reflectorch 1.5.1__py3-none-any.whl

reflectorch/data_generation/priors/parametric_subpriors.py

@@ -0,0 +1,369 @@
+from typing import Tuple, Dict, Type, List
+
+import torch
+from torch import Tensor
+
+from reflectorch.data_generation.priors.base import PriorSampler
+from reflectorch.data_generation.priors.params import AbstractParams
+from reflectorch.data_generation.priors.no_constraints import (
+    DEFAULT_DEVICE,
+    DEFAULT_DTYPE,
+)
+
+from reflectorch.data_generation.priors.parametric_models import (
+    MULTILAYER_MODELS,
+    NuisanceParamsWrapper,
+    ParametricModel,
+)
+from reflectorch.data_generation.priors.scaler_mixin import ScalerMixin
+
+
+class BasicParams(AbstractParams):
+    """Parameter class compatible with different parameterizations of the SLD profile. It stores the parameters as well as their minimum and maximum subprior bounds.
+
+        Args:
+            parameters (Tensor): the values of the thin film parameters
+            min_bounds (Tensor): the minimum subprior bounds of the parameters
+            max_bounds (Tensor): the maximum subprior bounds of the parameters
+            max_num_layers (int, optional): the maximum number of layers (for box model parameterizations it is the number of layers). Defaults to None.
+            param_model (ParametricModel, optional): the parametric model. Defaults to the box model parameterization with number of layers given by max_num_layers.
+        """
+    
+    __slots__ = (
+        'parameters',
+        'min_bounds',
+        'max_bounds',
+        'max_num_layers',
+        'param_model',
+    )
+    PARAM_NAMES = __slots__
+    PARAM_MODEL_CLS: Type[ParametricModel]
+    MAX_NUM_LAYERS: int = 30
+
+    def __init__(self,
+                 parameters: Tensor,
+                 min_bounds: Tensor,
+                 max_bounds: Tensor,
+                 max_num_layers: int = None,
+                 param_model: ParametricModel = None,
+                 ):
+
+        max_num_layers = max_num_layers or self.MAX_NUM_LAYERS
+        self.param_model = param_model or self.PARAM_MODEL_CLS(max_num_layers)
+        self.max_num_layers = max_num_layers
+        self.parameters = parameters
+        self.min_bounds = min_bounds
+        self.max_bounds = max_bounds
+
+    def get_param_labels(self, **kwargs) -> List[str]:
+        """gets the parameter labels"""
+        return self.param_model.get_param_labels(**kwargs)
+
+    def reflectivity(self, q: Tensor, log: bool = False, **kwargs):
+        r"""computes the reflectivity curves directly from the parameters
+
+        Args:
+            q (Tensor): the q values
+            log (bool, optional): whether to apply logarithm to the curves. Defaults to False.
+
+        Returns:
+            Tensor: the simulated reflectivity curves
+        """
+        return self.param_model.reflectivity(q, self.parameters, log=log, **kwargs)
+
+    @property
+    def max_layer_num(self) -> int:  # keep for back compatibility but TODO: unify api among different params
+        """gets the maximum number of layers"""
+        return self.max_num_layers
+
+    @property
+    def num_params(self) -> int:
+        """get the number of parameters (parameter dimensionality)"""
+        return self.param_model.param_dim
+
+    @property
+    def thicknesses(self):
+        """gets the thicknesses"""
+        params = self.param_model.to_standard_params(self.parameters)
+        return params['thickness']
+
+    @property
+    def roughnesses(self):
+        """gets the roughnesses"""
+        params = self.param_model.to_standard_params(self.parameters)
+        return params['roughness']
+
+    @property
+    def slds(self):
+        """gets the slds"""
+        params = self.param_model.to_standard_params(self.parameters)
+        return params['sld']
+        
+    @property
+    def real_slds(self):
+        """gets the real part of the slds"""
+        params = self.param_model.to_standard_params(self.parameters)
+        return params['sld'].real
+    
+    @property
+    def imag_slds(self):
+        """gets the imaginary part of the slds (only for complex dtypes)"""
+        params = self.param_model.to_standard_params(self.parameters)
+        return params['sld'].imag
+
+    @staticmethod
+    def rearrange_context_from_params(
+            scaled_params: Tensor,
+            context: Tensor,
+            inference: bool = False,
+            from_params: bool = False,
+    ):
+        if inference:
+            if from_params:
+                num_params = scaled_params.shape[-1] // 3
+                scaled_params = scaled_params[:, num_params:]
+            context = torch.cat([context, scaled_params], dim=-1)
+            return context
+
+        num_params = scaled_params.shape[-1] // 3
+        assert num_params * 3 == scaled_params.shape[-1]
+        scaled_params, bound_context = torch.split(scaled_params, [num_params, 2 * num_params], dim=-1)
+        context = torch.cat([context, bound_context], dim=-1)
+        return scaled_params, context
+
+    @staticmethod
+    def restore_params_from_context(scaled_params: Tensor, context: Tensor):
+        num_params = scaled_params.shape[-1]
+        scaled_bounds = context[:, -2 * num_params:]
+        scaled_params = torch.cat([scaled_params, scaled_bounds], dim=-1)
+        return scaled_params
+
+    def as_tensor(self, add_bounds: bool = True, **kwargs) -> Tensor:
+        """converts the instance of the class to a Pytorch tensor
+
+        Args:
+            add_bounds (bool, optional): whether to add the subprior bounds to the tensor. Defaults to True.
+
+        Returns:
+            Tensor: the Pytorch tensor obtained from the instance of the class
+        """
+        if not add_bounds:
+            return self.parameters
+        return torch.cat([self.parameters, self.min_bounds, self.max_bounds], -1)
+
+    @classmethod
+    def from_tensor(cls, params: Tensor, **kwargs):
+        """initializes an instance of the class from a Pytorch tensor
+
+        Args:
+            params (Tensor): Pytorch tensor containing the parameter values, min subprior bounds and max subprior bounds
+
+        Returns:
+            BasicParams: the instance of the class
+        """
+        num_params = params.shape[-1] // 3
+
+        params, min_bounds, max_bounds = torch.split(
+            params, [num_params, num_params, num_params], dim=-1
+        )
+
+        return cls(
+            params,
+            min_bounds,
+            max_bounds,
+            **kwargs
+        )
+
+    def scale_with_q(self, q_ratio: float):
+        """scales the parameters based on the q ratio
+
+        Args:
+            q_ratio (float): the scaling ratio
+        """
+        self.parameters = self.param_model.scale_with_q(self.parameters, q_ratio)
+        self.min_bounds = self.param_model.scale_with_q(self.min_bounds, q_ratio)
+        self.max_bounds = self.param_model.scale_with_q(self.max_bounds, q_ratio)
+
+
+class SubpriorParametricSampler(PriorSampler, ScalerMixin):
+    PARAM_CLS = BasicParams
+
+    def __init__(self,
+                 param_ranges: Dict[str, Tuple[float, float]],
+                 bound_width_ranges: Dict[str, Tuple[float, float]],
+                 model_name: str,
+                 device: torch.device = DEFAULT_DEVICE,
+                 dtype: torch.dtype = DEFAULT_DTYPE,
+                 max_num_layers: int = 50,
+                 logdist: bool = False,
+                 scale_params_by_ranges = False,
+                 scaled_range: Tuple[float, float] = (-1., 1.),
+                 **kwargs
+                 ):
+        """Prior sampler for the parameters of a parametric model and their subprior bounds
+
+        Args:
+            param_ranges (Dict[str, Tuple[float, float]]): dictionary containing the name of each type of parameter together with its range
+            bound_width_ranges (Dict[str, Tuple[float, float]]): dictionary containing the name of each type of parameter together with the range for sampling the widths of the subprior interval
+            model_name (str): the name of the parametric model
+            device (torch.device, optional): the Pytorch device. Defaults to DEFAULT_DEVICE.
+            dtype (torch.dtype, optional): the Pytorch data type. Defaults to DEFAULT_DTYPE.
+            max_num_layers (int, optional): the maximum number of layers (for box model parameterizations it is the number of layers). Defaults to 50.
+            logdist (bool, optional): if True the relative widths of the subprior intervals are sampled uniformly on a logarithmic scale instead of uniformly. Defaults to False.
+            scale_params_by_ranges (bool, optional): if True the parameters are scaled with respect to their ranges instead of being scaled with respect to their prior bounds. Defaults to False.
+            scaled_range (Tuple[float, float], optional): the range for scaling the parameters. Defaults to (-1., 1.)
+        """
+        self.scaled_range = scaled_range
+
+        self.shift_param_config = kwargs.pop('shift_param_config', {})
+
+        base_model: ParametricModel = MULTILAYER_MODELS[model_name](max_num_layers, logdist=logdist, **kwargs)
+        if any(self.shift_param_config.values()):
+            self.param_model = NuisanceParamsWrapper(
+                base_model=base_model,
+                nuisance_params_config=self.shift_param_config,
+                **kwargs,
+            )
+        else:
+            self.param_model = base_model
+
+        self.device = device
+        self.dtype = dtype
+        self.num_layers = max_num_layers
+
+        self.PARAM_CLS.PARAM_MODEL_CLS = MULTILAYER_MODELS[model_name]
+        self.PARAM_CLS.MAX_NUM_LAYERS = max_num_layers
+
+        self._param_dim = self.param_model.param_dim
+        self.min_bounds, self.max_bounds, self.min_delta, self.max_delta = self.param_model.init_bounds(
+            param_ranges, bound_width_ranges, device=device, dtype=dtype
+        )
+        
+        self.param_ranges = param_ranges
+        self.bound_width_ranges = bound_width_ranges
+        self.model_name = model_name
+        self.logdist = logdist
+        self.scale_params_by_ranges = scale_params_by_ranges
+
+    @property
+    def max_num_layers(self) -> int:
+        """gets the maximum number of layers"""
+        return self.num_layers
+
+    @property
+    def param_dim(self) -> int:
+        """get the number of parameters (parameter dimensionality)"""
+        return self._param_dim
+
+    def sample(self, batch_size: int) -> BasicParams:
+        """sample a batch of parameters
+
+        Args:
+            batch_size (int): the batch size
+
+        Returns:
+            BasicParams: sampled parameters
+        """
+        params, min_bounds, max_bounds = self.param_model.sample(
+            batch_size, self.min_bounds, self.max_bounds, self.min_delta, self.max_delta
+        )
+
+        params = BasicParams(
+            parameters=params,
+            min_bounds=min_bounds,
+            max_bounds=max_bounds,
+            max_num_layers=self.max_num_layers,
+            param_model=self.param_model,
+        )
+
+        return params
+
+    def scale_params(self, params: BasicParams) -> Tensor:
+        """scale the parameters to a ML-friendly range
+
+        Args:
+            params (BasicParams): the parameters to be scaled
+
+        Returns:
+            Tensor: the scaled parameters
+        """
+        if self.scale_params_by_ranges:
+            scaled_params = torch.cat([
+                self._scale(params.parameters, self.min_bounds, self.max_bounds), #parameters and subprior bounds are scaled with respect to the parameter ranges
+                self._scale(params.min_bounds, self.min_bounds, self.max_bounds), 
+                self._scale(params.max_bounds, self.min_bounds, self.max_bounds),
+            ], -1)
+            return scaled_params
+        else:
+            scaled_params = torch.cat([
+                self._scale(params.parameters, params.min_bounds, params.max_bounds), #each parameter scaled with respect to its subprior bounds
+                self._scale(params.min_bounds, self.min_bounds, self.max_bounds), #the subprior bounds are scaled with respect to the parameter ranges
+                self._scale(params.max_bounds, self.min_bounds, self.max_bounds),
+            ], -1)
+            return scaled_params
+
+    def restore_params(self, scaled_params: Tensor) -> BasicParams:
+        """restore the parameters to their original range
+
+        Args:
+            scaled_params (Tensor): the scaled parameters
+
+        Returns:
+            BasicParams: the parameters restored to their original range
+        """
+        num_params = scaled_params.shape[-1] // 3
+        scaled_params, scaled_min_bounds, scaled_max_bounds = torch.split(
+            scaled_params, num_params, -1
+        )
+        if self.scale_params_by_ranges:
+            min_bounds = self._restore(scaled_min_bounds, self.min_bounds, self.max_bounds)
+            max_bounds = self._restore(scaled_max_bounds, self.min_bounds, self.max_bounds)
+            params = self._restore(scaled_params, self.min_bounds, self.max_bounds)
+        else:
+            min_bounds = self._restore(scaled_min_bounds, self.min_bounds, self.max_bounds)
+            max_bounds = self._restore(scaled_max_bounds, self.min_bounds, self.max_bounds)
+            params = self._restore(scaled_params, min_bounds, max_bounds)
+
+        return BasicParams(
+            parameters=params,
+            min_bounds=min_bounds,
+            max_bounds=max_bounds,
+            max_num_layers=self.max_num_layers,
+            param_model=self.param_model,
+        )
+    
+    def scale_bounds(self, bounds: Tensor) -> Tensor:
+        return self._scale(bounds, self.min_bounds, self.max_bounds)
+
+    def log_prob(self, params: BasicParams) -> Tensor:
+        log_prob = torch.zeros(params.batch_size, device=self.device, dtype=self.dtype)
+        log_prob[~self.get_indices_within_bounds(params)] = -float('inf')
+        return log_prob
+
+    def get_indices_within_domain(self, params: BasicParams) -> Tensor:
+        return self.get_indices_within_bounds(params)
+
+    def get_indices_within_bounds(self, params: BasicParams) -> Tensor:
+        return (
+                torch.all(params.parameters >= params.min_bounds, -1) &
+                torch.all(params.parameters <= params.max_bounds, -1)
+        )
+
+    def filter_params(self, params: BasicParams) -> BasicParams:
+        indices = self.get_indices_within_domain(params)
+        return params[indices]
+
+    def clamp_params(
+            self, params: BasicParams, inplace: bool = False
+    ) -> BasicParams:
+        if inplace:
+            params.parameters = torch.clamp_(params.parameters, params.min_bounds, params.max_bounds)
+            return params
+
+        return BasicParams(
+            parameters=torch.clamp(params.parameters, params.min_bounds, params.max_bounds),
+            min_bounds=params.min_bounds.clone(),
+            max_bounds=params.max_bounds.clone(),
+            max_num_layers=self.max_num_layers,
+            param_model=self.param_model,
+        )

reflectorch/data_generation/priors/params.py

@@ -0,0 +1,252 @@
+from typing import List, Tuple
+
+import torch
+from torch import Tensor
+
+from reflectorch.data_generation.utils import get_d_rhos, get_param_labels
+from reflectorch.data_generation.reflectivity import reflectivity
+
+__all__ = [
+    "Params",
+    "AbstractParams",
+]
+
+
+class AbstractParams(object):
+    """Base class for parameters"""
+    PARAM_NAMES: Tuple[str, ...]
+
+    @staticmethod
+    def rearrange_context_from_params(
+            scaled_params: Tensor, context: Tensor, inference: bool = False, from_params: bool = False
+    ):
+        if inference:
+            return context
+        return scaled_params, context
+
+    @staticmethod
+    def restore_params_from_context(scaled_params: Tensor, context: Tensor):
+        return scaled_params
+
+    def reflectivity(self, q: Tensor, log: bool = False, **kwargs):
+        """computes the reflectivity curves directly from the parameters"""
+        raise NotImplementedError
+
+    def __iter__(self):
+        for name in self.PARAM_NAMES:
+            yield getattr(self, name)
+
+    def to_(self, tgt):
+        for name, arr in zip(self.PARAM_NAMES, self):
+            setattr(self, name, _to(arr, tgt))
+
+    def to(self, tgt):
+        """performs Pytorch Tensor dtype and/or device conversion"""
+        return self.__class__(*[_to(arr, tgt) for arr in self])
+
+    def cuda(self):
+        """moves the parameters to the GPU"""
+        return self.to('cuda')
+
+    def cpu(self):
+        """moves the parameters to the CPU"""
+        return self.to('cpu')
+
+    def __getitem__(self, item) -> 'AbstractParams':
+        return self.__class__(*[
+            arr.__getitem__(item) if isinstance(arr, Tensor) else arr for arr in self
+        ])
+
+    def __setitem__(self, key, other):
+        if not isinstance(other, AbstractParams):
+            raise ValueError
+
+        for param, other_param in zip(self, other):
+            if isinstance(param, Tensor):
+                param[key] = other_param
+
+    def __add__(self, other):
+        if not isinstance(other, AbstractParams):
+            raise NotImplemented
+
+        return self.__class__(*[
+            torch.cat([param, other_param], 0)
+            if isinstance(param, Tensor) else param
+            for param, other_param in zip(self, other)
+        ])
+
+    def __eq__(self, other):
+        if not isinstance(other, AbstractParams):
+            raise NotImplemented
+
+        return all([torch.allclose(param, other_param) for param, other_param in zip(self, other)])
+
+    @classmethod
+    def cat(cls, *params: 'AbstractParams'):
+        return cls(
+            *[torch.cat([getattr(p, name) for p in params], 0)
+              if isinstance(getattr(params[0], name), Tensor) else
+              getattr(params[0], name)
+              for name in cls.PARAM_NAMES]
+        )
+
+    @property
+    def _ref_tensor(self):
+        return getattr(self, self.PARAM_NAMES[0])
+
+    def scale_with_q(self, q_ratio: float):
+        raise NotImplementedError
+
+    @property
+    def max_layer_num(self) -> int:
+        """gets the number of layers"""
+        return self._ref_tensor.shape[-1]
+
+    @property
+    def batch_size(self) -> int:
+        """gets the batch size"""
+        return self._ref_tensor.shape[0]
+
+    @property
+    def device(self):
+        """gets the Pytorch device"""
+        return self._ref_tensor.device
+
+    @property
+    def dtype(self):
+        """gets the Pytorch data type"""
+        return self._ref_tensor.dtype
+
+    @property
+    def d_rhos(self):
+        raise NotImplementedError
+
+    def as_tensor(self, use_drho: bool = False) -> Tensor:
+        """converts the instance of the class to a Pytorch tensor"""
+        raise NotImplementedError
+
+    @classmethod
+    def from_tensor(cls, params: Tensor):
+        """initializes an instance of the class from a Pytorch tensor"""
+        raise NotImplementedError
+
+    @property
+    def num_params(self) -> int:
+        """get the number of parameters"""
+        return self.layers_num2size(self.max_layer_num)
+
+    @staticmethod
+    def size2layers_num(size: int) -> int:
+        """converts the number of parameters to the number of layers"""
+        raise NotImplementedError
+
+    @staticmethod
+    def layers_num2size(layers_num: int) -> int:
+        """converts the number of layers to the number of parameters"""
+        raise NotImplementedError
+
+    def get_param_labels(self) -> List[str]:
+        """gets the parameter labels"""
+        raise NotImplementedError
+
+    def __repr__(self):
+        return f'{self.__class__.__name__}(' \
+               f'batch_size={self.batch_size}, ' \
+               f'max_layer_num={self.max_layer_num}, ' \
+               f'device={str(self.device)})'
+
+
+def _to(arr, dest):
+    if hasattr(arr, 'to'):
+        arr = arr.to(dest)
+    return arr
+
+
+class Params(AbstractParams):
+    """Parameter class for thickness, roughness and sld parameters
+
+        Args:
+            thicknesses (Tensor): batch of thicknesses (top to bottom)
+            roughnesses (Tensor): batch of roughnesses (top to bottom)
+            slds (Tensor): batch of slds (top to bottom)
+        """
+    MIN_THICKNESS: float = 0.5
+
+    __slots__ = ('thicknesses', 'roughnesses', 'slds')
+    PARAM_NAMES = __slots__
+
+    def __init__(self, thicknesses: Tensor, roughnesses: Tensor, slds: Tensor):
+
+        self.thicknesses = thicknesses
+        self.roughnesses = roughnesses
+        self.slds = slds
+
+    @staticmethod
+    def rearrange_context_from_params(
+            scaled_params: Tensor, context: Tensor, inference: bool = False, from_params: bool = False
+    ):
+        if inference:
+            return context
+        return scaled_params, context
+
+    @staticmethod
+    def restore_params_from_context(scaled_params: Tensor, context: Tensor):
+        return scaled_params
+
+    def reflectivity(self, q: Tensor, log: bool = False, **kwargs):
+        """computes the reflectivity curves directly from the parameters
+
+        Args:
+            q (Tensor): the q values
+            log (bool, optional): whether to apply logarithm to the curves. Defaults to False.
+
+        Returns:
+            Tensor: the simulated reflectivity curves
+        """
+        return reflectivity(q, self.thicknesses, self.roughnesses, self.slds, log=log, **kwargs)
+
+    def scale_with_q(self, q_ratio: float):
+        """scales the parameters based on the q ratio
+
+        Args:
+            q_ratio (float): the scaling ratio
+        """
+        self.thicknesses /= q_ratio
+        self.roughnesses /= q_ratio
+        self.slds *= q_ratio ** 2
+
+    @property
+    def d_rhos(self):
+        """computes the differences in SLD values of the neighboring layers"""
+        return get_d_rhos(self.slds)
+
+    def as_tensor(self, use_drho: bool = False) -> Tensor:
+        """converts the instance of the class to a Pytorch tensor"""
+        if use_drho:
+            return torch.cat([self.thicknesses, self.roughnesses, self.d_rhos], -1)
+        else:
+            return torch.cat([self.thicknesses, self.roughnesses, self.slds], -1)
+
+    @classmethod
+    def from_tensor(cls, params: Tensor):
+        """initializes an instance of the class from a Pytorch tensor containing the values of the parameters"""
+        layers_num = (params.shape[-1] - 2) // 3
+
+        thicknesses, roughnesses, slds = torch.split(params, [layers_num, layers_num + 1, layers_num + 1], dim=-1)
+
+        return cls(thicknesses, roughnesses, slds)
+
+    @staticmethod
+    def size2layers_num(size: int) -> int:
+        """converts the number of parameters to the number of layers"""
+        return (size - 2) // 3
+
+    @staticmethod
+    def layers_num2size(layers_num: int) -> int:
+        """converts the number of layers to the number of parameters"""
+        return layers_num * 3 + 2
+
+    def get_param_labels(self, **kwargs) -> List[str]:
+        """gets the parameter labels, the layers are numbered from the bottom to the top 
+           (i.e. opposite to the order in the Tensors)"""
+        return get_param_labels(self.max_layer_num, **kwargs)