variable names to transformed tensors.

        Returns:
            Dictionary of output variable names to values depending on output_format.
        """
        # for var in self.output_variables:
        #     if parsed_outputs[var.name].dim() == 0:
        #         idx = self.output_names.index(var.name)
        #         if isinstance(var, ScalarOutputVariable):
        #             self.output_variables[idx].value = parsed_outputs[var.name].item()
        #         elif isinstance(var, ImageOutputVariable):
        #             # OutputVariables should be numpy arrays
        #             self.output_variables[idx].value = (parsed_outputs[var.name].reshape(var.shape).numpy())
        #             self._update_image_limits(var, parsed_outputs)

        if self.output_format == "tensor":
            return parsed_outputs
        elif self.output_format == "variable":
            output_dict = {var.name: var for var in self.output_variables}
            for var in output_dict.values():
                var.value = parsed_outputs[var.name].item()
            return output_dict
            # return {var.name: var for var in self.output_variables}
        else:
            return {key: value.item() if value.squeeze().dim() == 0 else value
                    for key, value in parsed_outputs.items()}
            # return {var.name: var.value for var in self.output_variables}

    def _update_image_limits(
            self,
            variable: OutputVariable, predicted_output: dict[str, torch.Tensor],
    ):
        output_idx = self.output_names.index(variable.name)
        if self.output_variables[output_idx].x_min_variable:
            self.output_variables[output_idx].x_min = predicted_output[
                self.output_variables[output_idx].x_min_variable
            ].item()

        if self.output_variables[output_idx].x_max_variable:
            self.output_variables[output_idx].x_max = predicted_output[
                self.output_variables[output_idx].x_max_variable
            ].item()

        if self.output_variables[output_idx].y_min_variable:
            self.output_variables[output_idx].y_min = predicted_output[
                self.output_variables[output_idx].y_min_variable
            ].item()

        if self.output_variables[output_idx].y_max_variable:
            self.output_variables[output_idx].y_max = predicted_output[
                self.output_variables[output_idx].y_max_variable
            ].item()

__init__(*args, **kwargs)

Initializes TorchModel.

Parameters:

Name	Type	Description	Default
*args		Accepts a single argument which is the model configuration as dictionary, YAML or JSON formatted string or file path.	()
**kwargs		See class attributes.	{}

Source code in lume_model/models/torch_model.py

def __init__(self, *args, **kwargs):
    """Initializes TorchModel.

    Args:
        *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
          formatted string or file path.
        **kwargs: See class attributes.
    """
    super().__init__(*args, **kwargs)

    # set precision
    self.model.to(dtype=self.dtype)
    for t in self.input_transformers + self.output_transformers:
        if isinstance(t, torch.nn.Module):
            t.to(dtype=self.dtype)

    # fixed model: set full model in eval mode and deactivate all gradients
    if self.fixed_model:
        self.model.eval().requires_grad_(False)
        for t in self.input_transformers + self.output_transformers:
            if isinstance(t, torch.nn.Module):
                t.eval().requires_grad_(False)

    # ensure consistent device
    self.to(self.device)

evaluate(input_dict)

Evaluates model on the given input dictionary.

Parameters:

Name	Type	Description	Default
input_dict	dict[str, Union[InputVariable, float, Tensor]]	Input dictionary on which to evaluate the model.	required

Returns:

Type	Description
dict[str, Union[OutputVariable, float, Tensor]]	Dictionary of output variable names to values.

Source code in lume_model/models/torch_model.py

def evaluate(
        self,
        input_dict: dict[str, Union[InputVariable, float, torch.Tensor]],
) -> dict[str, Union[OutputVariable, float, torch.Tensor]]:
    """Evaluates model on the given input dictionary.

    Args:
        input_dict: Input dictionary on which to evaluate the model.

    Returns:
        Dictionary of output variable names to values.
    """
    formatted_inputs = self._format_inputs(input_dict)
    input_tensor = self._arrange_inputs(formatted_inputs)
    input_tensor = self._transform_inputs(input_tensor)
    output_tensor = self.model(input_tensor)
    output_tensor = self._transform_outputs(output_tensor)
    parsed_outputs = self._parse_outputs(output_tensor)
    output_dict = self._prepare_outputs(parsed_outputs)
    return output_dict

insert_input_transformer(new_transformer, loc)

Inserts an additional input transformer at the given location.

Parameters:

Name	Type	Description	Default
new_transformer	ReversibleInputTransform	New transformer to add.	required
loc	int	Location where the new transformer shall be added to the transformer list.	required

Source code in lume_model/models/torch_model.py

def insert_input_transformer(self, new_transformer: ReversibleInputTransform, loc: int):
    """Inserts an additional input transformer at the given location.

    Args:
        new_transformer: New transformer to add.
        loc: Location where the new transformer shall be added to the transformer list.
    """
    self.input_transformers = (self.input_transformers[:loc] + [new_transformer] +
                               self.input_transformers[loc:])

insert_output_transformer(new_transformer, loc)

Inserts an additional output transformer at the given location.

Parameters:

Name	Type	Description	Default
new_transformer	ReversibleInputTransform	New transformer to add.	required
loc	int	Location where the new transformer shall be added to the transformer list.	required

Source code in lume_model/models/torch_model.py

def insert_output_transformer(self, new_transformer: ReversibleInputTransform, loc: int):
    """Inserts an additional output transformer at the given location.

    Args:
        new_transformer: New transformer to add.
        loc: Location where the new transformer shall be added to the transformer list.
    """
    self.output_transformers = (self.output_transformers[:loc] + [new_transformer] +
                                self.output_transformers[loc:])

random_evaluate(n_samples=1)

Returns random evaluation(s) of the model.

Parameters:

Name	Type	Description	Default
n_samples	int	Number of random samples to evaluate.	1

Returns:

Type	Description
dict[str, Union[OutputVariable, float, Tensor]]	Dictionary of variable names to outputs.

Source code in lume_model/models/torch_model.py

def random_evaluate(self, n_samples: int = 1) -> dict[str, Union[OutputVariable, float, torch.Tensor]]:
    """Returns random evaluation(s) of the model.

    Args:
        n_samples: Number of random samples to evaluate.

    Returns:
        Dictionary of variable names to outputs.
    """
    random_input = self.random_input(n_samples)
    return self.evaluate(random_input)

random_input(n_samples=1)

Generates random input(s) for the model.

Parameters:

Name	Type	Description	Default
n_samples	int	Number of random samples to generate.	1

Returns:

Type	Description
dict[str, Tensor]	Dictionary of input variable names to tensors.

Source code in lume_model/models/torch_model.py

def random_input(self, n_samples: int = 1) -> dict[str, torch.Tensor]:
    """Generates random input(s) for the model.

    Args:
        n_samples: Number of random samples to generate.

    Returns:
        Dictionary of input variable names to tensors.
    """
    input_dict = {}
    for var in self.input_variables:
        if isinstance(var, ScalarInputVariable):
            input_dict[var.name] = var.value_range[0] + torch.rand(size=(n_samples,)) * (
                        var.value_range[1] - var.value_range[0])
        else:
            torch.tensor(var.default, **self._tkwargs).repeat((n_samples, 1))
    return input_dict

to(device)

Updates the device for the model, transformers and default values.

Parameters:

Name	Type	Description	Default
device	Union[device, str]	Device on which the model will be evaluated.	required

Source code in lume_model/models/torch_model.py

def to(self, device: Union[torch.device, str]):
    """Updates the device for the model, transformers and default values.

    Args:
        device: Device on which the model will be evaluated.
    """
    self.model.to(device)
    for t in self.input_transformers + self.output_transformers:
        if isinstance(t, torch.nn.Module):
            t.to(device)
    self.device = device

update_input_variables_to_transformer(transformer_loc)

Returns input variables updated to the transformer at the given location.

Updated are the value ranges and default of the input variables. This allows, e.g., to add a calibration transformer and to update the input variable specification accordingly.

Parameters:

Name	Type	Description	Default
transformer_loc	int	The location of the input transformer to adjust for.	required

Returns:

Type	Description
list[InputVariable]	The updated input variables.

Source code in lume_model/models/torch_model.py

def update_input_variables_to_transformer(self, transformer_loc: int) -> list[InputVariable]:
    """Returns input variables updated to the transformer at the given location.

    Updated are the value ranges and default of the input variables. This allows, e.g., to add a
    calibration transformer and to update the input variable specification accordingly.

    Args:
        transformer_loc: The location of the input transformer to adjust for.

    Returns:
        The updated input variables.
    """
    x_old = {
        "min": torch.tensor([var.value_range[0] for var in self.input_variables], dtype=self.dtype),
        "max": torch.tensor([var.value_range[1] for var in self.input_variables], dtype=self.dtype),
        "default": torch.tensor([var.default for var in self.input_variables], dtype=self.dtype),
    }
    x_new = {}
    for key in x_old.keys():
        x = x_old[key]
        # compute previous limits at transformer location
        for i in range(transformer_loc):
            x = self.input_transformers[i].transform(x)
        # untransform of transformer to adjust for
        x = self.input_transformers[transformer_loc].untransform(x)
        # backtrack through transformers
        for transformer in self.input_transformers[:transformer_loc][::-1]:
            x = transformer.untransform(x)
        x_new[key] = x
    updated_variables = deepcopy(self.input_variables)
    for i, var in enumerate(updated_variables):
        var.value_range = [x_new["min"][i].item(), x_new["max"][i].item()]
        var.default = x_new["default"][i].item()
    return updated_variables

TorchModule

Bases: Module

Wrapper to allow a LUME TorchModel to be used like a torch.nn.Module.

As the base model within the TorchModel is assumed to be fixed during instantiation, so is the TorchModule.

Source code in lume_model/models/torch_module.py

class TorchModule(torch.nn.Module):
    """Wrapper to allow a LUME TorchModel to be used like a torch.nn.Module.

    As the base model within the TorchModel is assumed to be fixed during instantiation,
    so is the TorchModule.
    """
    def __init__(
        self,
        *args,
        model: TorchModel = None,
        input_order: list[str] = None,
        output_order: list[str] = None,
    ):
        """Initializes TorchModule.

        Args:
            *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
              formatted string or file path.

        Keyword Args:
            model: The TorchModel instance to wrap around. If config is None, this has to be defined.
            input_order: Input names in the order they are passed to the model. If None, the input order of the
              TorchModel is used.
            output_order: Output names in the order they are returned by the model. If None, the output order of
              the TorchModel is used.
        """
        if all(arg is None for arg in [*args, model]):
            raise ValueError("Either a YAML string has to be given or model has to be defined.")
        super().__init__()
        if len(args) == 1:
            if not all(v is None for v in [model, input_order, output_order]):
                raise ValueError("Cannot specify YAML string and keyword arguments for TorchModule init.")
            model_fields = {f"model.{k}": v for k, v in TorchModel.model_fields.items()}
            kwargs = parse_config(args[0], model_fields)
            kwargs["model"] = TorchModel(kwargs["model"])
            self.__init__(**kwargs)
        elif len(args) > 1:
            raise ValueError(
                "Arguments to TorchModule must be either a single YAML string or keyword arguments."
            )
        else:
            self._model = model
            self._input_order = input_order
            self._output_order = output_order
            self.register_module("base_model", self._model.model)
            for i, input_transformer in enumerate(self._model.input_transformers):
                self.register_module(f"input_transformers_{i}", input_transformer)
            for i, output_transformer in enumerate(self._model.output_transformers):
                self.register_module(f"output_transformers_{i}", output_transformer)
            if not model.model.training:  # TorchModel defines train/eval mode
                self.eval()

    @property
    def model(self):
        return self._model

    @property
    def input_order(self):
        if self._input_order is None:
            return self._model.input_names
        else:
            return self._input_order

    @property
    def output_order(self):
        if self._output_order is None:
            return self._model.output_names
        else:
            return self._output_order

    def forward(self, x: torch.Tensor):
        # input shape: [n_batch, n_samples, n_dim]
        x = self._validate_input(x)
        model_input = self._tensor_to_dictionary(x)
        y_model = self.evaluate_model(model_input)
        y_model = self.manipulate_output(y_model)
        # squeeze for use as prior mean in botorch GPs
        y = self._dictionary_to_tensor(y_model).squeeze()
        return y

    def yaml(
            self,
            base_key: str = "",
            file_prefix: str = "",
            save_models: bool = False,
    ) -> str:
        """Serializes the object and returns a YAML formatted string defining the TorchModule instance.

        Args:
            base_key: Base key for serialization.
            file_prefix: Prefix for generated filenames.
            save_models: Determines whether models are saved to file.

        Returns:
            YAML formatted string defining the TorchModule instance.
        """
        d = {}
        for k, v in inspect.signature(TorchModule.__init__).parameters.items():
            if k not in ["self", "args", "model"]:
                d[k] = getattr(self, k)
        output = json.loads(
            json.dumps(recursive_serialize(d, base_key, file_prefix, save_models))
        )
        model_output = json.loads(
            self._model.to_json(
                base_key=base_key,
                file_prefix=file_prefix,
                save_models=save_models,
            )
        )
        output["model"] = model_output
        # create YAML formatted string
        s = yaml.dump({"model_class": self.__class__.__name__} | output,
                      default_flow_style=None, sort_keys=False)
        return s

    def dump(
            self,
            file: Union[str, os.PathLike],
            save_models: bool = True,
            base_key: str = "",
    ):
        """Returns and optionally saves YAML formatted string defining the model.

        Args:
            file: File path to which the YAML formatted string and corresponding files are saved.
            base_key: Base key for serialization.
            save_models: Determines whether models are saved to file.
        """
        file_prefix = os.path.splitext(file)[0]
        with open(file, "w") as f:
            f.write(
                self.yaml(
                    save_models=save_models,
                    base_key=base_key,
                    file_prefix=file_prefix,
                )
            )

    def evaluate_model(self, x: dict[str, torch.Tensor]):
        """Placeholder method to modify model calls."""
        return self._model.evaluate(x)

    def manipulate_output(self, y_model: dict[str, torch.Tensor]):
        """Placeholder method to modify the model output."""
        return y_model

    def _tensor_to_dictionary(self, x: torch.Tensor):
        input_dict = {}
        for idx, input_name in enumerate(self.input_order):
            input_dict[input_name] = x[..., idx].unsqueeze(-1)
        return input_dict

    def _dictionary_to_tensor(self, y_model: dict[str, torch.Tensor]):
        output_tensor = torch.stack(
            [y_model[output_name].unsqueeze(-1) for output_name in self.output_order], dim=-1
        )
        return output_tensor

    @staticmethod
    def _validate_input(x: torch.Tensor) -> torch.Tensor:
        if x.dim() <= 1:
            raise ValueError(
                f"Expected input dim to be at least 2 ([n_samples, n_features]), received: {tuple(x.shape)}"
            )
        else:
            return x

__init__(*args, model=None, input_order=None, output_order=None)

Initializes TorchModule.

Parameters:

Name	Type	Description	Default
*args		Accepts a single argument which is the model configuration as dictionary, YAML or JSON formatted string or file path.	()

Other Parameters:

Name	Type	Description
model	TorchModel	The TorchModel instance to wrap around. If config is None, this has to be defined.
input_order	list[str]	Input names in the order they are passed to the model. If None, the input order of the TorchModel is used.
output_order	list[str]	Output names in the order they are returned by the model. If None, the output order of the TorchModel is used.

Source code in lume_model/models/torch_module.py

def __init__(
    self,
    *args,
    model: TorchModel = None,
    input_order: list[str] = None,
    output_order: list[str] = None,
):
    """Initializes TorchModule.

    Args:
        *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
          formatted string or file path.

    Keyword Args:
        model: The TorchModel instance to wrap around. If config is None, this has to be defined.
        input_order: Input names in the order they are passed to the model. If None, the input order of the
          TorchModel is used.
        output_order: Output names in the order they are returned by the model. If None, the output order of
          the TorchModel is used.
    """
    if all(arg is None for arg in [*args, model]):
        raise ValueError("Either a YAML string has to be given or model has to be defined.")
    super().__init__()
    if len(args) == 1:
        if not all(v is None for v in [model, input_order, output_order]):
            raise ValueError("Cannot specify YAML string and keyword arguments for TorchModule init.")
        model_fields = {f"model.{k}": v for k, v in TorchModel.model_fields.items()}
        kwargs = parse_config(args[0], model_fields)
        kwargs["model"] = TorchModel(kwargs["model"])
        self.__init__(**kwargs)
    elif len(args) > 1:
        raise ValueError(
            "Arguments to TorchModule must be either a single YAML string or keyword arguments."
        )
    else:
        self._model = model
        self._input_order = input_order
        self._output_order = output_order
        self.register_module("base_model", self._model.model)
        for i, input_transformer in enumerate(self._model.input_transformers):
            self.register_module(f"input_transformers_{i}", input_transformer)
        for i, output_transformer in enumerate(self._model.output_transformers):
            self.register_module(f"output_transformers_{i}", output_transformer)
        if not model.model.training:  # TorchModel defines train/eval mode
            self.eval()

dump(file, save_models=True, base_key='')

Returns and optionally saves YAML formatted string defining the model.

Parameters:

Name	Type	Description	Default
file	Union[str, PathLike]	File path to which the YAML formatted string and corresponding files are saved.	required
base_key	str	Base key for serialization.	''
save_models	bool	Determines whether models are saved to file.	True

Source code in lume_model/models/torch_module.py

def dump(
        self,
        file: Union[str, os.PathLike],
        save_models: bool = True,
        base_key: str = "",
):
    """Returns and optionally saves YAML formatted string defining the model.

    Args:
        file: File path to which the YAML formatted string and corresponding files are saved.
        base_key: Base key for serialization.
        save_models: Determines whether models are saved to file.
    """
    file_prefix = os.path.splitext(file)[0]
    with open(file, "w") as f:
        f.write(
            self.yaml(
                save_models=save_models,
                base_key=base_key,
                file_prefix=file_prefix,
            )
        )

evaluate_model(x)

Placeholder method to modify model calls.

Source code in lume_model/models/torch_module.py

def evaluate_model(self, x: dict[str, torch.Tensor]):
    """Placeholder method to modify model calls."""
    return self._model.evaluate(x)

manipulate_output(y_model)

Placeholder method to modify the model output.

Source code in lume_model/models/torch_module.py

def manipulate_output(self, y_model: dict[str, torch.Tensor]):
    """Placeholder method to modify the model output."""
    return y_model

yaml(base_key='', file_prefix='', save_models=False)

Serializes the object and returns a YAML formatted string defining the TorchModule instance.

Parameters:

Name	Type	Description	Default
base_key	str	Base key for serialization.	''
file_prefix	str	Prefix for generated filenames.	''
save_models	bool	Determines whether models are saved to file.	False

Returns:

Type	Description
str	YAML formatted string defining the TorchModule instance.

Source code in lume_model/models/torch_module.py

def yaml(
        self,
        base_key: str = "",
        file_prefix: str = "",
        save_models: bool = False,
) -> str:
    """Serializes the object and returns a YAML formatted string defining the TorchModule instance.

    Args:
        base_key: Base key for serialization.
        file_prefix: Prefix for generated filenames.
        save_models: Determines whether models are saved to file.

    Returns:
        YAML formatted string defining the TorchModule instance.
    """
    d = {}
    for k, v in inspect.signature(TorchModule.__init__).parameters.items():
        if k not in ["self", "args", "model"]:
            d[k] = getattr(self, k)
    output = json.loads(
        json.dumps(recursive_serialize(d, base_key, file_prefix, save_models))
    )
    model_output = json.loads(
        self._model.to_json(
            base_key=base_key,
            file_prefix=file_prefix,
            save_models=save_models,
        )
    )
    output["model"] = model_output
    # create YAML formatted string
    s = yaml.dump({"model_class": self.__class__.__name__} | output,
                  default_flow_style=None, sort_keys=False)
    return s

KerasModel

Bases: LUMEBaseModel

LUME-model class for keras models.

Attributes:

Name	Type	Description
model	Model	The keras base model.
output_format	str	Determines format of outputs: "array", "variable" or "raw".
output_transforms	list[str]	List of strings defining additional transformations applied to the outputs. For now, only "softmax" is supported.

Source code in lume_model/models/keras_model.py

class KerasModel(LUMEBaseModel):
    """LUME-model class for keras models.

    Attributes:
        model: The keras base model.
        output_format: Determines format of outputs: "array", "variable" or "raw".
        output_transforms: List of strings defining additional transformations applied to the outputs. For now,
          only "softmax" is supported.
    """
    model: keras.Model
    output_format: str = "array"
    output_transforms: list[str] = []

    def __init__(
            self,
            *args,
            **kwargs,
    ):
        """Initializes KerasModel.

        Args:
            *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
              formatted string or file path.
            **kwargs: See class attributes.
        """
        super().__init__(*args, **kwargs)

    @field_validator("model", mode="before")
    def validate_keras_model(cls, v):
        if isinstance(v, (str, os.PathLike)):
            if os.path.exists(v):
                v = keras.models.load_model(v)
            else:
                raise OSError(f"File {v} is not found.")
        return v

    @field_validator("output_format")
    def validate_output_format(cls, v):
        supported_formats = ["array", "variable", "raw"]
        if v not in supported_formats:
            raise ValueError(f"Unknown output format {v}, expected one of {supported_formats}.")
        return v

    @property
    def dtype(self):
        return np.double

    def evaluate(
            self,
            input_dict: dict[str, Union[InputVariable, float, np.ndarray]],
    ) -> dict[str, Union[OutputVariable, float, np.ndarray]]:
        """Evaluates model on the given input dictionary.

        Args:
            input_dict: Input dictionary on which to evaluate the model.

        Returns:
            Dictionary of output variable names to values.
        """
        formatted_inputs = self._format_inputs(input_dict)
        complete_input_dict = self._complete_inputs(formatted_inputs)
        output_array = self.model.predict(complete_input_dict).astype(self.dtype)
        output_array = self._output_transform(output_array)
        parsed_outputs = self._parse_outputs(output_array)
        output_dict = self._prepare_outputs(parsed_outputs)
        return output_dict

    def random_input(self, n_samples: int = 1) -> dict[str, np.ndarray]:
        """Generates random input(s) for the model.

        Args:
            n_samples: Number of random samples to generate.

        Returns:
            Dictionary of input variable names to arrays.
        """
        input_dict = {}
        for var in self.input_variables:
            if isinstance(var, ScalarInputVariable):
                input_dict[var.name] = np.random.uniform(*var.value_range, size=n_samples)
            else:
                default_array = np.array(var.default, dtype=self.dtype)
                input_dict[var.name] = np.repeat(default_array.reshape((1, *default_array.shape)),
                                                 n_samples, axis=0)
        return input_dict

    def random_evaluate(self, n_samples: int = 1) -> dict[str, Union[OutputVariable, float, np.ndarray]]:
        """Returns random evaluation(s) of the model.

        Args:
            n_samples: Number of random samples to evaluate.

        Returns:
            Dictionary of variable names to outputs.
        """
        random_input = self.random_input(n_samples)
        return self.evaluate(random_input)

    def _format_inputs(
            self,
            input_dict: dict[str, Union[InputVariable, float, np.ndarray]],
    ) -> dict[str, np.ndarray]:
        """Formats values of the input dictionary as arrays.

        Args:
            input_dict: Dictionary of input variable names to values.

        Returns:
            Dictionary of input variable names to arrays.
        """
        # NOTE: The input variable is only updated if a singular value is given (ambiguous otherwise)
        formatted_inputs = {}
        for var_name, var in input_dict.items():
            if isinstance(var, InputVariable):
                formatted_inputs[var_name] = np.array(var.value, dtype=self.dtype)
                # self.input_variables[self.input_names.index(var_name)].value = var.value
            elif isinstance(var, float):
                formatted_inputs[var_name] = np.array(var, dtype=self.dtype)
                # self.input_variables[self.input_names.index(var_name)].value = var
            elif isinstance(var, np.ndarray):
                var = var.astype(self.dtype).squeeze()
                formatted_inputs[var_name] = var
                # if var.ndim == 0:
                #     self.input_variables[self.input_names.index(var_name)].value = var.item()
            else:
                TypeError(
                    f"Unknown type {type(var)} passed to evaluate."
                    f"Should be one of InputVariable, float or np.ndarray."
                )
        return formatted_inputs

    def _complete_inputs(self, formatted_inputs: dict[str, np.ndarray]) -> dict[str, np.ndarray]:
        """Completes input dictionary by filling in default values.

        Args:
            formatted_inputs: Dictionary of input variable names to arrays.

        Returns:
            Completed input dictionary to be passed to the model.
        """
        # determine input shape
        input_shapes = [formatted_inputs[k].shape for k in formatted_inputs.keys()]
        if not all(ele == input_shapes[0] for ele in input_shapes):
            raise ValueError("Inputs have inconsistent shapes.")

        for i, key in enumerate(self.input_names):
            if key not in formatted_inputs.keys():
                default_array = np.array(self.input_variables[i].default, dtype=self.dtype)
                formatted_inputs[key] = np.tile(default_array, reps=input_shapes[0])

        if not input_shapes[0]:
            for key in self.input_names:
                formatted_inputs[key] = formatted_inputs[key].reshape((1, *formatted_inputs[key].shape))
        return formatted_inputs

    def _output_transform(self, output_array: np.ndarray) -> np.ndarray:
        """Applies additional transformations to the model output array.

        Args:
            output_array: Output array from the model.

        Returns:
            Transformed output array.
        """
        if "softmax" in self.output_transforms:
            output_array = np.argmax(output_array, axis=-1)
        return output_array

    def _parse_outputs(self, output_array: np.ndarray) -> dict[str, np.ndarray]:
        """Constructs dictionary from model output array.

        Args:
            output_array: Transformed output array from the model.

        Returns:
            Dictionary of output variable names to transformed arrays.
        """
        parsed_outputs = {}
        if output_array.ndim in [0, 1]:
            output_array = output_array.reshape((1, *output_array.shape))
        if len(self.output_names) == 1:
            parsed_outputs[self.output_names[0]] = output_array.squeeze()
        else:
            for idx, output_name in enumerate(self.output_names):
                parsed_outputs[output_name] = output_array[..., idx].squeeze()
        return parsed_outputs

    def _prepare_outputs(
            self,
            parsed_outputs: dict[str, np.ndarray],
    ) -> dict[str, Union[OutputVariable, np.ndarray]]:
        """Updates and returns outputs according to output_format.

        Updates the output variables within the model to reflect the new values.

        Args:
            parsed_outputs: Dictionary of output variable names to transformed arrays.

        Returns:
            Dictionary of output variable names to values depending on output_format.
        """
        # for var in self.output_variables:
        #     if parsed_outputs[var.name].ndim == 0:
        #         idx = self.output_names.index(var.name)
        #         if isinstance(var, ScalarOutputVariable):
        #             self.output_variables[idx].value = parsed_outputs[var.name].item()
        #         elif isinstance(var, ImageOutputVariable):
        #             # OutputVariables should be arrays
        #             self.output_variables[idx].value = (parsed_outputs[var.name].reshape(var.shape).numpy())
        #             self._update_image_limits(var, parsed_outputs)

        if self.output_format == "array":
            return parsed_outputs
        elif self.output_format == "variable":
            output_dict = {var.name: var for var in self.output_variables}
            for var in output_dict.values():
                var.value = parsed_outputs[var.name].item()
            return output_dict
            # return {var.name: var for var in self.output_variables}
        else:
            return {key: value.item() if value.squeeze().ndim == 0 else value
                    for key, value in parsed_outputs.items()}
            # return {var.name: var.value for var in self.output_variables}

    def _update_image_limits(
            self,
            variable: OutputVariable, predicted_output: dict[str, np.ndarray],
    ):
        output_idx = self.output_names.index(variable.name)
        if self.output_variables[output_idx].x_min_variable:
            self.output_variables[output_idx].x_min = predicted_output[
                self.output_variables[output_idx].x_min_variable
            ].item()

        if self.output_variables[output_idx].x_max_variable:
            self.output_variables[output_idx].x_max = predicted_output[
                self.output_variables[output_idx].x_max_variable
            ].item()

        if self.output_variables[output_idx].y_min_variable:
            self.output_variables[output_idx].y_min = predicted_output[
                self.output_variables[output_idx].y_min_variable
            ].item()

        if self.output_variables[output_idx].y_max_variable:
            self.output_variables[output_idx].y_max = predicted_output[
                self.output_variables[output_idx].y_max_variable
            ].item()

__init__(*args, **kwargs)

Initializes KerasModel.

Parameters:

Name	Type	Description	Default
*args		Accepts a single argument which is the model configuration as dictionary, YAML or JSON formatted string or file path.	()
**kwargs		See class attributes.	{}

Source code in lume_model/models/keras_model.py

def __init__(
        self,
        *args,
        **kwargs,
):
    """Initializes KerasModel.

    Args:
        *args: Accepts a single argument which is the model configuration as dictionary, YAML or JSON
          formatted string or file path.
        **kwargs: See class attributes.
    """
    super().__init__(*args, **kwargs)

evaluate(input_dict)

Evaluates model on the given input dictionary.

Parameters:

Name	Type	Description	Default
input_dict	dict[str, Union[InputVariable, float, ndarray]]	Input dictionary on which to evaluate the model.	required

Returns:

Type	Description
dict[str, Union[OutputVariable, float, ndarray]]	Dictionary of output variable names to values.

Source code in lume_model/models/keras_model.py

def evaluate(
        self,
        input_dict: dict[str, Union[InputVariable, float, np.ndarray]],
) -> dict[str, Union[OutputVariable, float, np.ndarray]]:
    """Evaluates model on the given input dictionary.

    Args:
        input_dict: Input dictionary on which to evaluate the model.

    Returns:
        Dictionary of output variable names to values.
    """
    formatted_inputs = self._format_inputs(input_dict)
    complete_input_dict = self._complete_inputs(formatted_inputs)
    output_array = self.model.predict(complete_input_dict).astype(self.dtype)
    output_array = self._output_transform(output_array)
    parsed_outputs = self._parse_outputs(output_array)
    output_dict = self._prepare_outputs(parsed_outputs)
    return output_dict

random_evaluate(n_samples=1)

Returns random evaluation(s) of the model.

Parameters:

Name	Type	Description	Default
n_samples	int	Number of random samples to evaluate.	1

Returns:

Type	Description
dict[str, Union[OutputVariable, float, ndarray]]	Dictionary of variable names to outputs.

Source code in lume_model/models/keras_model.py

def random_evaluate(self, n_samples: int = 1) -> dict[str, Union[OutputVariable, float, np.ndarray]]:
    """Returns random evaluation(s) of the model.

    Args:
        n_samples: Number of random samples to evaluate.

    Returns:
        Dictionary of variable names to outputs.
    """
    random_input = self.random_input(n_samples)
    return self.evaluate(random_input)

random_input(n_samples=1)

Generates random input(s) for the model.

Parameters:

Name	Type	Description	Default
n_samples	int	Number of random samples to generate.	1

Returns:

Type	Description
dict[str, ndarray]	Dictionary of input variable names to arrays.

Source code in lume_model/models/keras_model.py

def random_input(self, n_samples: int = 1) -> dict[str, np.ndarray]:
    """Generates random input(s) for the model.

    Args:
        n_samples: Number of random samples to generate.

    Returns:
        Dictionary of input variable names to arrays.
    """
    input_dict = {}
    for var in self.input_variables:
        if isinstance(var, ScalarInputVariable):
            input_dict[var.name] = np.random.uniform(*var.value_range, size=n_samples)
        else:
            default_array = np.array(var.default, dtype=self.dtype)
            input_dict[var.name] = np.repeat(default_array.reshape((1, *default_array.shape)),
                                             n_samples, axis=0)
    return input_dict