reflectorch 1.0.0__py3-none-any.whl

reflectorch/__init__.py

@@ -0,0 +1,23 @@
+# -*- coding: utf-8 -*-
+#
+#
+# This source code is licensed under the GPL license found in the
+# LICENSE file in the root directory of this source tree.
+
+from reflectorch.data_generation import *
+from reflectorch.ml import *
+from reflectorch.models import *
+from reflectorch.utils import *
+from reflectorch.paths import *
+from reflectorch.runs import *
+from reflectorch.inference import *
+
+from reflectorch.data_generation import __all__ as all_data_generation
+from reflectorch.ml import __all__ as all_ml
+from reflectorch.models import __all__ as all_models
+from reflectorch.utils import __all__ as all_utils
+from reflectorch.paths import __all__ as all_paths
+from reflectorch.runs import __all__ as all_runs
+from reflectorch.inference import __all__ as all_inference
+
+__all__ = all_data_generation + all_ml + all_models + all_utils + all_paths + all_runs + all_inference

reflectorch/data_generation/__init__.py

@@ -0,0 +1,130 @@
+# -*- coding: utf-8 -*-
+#
+#
+# This source code is licensed under the GPL license found in the
+# LICENSE file in the root directory of this source tree.
+
+from reflectorch.data_generation.dataset import BasicDataset, BATCH_DATA_TYPE
+from reflectorch.data_generation.priors import (
+    Params,
+    PriorSampler,
+    BasicPriorSampler,
+    SingleParamPrior,
+    SimplePriorSampler,
+    UniformParamPrior,
+    GaussianParamPrior,
+    TruncatedGaussianParamPrior,
+    UniformSubPriorParams,
+    UniformSubPriorSampler,
+    NarrowSldUniformSubPriorSampler,
+    ExpUniformSubPriorSampler,
+    SimpleMultilayerSampler,
+    SubpriorParametricSampler,
+    BasicParams,
+    ParametricModel,
+    MULTILAYER_MODELS,
+    SamplerStrategy,
+    BasicSamplerStrategy,
+    ConstrainedRoughnessSamplerStrategy,
+    ConstrainedRoughnessAndImgSldSamplerStrategy,
+)
+from reflectorch.data_generation.process_data import ProcessData, ProcessPipeline
+from reflectorch.data_generation.q_generator import (
+    QGenerator,
+    ConstantAngle,
+    ConstantQ,
+    VariableQ,
+    EquidistantQ,
+)
+from reflectorch.data_generation.noise import (
+    QNoiseGenerator,
+    IntensityNoiseGenerator,
+    QNormalNoiseGenerator,
+    QSystematicShiftGenerator,
+    PoissonNoiseGenerator,
+    MultiplicativeLogNormalNoiseGenerator,
+    ShiftNoise,
+    ScalingNoise,
+    BackgroundNoise,
+    BasicExpIntensityNoise,
+    BasicQNoiseGenerator,
+)
+from reflectorch.data_generation.scale_curves import (
+    CurvesScaler,
+    LogAffineCurvesScaler,
+    MeanNormalizationCurvesScaler,
+)
+from reflectorch.data_generation.utils import (
+    get_reversed_params,
+    get_density_profiles,
+    uniform_sampler,
+    logdist_sampler,
+    triangular_sampler,
+    get_param_labels,
+)
+
+from reflectorch.data_generation.smearing import Smearing
+
+from reflectorch.data_generation.reflectivity import reflectivity
+
+from reflectorch.data_generation.likelihoods import (
+    LogLikelihood,
+    PoissonLogLikelihood,
+)
+
+__all__ = [
+    "Params",
+    "PriorSampler",
+    "BasicPriorSampler",
+    "BasicDataset",
+    "ProcessData",
+    "ProcessPipeline",
+    "QGenerator",
+    "ConstantQ",
+    "VariableQ",
+    "EquidistantQ",
+    "QNoiseGenerator",
+    "IntensityNoiseGenerator",
+    "MultiplicativeLogNormalNoiseGenerator",
+    "PoissonNoiseGenerator",
+    "CurvesScaler",
+    "ShiftNoise",
+    "ScalingNoise",
+    "BackgroundNoise",
+    "QNormalNoiseGenerator",
+    "QSystematicShiftGenerator",
+    "LogAffineCurvesScaler",
+    "MeanNormalizationCurvesScaler",
+    "get_reversed_params",
+    "get_density_profiles",
+    "logdist_sampler",
+    "uniform_sampler",
+    "triangular_sampler",
+    "get_param_labels",
+    "reflectivity",
+    "Smearing",
+    "SingleParamPrior",
+    "SimplePriorSampler",
+    "UniformParamPrior",
+    "GaussianParamPrior",
+    "TruncatedGaussianParamPrior",
+    "UniformSubPriorParams",
+    "UniformSubPriorSampler",
+    "NarrowSldUniformSubPriorSampler",
+    "ExpUniformSubPriorSampler",
+    "SimpleMultilayerSampler",
+    "BATCH_DATA_TYPE",
+    "LogLikelihood",
+    "PoissonLogLikelihood",
+    "BasicExpIntensityNoise",
+    "BasicQNoiseGenerator",
+    "ConstantAngle",
+    "SubpriorParametricSampler",
+    "BasicParams",
+    "ParametricModel",
+    "MULTILAYER_MODELS",
+    "SamplerStrategy",
+    "BasicSamplerStrategy",
+    "ConstrainedRoughnessSamplerStrategy",
+    "ConstrainedRoughnessAndImgSldSamplerStrategy",
+]

reflectorch/data_generation/dataset.py

@@ -0,0 +1,196 @@
+# -*- coding: utf-8 -*-
+#
+#
+# This source code is licensed under the GPL license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import Dict, Union
+import warnings
+
+from torch import Tensor
+import torch
+
+from reflectorch.data_generation.priors import PriorSampler, BasicParams
+from reflectorch.data_generation.noise import QNoiseGenerator, IntensityNoiseGenerator
+from reflectorch.data_generation.q_generator import QGenerator
+from reflectorch.data_generation.scale_curves import CurvesScaler, LogAffineCurvesScaler
+from reflectorch.data_generation.smearing import Smearing
+
+BATCH_DATA_TYPE = Dict[str, Union[Tensor, BasicParams]]
+
+
+class BasicDataset(object):
+    """Reflectometry dataset. It generates the q positions, samples the thin film parameters from the prior, 
+    simulates the reflectivity curves and applies noise to the curves.
+
+    Args:
+        q_generator (QGenerator): the momentum transfer (q) generator
+        prior_sampler (PriorSampler): the prior sampler
+        intensity_noise (IntensityNoiseGenerator, optional): the intensity noise generator. Defaults to None.
+        q_noise (QNoiseGenerator, optional): the q noise generator. Defaults to None.
+        curves_scaler (CurvesScaler, optional): the reflectivity curve scaler. Defaults to an instance of LogAffineCurvesScaler, 
+                                                which scales the curves to the range [-1, 1], the minimum considered intensity being 1e-10.
+        calc_denoised_curves (bool, optional): whether to add the curves without noise to the dictionary. Defaults to False.
+        smearing (Smearing, optional): curve smearing generator. Defaults to None.
+    """
+    def __init__(self,
+                 q_generator: QGenerator,
+                 prior_sampler: PriorSampler,
+                 intensity_noise: IntensityNoiseGenerator = None,
+                 q_noise: QNoiseGenerator = None,
+                 curves_scaler: CurvesScaler = None,
+                 calc_denoised_curves: bool = False,
+                 smearing: Smearing = None,
+                 ):
+        self.q_generator = q_generator
+        self.intensity_noise = intensity_noise
+        self.q_noise = q_noise
+        self.curves_scaler = curves_scaler or LogAffineCurvesScaler()
+        self.prior_sampler = prior_sampler
+        self.smearing = smearing
+        self.calc_denoised_curves = calc_denoised_curves
+
+    def update_batch_data(self, batch_data: BATCH_DATA_TYPE) -> None:
+        """implement in a subclass to edit batch_data dict inplace"""
+        pass
+
+    def _sample_from_prior(self, batch_size: int):
+        params: BasicParams = self.prior_sampler.sample(batch_size)
+        scaled_params: Tensor = self.prior_sampler.scale_params(params)
+        return params, scaled_params
+
+    def get_batch(self, batch_size: int) -> BATCH_DATA_TYPE:
+        """get a batch of data as a dictionary with keys ``params``, ``scaled_params``, ``q_values``, ``curves``, ``scaled_noisy_curves``
+
+        Args:
+            batch_size (int): the batch size
+        """
+        batch_data = {}
+
+        params, scaled_params = self._sample_from_prior(batch_size)
+
+        batch_data['params'] = params
+        batch_data['scaled_params'] = scaled_params
+
+        q_values: Tensor = self.q_generator.get_batch(batch_size, batch_data)
+
+        if self.q_noise:
+            batch_data['original_q_values'] = q_values
+            q_values = self.q_noise.apply(q_values, batch_data)
+
+        batch_data['q_values'] = q_values
+
+        curves = self._calc_curves(q_values, params)
+
+        if self.calc_denoised_curves:
+            batch_data['curves'] = curves
+
+        noisy_curves = curves
+
+        if self.intensity_noise:
+            noisy_curves = self.intensity_noise(noisy_curves, batch_data)
+
+        scaled_noisy_curves = self.curves_scaler.scale(noisy_curves)
+        batch_data['scaled_noisy_curves'] = scaled_noisy_curves
+
+        is_finite = torch.all(torch.isfinite(scaled_noisy_curves), -1)
+        if not torch.all(is_finite).item():
+            infinite_indices = ~is_finite
+            warnings.warn(f'Batch with {infinite_indices.sum().item()} curves with infinities skipped.')
+            return self.get_batch(batch_size = batch_size)
+
+        is_finite = torch.all(torch.isfinite(batch_data['scaled_noisy_curves']), -1)
+        assert torch.all(is_finite).item()
+
+        self.update_batch_data(batch_data)
+
+        return batch_data
+
+    def _calc_curves(self, q_values: Tensor, params: BasicParams):
+        if self.smearing:
+            curves = self.smearing.get_curves(q_values, params)
+        else:
+            curves = params.reflectivity(q_values)
+        curves = curves.to(q_values)
+        return curves
+
+
+def _insert_batch_data(tgt_batch_data, add_batch_data, indices):
+    for key in tuple(tgt_batch_data.keys()):
+        value = tgt_batch_data[key]
+        if isinstance(value, BasicParams) or len(value.shape) == 2:
+            value[indices] = add_batch_data[key]
+        else:
+            warnings.warn(f'Ignore {key} while merging batch_data.')
+
+
+if __name__ == '__main__':
+    from reflectorch.data_generation.q_generator import ConstantQ
+    from reflectorch.data_generation.priors import BasicPriorSampler, UniformSubPriorSampler
+    from reflectorch.data_generation.noise import BasicExpIntensityNoise
+    from reflectorch.data_generation.noise import BasicQNoiseGenerator
+    from reflectorch.utils import to_np
+    from time import perf_counter
+
+    q_generator = ConstantQ((0, 0.2, 65), device='cpu')
+    noise_gen = BasicExpIntensityNoise(
+        relative_errors=(0.05, 0.2),
+        # scale_range=(-1e-2, 1e-2),
+        logdist=True,
+        apply_shift=True,
+    )
+    q_noise_gen = BasicQNoiseGenerator(
+        shift_std=5e-4,
+        noise_std=(0, 1e-3),
+    )
+    prior_sampler = UniformSubPriorSampler(
+        thickness_range=(0, 250),
+        roughness_range=(0, 40),
+        sld_range=(0, 60),
+        num_layers=2,
+        device=torch.device('cpu'),
+        dtype=torch.float64,
+        smaller_roughnesses=True,
+        logdist=True,
+        relative_min_bound_width=5e-4,
+    )
+    smearing = Smearing(
+        sigma_range=(0.8e-3, 5e-3),
+        gauss_num=31,
+        share_smeared=0.5,
+    )
+
+    dataset = BasicDataset(
+        q_generator,
+        prior_sampler,
+        noise_gen,
+        q_noise=q_noise_gen,
+        smearing=smearing
+    )
+    start = perf_counter()
+    batch_data = dataset.get_batch(32)
+    print(f'Total time = {(perf_counter() - start):.3f} sec ')
+    print(batch_data['params'].roughnesses[:10])
+    print(batch_data['scaled_noisy_curves'].min().item())
+
+    scaled_noisy_curves = batch_data['scaled_noisy_curves']
+    scaled_curves = dataset.curves_scaler.scale(
+        batch_data['params'].reflectivity(q_generator.q)
+    )
+
+    try:
+        import matplotlib.pyplot as plt
+
+        for i in range(16):
+            plt.plot(
+                to_np(q_generator.q.squeeze().cpu().numpy()),
+                to_np(scaled_curves[i])
+            )
+            plt.plot(
+                to_np(q_generator.q.squeeze().cpu().numpy()),
+                to_np(scaled_noisy_curves[i])
+            )
+
+            plt.show()
+    except ImportError:
+        pass

reflectorch/data_generation/likelihoods.py

@@ -0,0 +1,86 @@
+# -*- coding: utf-8 -*-
+#
+#
+# This source code is licensed under the GPL license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import Union, Tuple
+
+import torch
+from torch import Tensor
+
+from reflectorch.data_generation import (
+    PriorSampler,
+    Params,
+)
+
+
+class LogLikelihood(object):
+    """Computes the gaussian log likelihood of the thin film parameters
+
+    Args:
+        q (Tensor): the q values
+        exp_curve (Tensor): the experimental reflectivity curve
+        priors (PriorSampler): the prior sampler
+        sigmas (Union[float, Tensor]): the sigmas (i.e. intensity error bars)
+    """
+    def __init__(self, q: Tensor, exp_curve: Tensor, priors: PriorSampler, sigmas: Union[float, Tensor]):
+        self.exp_curve = torch.atleast_2d(exp_curve)
+        self.priors: PriorSampler = priors
+        self.q = q
+        self.sigmas = sigmas
+        self.sigmas2 = self.sigmas ** 2
+
+    def calc_log_likelihood(self, curves: Tensor):
+        "computes the gaussian log likelihood"
+        log_probs = - (self.exp_curve - curves) ** 2 / self.sigmas2 / 2
+        return log_probs.sum(-1)
+
+    def __call__(self, params: Union[Params, Tensor], curves: Tensor = None):
+        if not isinstance(params, Params):
+            params: Params = self.priors.PARAM_CLS.from_tensor(params)
+        log_priors: Tensor = self.priors.log_prob(params)
+        indices: Tensor = torch.isfinite(log_priors)
+
+        if not indices.sum().item():
+            return log_priors
+
+        finite_params: Params = params[indices]
+
+        if curves is None:
+            curves: Tensor = finite_params.reflectivity(self.q)
+        else:
+            curves = curves[indices]
+
+        log_priors[indices] += self.calc_log_likelihood(curves)
+
+        return log_priors
+
+    calc_log_posterior = __call__
+
+    def get_importance_sampling_weights(
+            self, sampled_params: Params, nf_log_probs: Tensor, curves: Tensor = None
+    ) -> Tuple[Tensor, Tensor, Tensor]:
+        log_probs = self.calc_log_posterior(sampled_params, curves=curves)
+        log_weights = log_probs - nf_log_probs
+        log_weights = log_weights - log_weights.max()
+
+        weights = torch.exp(log_weights.to(torch.float64)).to(log_weights)
+        weights = weights / weights.sum()
+
+        return weights, log_weights, log_probs
+
+
+class PoissonLogLikelihood(LogLikelihood):
+    """Computes the Poisson log likelihood of the thin film parameters
+
+    Args:
+        q (Tensor): the q values
+        exp_curve (Tensor): the experimental reflectivity curve
+        priors (PriorSampler): the prior sampler
+        sigmas (Union[float, Tensor]): the sigmas (i.e. intensity error bars)
+    """    
+    def calc_log_likelihood(self, curves: Tensor):
+        """computes the Poisson log likelihood"""
+        log_probs = self.exp_curve / self.sigmas2 * (self.exp_curve * torch.log(curves) - curves)
+        return log_probs.sum(-1)