Changes in / [e589e9a:9150036] in sasmodels


Ignore:
Files:
7 edited

Legend:

Unmodified
Added
Removed
  • doc/guide/plugin.rst

    r81751c2 r94bfa42  
    272272structure factor to account for interactions between particles.  See 
    273273`Form_Factors`_ for more details. 
     274 
     275**model_info = ...** lets you define a model directly, for example, by 
     276loading and modifying existing models.  This is done implicitly by 
     277:func:`sasmodels.core.load_model_info`, which can create a mixture model 
     278from a pair of existing models.  For example:: 
     279 
     280    from sasmodels.core import load_model_info 
     281    model_info = load_model_info('sphere+cylinder') 
     282 
     283See :class:`sasmodels.modelinfo.ModelInfo` for details about the model 
     284attributes that are defined. 
    274285 
    275286Model Parameters 
     
    894905             - \frac{\sin(x)}{x}\left(\frac{1}{x} - \frac{3!}{x^3} + \frac{5!}{x^5} - \frac{7!}{x^7}\right) 
    895906 
    896         For small arguments , 
     907        For small arguments, 
    897908 
    898909        .. math:: 
  • example/multiscatfit.py

    r49d1f8b8 r2c4a190  
    1515 
    1616    # Show the model without fitting 
    17     PYTHONPATH=..:../explore:../../bumps:../../sasview/src python multiscatfit.py 
     17    PYTHONPATH=..:../../bumps:../../sasview/src python multiscatfit.py 
    1818 
    1919    # Run the fit 
    20     PYTHONPATH=..:../explore:../../bumps:../../sasview/src ../../bumps/run.py \ 
     20    PYTHONPATH=..:../../bumps:../../sasview/src ../../bumps/run.py \ 
    2121    multiscatfit.py --store=/tmp/t1 
    2222 
     
    5555    ) 
    5656 
     57# Tie the model to the data 
     58M = Experiment(data=data, model=model) 
     59 
     60# Stack mulitple scattering on top of the existing resolution function. 
     61M.resolution = MultipleScattering(resolution=M.resolution, probability=0.) 
     62 
    5763# SET THE FITTING PARAMETERS 
    5864model.radius_polar.range(15, 3000) 
     
    6571model.scale.range(0, 0.1) 
    6672 
    67 # Mulitple scattering probability parameter 
    68 # HACK: the probability is stuffed in as an extra parameter to the experiment. 
    69 probability = Parameter(name="probability", value=0.0) 
    70 probability.range(0.0, 0.9) 
     73# The multiple scattering probability parameter is in the resolution function 
     74# instead of the scattering function, so access it through M.resolution 
     75M.scattering_probability.range(0.0, 0.9) 
    7176 
    72 M = Experiment(data=data, model=model, extra_pars={'probability': probability}) 
    73  
    74 # Stack mulitple scattering on top of the existing resolution function. 
    75 # Because resolution functions in sasview don't have fitting parameters, 
    76 # we instead allow the multiple scattering calculator to take a function 
    77 # instead of a probability.  This function returns the current value of 
    78 # the parameter. ** THIS IS TEMPORARY ** when multiple scattering is 
    79 # properly integrated into sasmodels and sasview, its fittable parameter 
    80 # will be treated like the model parameters. 
    81 M.resolution = MultipleScattering(resolution=M.resolution, 
    82                                   probability=lambda: probability.value, 
    83                                   ) 
    84 M._kernel_inputs = M.resolution.q_calc 
     77# Let bumps know that we are fitting this experiment 
    8578problem = FitProblem(M) 
    8679 
  • sasmodels/__init__.py

    ra1ec908 r37f38ff  
    1414defining new models. 
    1515""" 
    16 __version__ = "0.98" 
     16__version__ = "0.99" 
    1717 
    1818def data_files(): 
  • sasmodels/bumps_model.py

    r49d1f8b8 r2c4a190  
    3535    # when bumps is not on the path. 
    3636    from bumps.names import Parameter # type: ignore 
     37    from bumps.parameter import Reference # type: ignore 
    3738except ImportError: 
    3839    pass 
     
    139140    def __init__(self, data, model, cutoff=1e-5, name=None, extra_pars=None): 
    140141        # type: (Data, Model, float) -> None 
     142        # Allow resolution function to define fittable parameters.  We do this 
     143        # by creating reference parameters within the resolution object rather 
     144        # than modifying the object itself to use bumps parameters.  We need 
     145        # to reset the parameters each time the object has changed.  These 
     146        # additional parameters need to be returned from the fitting engine. 
     147        # To make them available to the user, they are added as top-level 
     148        # attributes to the experiment object.  The only change to the 
     149        # resolution function is that it needs an optional 'fittable' attribute 
     150        # which maps the internal name to the user visible name for the 
     151        # for the parameter. 
     152        self._resolution = None 
     153        self._resolution_pars = {} 
    141154        # remember inputs so we can inspect from outside 
    142155        self.name = data.filename if name is None else name 
     
    145158        self._interpret_data(data, model.sasmodel) 
    146159        self._cache = {} 
     160        # CRUFT: no longer need extra parameters 
     161        # Multiple scattering probability is now retrieved directly from the 
     162        # multiple scattering resolution function. 
    147163        self.extra_pars = extra_pars 
    148164 
     
    162178        return len(self.Iq) 
    163179 
     180    @property 
     181    def resolution(self): 
     182        return self._resolution 
     183 
     184    @resolution.setter 
     185    def resolution(self, value): 
     186        self._resolution = value 
     187 
     188        # Remove old resolution fitting parameters from experiment 
     189        for name in self._resolution_pars: 
     190            delattr(self, name) 
     191 
     192        # Create new resolution fitting parameters 
     193        res_pars = getattr(self._resolution, 'fittable', {}) 
     194        self._resolution_pars = { 
     195            name: Reference(self._resolution, refname, name=name) 
     196            for refname, name in res_pars.items() 
     197        } 
     198 
     199        # Add new resolution fitting parameters as experiment attributes 
     200        for name, ref in self._resolution_pars.items(): 
     201            setattr(self, name, ref) 
     202 
    164203    def parameters(self): 
    165204        # type: () -> Dict[str, Parameter] 
     
    168207        """ 
    169208        pars = self.model.parameters() 
    170         if self.extra_pars: 
     209        if self.extra_pars is not None: 
    171210            pars.update(self.extra_pars) 
     211        pars.update(self._resolution_pars) 
    172212        return pars 
    173213 
  • sasmodels/direct_model.py

    r5024a56 r2c4a190  
    224224            else: 
    225225                Iq, dIq = None, None 
    226             #self._theory = np.zeros_like(q) 
    227             q_vectors = [res.q_calc] 
    228226        elif self.data_type == 'Iqxy': 
    229227            #if not model.info.parameters.has_2d: 
     
    242240            res = resolution2d.Pinhole2D(data=data, index=index, 
    243241                                         nsigma=3.0, accuracy=accuracy) 
    244             #self._theory = np.zeros_like(self.Iq) 
    245             q_vectors = res.q_calc 
    246242        elif self.data_type == 'Iq': 
    247243            index = (data.x >= data.qmin) & (data.x <= data.qmax) 
     
    268264            else: 
    269265                res = resolution.Perfect1D(data.x[index]) 
    270  
    271             #self._theory = np.zeros_like(self.Iq) 
    272             q_vectors = [res.q_calc] 
    273266        elif self.data_type == 'Iq-oriented': 
    274267            index = (data.x >= data.qmin) & (data.x <= data.qmax) 
     
    286279                                      qx_width=data.dxw[index], 
    287280                                      qy_width=data.dxl[index]) 
    288             q_vectors = res.q_calc 
    289281        else: 
    290282            raise ValueError("Unknown data type") # never gets here 
     
    292284        # Remember function inputs so we can delay loading the function and 
    293285        # so we can save/restore state 
    294         self._kernel_inputs = q_vectors 
    295286        self._kernel = None 
    296287        self.Iq, self.dIq, self.index = Iq, dIq, index 
     
    329320        # type: (ParameterSet, float) -> np.ndarray 
    330321        if self._kernel is None: 
    331             self._kernel = self._model.make_kernel(self._kernel_inputs) 
     322            # TODO: change interfaces so that resolution returns kernel inputs 
     323            # Maybe have resolution always return a tuple, or maybe have 
     324            # make_kernel accept either an ndarray or a pair of ndarrays. 
     325            kernel_inputs = self.resolution.q_calc 
     326            if isinstance(kernel_inputs, np.ndarray): 
     327                kernel_inputs = (kernel_inputs,) 
     328            self._kernel = self._model.make_kernel(kernel_inputs) 
    332329 
    333330        # Need to pull background out of resolution for multiple scattering 
  • sasmodels/multiscat.py

    rb3703f5 r2c4a190  
    342342 
    343343    *probability* is related to the expected number of scattering 
    344     events in the sample $\lambda$ as $p = 1 = e^{-\lambda}$.  As a 
    345     hack to allow probability to be a fitted parameter, the "value" 
    346     can be a function that takes no parameters and returns the current 
    347     value of the probability.  *coverage* determines how many scattering 
    348     steps to consider.  The default is 0.99, which sets $n$ such that 
    349     $1 \ldots n$ covers 99% of the Poisson probability mass function. 
     344    events in the sample $\lambda$ as $p = 1 - e^{-\lambda}$. 
     345    *coverage* determines how many scattering steps to consider.  The 
     346    default is 0.99, which sets $n$ such that $1 \ldots n$ covers 99% 
     347    of the Poisson probability mass function. 
    350348 
    351349    *is2d* is True then 2D scattering is used, otherwise it accepts 
     
    399397        self.qmin = qmin 
    400398        self.nq = nq 
    401         self.probability = probability 
     399        self.probability = 0. if probability is None else probability 
    402400        self.coverage = coverage 
    403401        self.is2d = is2d 
     
    456454        self.Iqxy = None # type: np.ndarray 
    457455 
     456        # Label probability as a fittable parameter, and give its external name 
     457        # Note that the external name must be a valid python identifier, since 
     458        # is will be set as an experiment attribute. 
     459        self.fittable = {'probability': 'scattering_probability'} 
     460 
    458461    def apply(self, theory): 
    459462        if self.is2d: 
     
    463466        Iq_calc = Iq_calc.reshape(self.nq, self.nq) 
    464467 
     468        # CRUFT: don't need probability as a function anymore 
    465469        probability = self.probability() if callable(self.probability) else self.probability 
    466470        coverage = self.coverage 
  • sasmodels/sasview_model.py

    r3a1afed r3a1afed  
    2525from . import core 
    2626from . import custom 
     27from . import kernelcl 
    2728from . import product 
    2829from . import generate 
     
    3031from . import modelinfo 
    3132from .details import make_kernel_args, dispersion_mesh 
     33from .kernelcl import reset_environment 
    3234 
    3335# pylint: disable=unused-import 
     
    6870#: has changed since we last reloaded. 
    6971_CACHED_MODULE = {}  # type: Dict[str, "module"] 
     72 
     73def reset_environment(): 
     74    # type: () -> None 
     75    """ 
     76    Clear the compute engine context so that the GUI can change devices. 
     77 
     78    This removes all compiled kernels, even those that are active on fit 
     79    pages, but they will be restored the next time they are needed. 
     80    """ 
     81    kernelcl.reset_environment() 
     82    for model in MODELS.values(): 
     83        model._model = None 
    7084 
    7185def find_model(modelname): 
     
    696710    def _calculate_Iq(self, qx, qy=None): 
    697711        if self._model is None: 
    698             self._model = core.build_model(self._model_info) 
     712            # Only need one copy of the compiled kernel regardless of how many 
     713            # times it is used, so store it in the class.  Also, to reset the 
     714            # compute engine, need to clear out all existing compiled kernels, 
     715            # which is much easier to do if we store them in the class. 
     716            self.__class__._model = core.build_model(self._model_info) 
    699717        if qy is not None: 
    700718            q_vectors = [np.asarray(qx), np.asarray(qy)] 
Note: See TracChangeset for help on using the changeset viewer.