source: sasview/src/sas/sascalc/pr/fit/BumpsFitting.py @ 59dfb53

ESS_GUIESS_GUI_DocsESS_GUI_batch_fittingESS_GUI_bumps_abstractionESS_GUI_iss1116ESS_GUI_iss879ESS_GUI_iss959ESS_GUI_openclESS_GUI_orderingESS_GUI_sync_sascalcmagnetic_scattrelease-4.2.2ticket-1009ticket-1094-headlessticket-1242-2d-resolutionticket-1243ticket-1249ticket885unittest-saveload
Last change on this file since 59dfb53 was 9a5097c, checked in by andyfaff, 8 years ago

MAINT: import numpy as np

  • Property mode set to 100644
File size: 13.6 KB
RevLine 
[6fe5100]1"""
2BumpsFitting module runs the bumps optimizer.
3"""
[249a7c6]4import os
[35086c3]5from datetime import timedelta, datetime
6
[9a5097c]7import numpy as np
[6fe5100]8
9from bumps import fitters
[7945367]10try:
11    from bumps.options import FIT_CONFIG
12    # Default bumps to use the Levenberg-Marquardt optimizer
13    FIT_CONFIG.selected_id = fitters.LevenbergMarquardtFit.id
14    def get_fitter():
15        return FIT_CONFIG.selected_fitter, FIT_CONFIG.selected_values
16except:
17    # CRUFT: Bumps changed its handling of fit options around 0.7.5.6
18    # Default bumps to use the Levenberg-Marquardt optimizer
19    fitters.FIT_DEFAULT = 'lm'
20    def get_fitter():
21        fitopts = fitters.FIT_OPTIONS[fitters.FIT_DEFAULT]
22        return fitopts.fitclass, fitopts.options.copy()
23
24
[249a7c6]25from bumps.mapper import SerialMapper, MPMapper
[e3efa6b3]26from bumps import parameter
27from bumps.fitproblem import FitProblem
[7945367]28
[6fe5100]29
[b699768]30from sas.sascalc.fit.AbstractFitEngine import FitEngine
31from sas.sascalc.fit.AbstractFitEngine import FResult
32from sas.sascalc.fit.expression import compile_constraints
[6fe5100]33
[35086c3]34class Progress(object):
35    def __init__(self, history, max_step, pars, dof):
36        remaining_time = int(history.time[0]*(float(max_step)/history.step[0]-1))
37        # Depending on the time remaining, either display the expected
38        # time of completion, or the amount of time remaining.  Use precision
39        # appropriate for the duration.
40        if remaining_time >= 1800:
41            completion_time = datetime.now() + timedelta(seconds=remaining_time)
42            if remaining_time >= 36000:
43                time = completion_time.strftime('%Y-%m-%d %H:%M')
44            else:
45                time = completion_time.strftime('%H:%M')
46        else:
47            if remaining_time >= 3600:
48                time = '%dh %dm'%(remaining_time//3600, (remaining_time%3600)//60)
49            elif remaining_time >= 60:
50                time = '%dm %ds'%(remaining_time//60, remaining_time%60)
51            else:
52                time = '%ds'%remaining_time
53        chisq = "%.3g"%(2*history.value[0]/dof)
54        step = "%d of %d"%(history.step[0], max_step)
55        header = "=== Steps: %s  chisq: %s  ETA: %s\n"%(step, chisq, time)
56        parameters = ["%15s: %-10.3g%s"%(k,v,("\n" if i%3==2 else " | "))
57                      for i,(k,v) in enumerate(zip(pars,history.point[0]))]
58        self.msg = "".join([header]+parameters)
59
60    def __str__(self):
61        return self.msg
62
63
[85f17f6]64class BumpsMonitor(object):
[35086c3]65    def __init__(self, handler, max_step, pars, dof):
[85f17f6]66        self.handler = handler
67        self.max_step = max_step
[35086c3]68        self.pars = pars
69        self.dof = dof
[ed4aef2]70
[85f17f6]71    def config_history(self, history):
72        history.requires(time=1, value=2, point=1, step=1)
[ed4aef2]73
[85f17f6]74    def __call__(self, history):
[e3efa6b3]75        if self.handler is None: return
[35086c3]76        self.handler.set_result(Progress(history, self.max_step, self.pars, self.dof))
[85f17f6]77        self.handler.progress(history.step[0], self.max_step)
78        if len(history.step)>1 and history.step[1] > history.step[0]:
79            self.handler.improvement()
80        self.handler.update_fit()
81
[ed4aef2]82class ConvergenceMonitor(object):
83    """
84    ConvergenceMonitor contains population summary statistics to show progress
85    of the fit.  This is a list [ (best, 0%, 25%, 50%, 75%, 100%) ] or
86    just a list [ (best, ) ] if population size is 1.
87    """
88    def __init__(self):
89        self.convergence = []
90
91    def config_history(self, history):
92        history.requires(value=1, population_values=1)
93
94    def __call__(self, history):
95        best = history.value[0]
96        try:
97            p = history.population_values[0]
[9a5097c]98            n,p = len(p), np.sort(p)
[ed4aef2]99            QI,Qmid, = int(0.2*n),int(0.5*n)
100            self.convergence.append((best, p[0],p[QI],p[Qmid],p[-1-QI],p[-1]))
101        except:
[e3efa6b3]102            self.convergence.append((best, best,best,best,best,best))
[ed4aef2]103
[e3efa6b3]104
[4e9f227]105# Note: currently using bumps parameters for each parameter object so that
106# a SasFitness can be used directly in bumps with the usual semantics.
107# The disadvantage of this technique is that we need to copy every parameter
108# back into the model each time the function is evaluated.  We could instead
[fd5ac0d]109# define reference parameters for each sas parameter, but then we would not
[4e9f227]110# be able to express constraints using python expressions in the usual way
111# from bumps, and would instead need to use string expressions.
[e3efa6b3]112class SasFitness(object):
[6fe5100]113    """
[e3efa6b3]114    Wrap SAS model as a bumps fitness object
[6fe5100]115    """
[5044543]116    def __init__(self, model, data, fitted=[], constraints={},
117                 initial_values=None, **kw):
[4e9f227]118        self.name = model.name
119        self.model = model.model
[6fe5100]120        self.data = data
[9f7fbd9]121        if self.data.smearer is not None:
122            self.data.smearer.model = self.model
[e3efa6b3]123        self._define_pars()
124        self._init_pars(kw)
[5044543]125        if initial_values is not None:
126            self._reset_pars(fitted, initial_values)
[4e9f227]127        self.constraints = dict(constraints)
[e3efa6b3]128        self.set_fitted(fitted)
[5044543]129        self.update()
130
131    def _reset_pars(self, names, values):
132        for k,v in zip(names, values):
133            self._pars[k].value = v
[e3efa6b3]134
135    def _define_pars(self):
136        self._pars = {}
137        for k in self.model.getParamList():
138            name = ".".join((self.name,k))
139            value = self.model.getParam(k)
140            bounds = self.model.details.get(k,["",None,None])[1:3]
141            self._pars[k] = parameter.Parameter(value=value, bounds=bounds,
142                                                fixed=True, name=name)
[4e9f227]143        #print parameter.summarize(self._pars.values())
[e3efa6b3]144
145    def _init_pars(self, kw):
146        for k,v in kw.items():
147            # dispersion parameters initialized with _field instead of .field
148            if k.endswith('_width'): k = k[:-6]+'.width'
149            elif k.endswith('_npts'): k = k[:-5]+'.npts'
150            elif k.endswith('_nsigmas'): k = k[:-7]+'.nsigmas'
151            elif k.endswith('_type'): k = k[:-5]+'.type'
152            if k not in self._pars:
153                formatted_pars = ", ".join(sorted(self._pars.keys()))
154                raise KeyError("invalid parameter %r for %s--use one of: %s"
155                               %(k, self.model, formatted_pars))
156            if '.' in k and not k.endswith('.width'):
157                self.model.setParam(k, v)
158            elif isinstance(v, parameter.BaseParameter):
159                self._pars[k] = v
160            elif isinstance(v, (tuple,list)):
161                low, high = v
162                self._pars[k].value = (low+high)/2
163                self._pars[k].range(low,high)
[95d58d3]164            else:
[e3efa6b3]165                self._pars[k].value = v
166
167    def set_fitted(self, param_list):
[6fe5100]168        """
[e3efa6b3]169        Flag a set of parameters as fitted parameters.
[6fe5100]170        """
[e3efa6b3]171        for k,p in self._pars.items():
[4e9f227]172            p.fixed = (k not in param_list or k in self.constraints)
[4a0dc427]173        self.fitted_par_names = [k for k in param_list if k not in self.constraints]
[bf5e985]174        self.computed_par_names = [k for k in param_list if k in self.constraints]
175        self.fitted_pars = [self._pars[k] for k in self.fitted_par_names]
176        self.computed_pars = [self._pars[k] for k in self.computed_par_names]
[6fe5100]177
[e3efa6b3]178    # ===== Fitness interface ====
179    def parameters(self):
180        return self._pars
[6fe5100]181
[e3efa6b3]182    def update(self):
183        for k,v in self._pars.items():
[4e9f227]184            #print "updating",k,v,v.value
[e3efa6b3]185            self.model.setParam(k,v.value)
186        self._dirty = True
[6fe5100]187
[e3efa6b3]188    def _recalculate(self):
189        if self._dirty:
[9f7fbd9]190            self._residuals, self._theory \
191                = self.data.residuals(self.model.evalDistribution)
[e3efa6b3]192            self._dirty = False
[6fe5100]193
[e3efa6b3]194    def numpoints(self):
[9a5097c]195        return np.sum(self.data.idx) # number of fitted points
[6fe5100]196
[e3efa6b3]197    def nllf(self):
[9a5097c]198        return 0.5*np.sum(self.residuals()**2)
[e3efa6b3]199
200    def theory(self):
201        self._recalculate()
202        return self._theory
203
204    def residuals(self):
205        self._recalculate()
206        return self._residuals
207
208    # Not implementing the data methods for now:
209    #
210    #     resynth_data/restore_data/save/plot
[6fe5100]211
[191c648]212class ParameterExpressions(object):
213    def __init__(self, models):
214        self.models = models
215        self._setup()
216
217    def _setup(self):
218        exprs = {}
219        for M in self.models:
220            exprs.update((".".join((M.name, k)), v) for k, v in M.constraints.items())
221        if exprs:
222            symtab = dict((".".join((M.name, k)), p)
223                          for M in self.models
224                          for k,p in M.parameters().items())
225            self.update = compile_constraints(symtab, exprs)
226        else:
227            self.update = lambda: 0
228
229    def __call__(self):
230        self.update()
231
232    def __getstate__(self):
233        return self.models
234
235    def __setstate__(self, state):
236        self.models = state
237        self._setup()
238
[6fe5100]239class BumpsFit(FitEngine):
240    """
241    Fit a model using bumps.
242    """
243    def __init__(self):
244        """
245        Creates a dictionary (self.fit_arrange_dict={})of FitArrange elements
246        with Uid as keys
247        """
248        FitEngine.__init__(self)
249        self.curr_thread = None
250
251    def fit(self, msg_q=None,
252            q=None, handler=None, curr_thread=None,
253            ftol=1.49012e-8, reset_flag=False):
[e3efa6b3]254        # Build collection of bumps fitness calculators
[bf5e985]255        models = [SasFitness(model=M.get_model(),
256                             data=M.get_data(),
257                             constraints=M.constraints,
[5044543]258                             fitted=M.pars,
259                             initial_values=M.vals if reset_flag else None)
[bf5e985]260                  for M in self.fit_arrange_dict.values()
261                  if M.get_to_fit()]
[233c121]262        if len(models) == 0:
263            raise RuntimeError("Nothing to fit")
[e3efa6b3]264        problem = FitProblem(models)
265
[191c648]266        # TODO: need better handling of parameter expressions and bounds constraints
267        # so that they are applied during polydispersity calculations.  This
268        # will remove the immediate need for the setp_hook in bumps, though
269        # bumps may still need something similar, such as a sane class structure
270        # which allows a subclass to override setp.
271        problem.setp_hook = ParameterExpressions(models)
[4e9f227]272
[e3efa6b3]273        # Run the fit
274        result = run_bumps(problem, handler, curr_thread)
[6fe5100]275        if handler is not None:
276            handler.update_fit(last=True)
[e3efa6b3]277
[eff93b8]278        # TODO: shouldn't reference internal parameters of fit problem
[e3efa6b3]279        varying = problem._parameters
280        # collect the results
281        all_results = []
282        for M in problem.models:
283            fitness = M.fitness
284            fitted_index = [varying.index(p) for p in fitness.fitted_pars]
[e1442d4]285            param_list = fitness.fitted_par_names + fitness.computed_par_names
[e3efa6b3]286            R = FResult(model=fitness.model, data=fitness.data,
[e1442d4]287                        param_list=param_list)
[e3efa6b3]288            R.theory = fitness.theory()
289            R.residuals = fitness.residuals()
[5044543]290            R.index = fitness.data.idx
[e3efa6b3]291            R.fitter_id = self.fitter_id
[eff93b8]292            # TODO: should scale stderr by sqrt(chisq/DOF) if dy is unknown
[e3efa6b3]293            R.success = result['success']
[e1442d4]294            if R.success:
[9a5097c]295                R.stderr = np.hstack((result['stderr'][fitted_index],
296                                      np.NaN*np.ones(len(fitness.computed_pars))))
297                R.pvec = np.hstack((result['value'][fitted_index],
[e1442d4]298                                      [p.value for p in fitness.computed_pars]))
[9a5097c]299                R.fitness = np.sum(R.residuals**2)/(fitness.numpoints() - len(fitted_index))
[e1442d4]300            else:
[9a5097c]301                R.stderr = np.NaN*np.ones(len(param_list))
302                R.pvec = np.asarray( [p.value for p in fitness.fitted_pars+fitness.computed_pars])
303                R.fitness = np.NaN
[e3efa6b3]304            R.convergence = result['convergence']
305            if result['uncertainty'] is not None:
306                R.uncertainty_state = result['uncertainty']
307            all_results.append(R)
308
[6fe5100]309        if q is not None:
[e3efa6b3]310            q.put(all_results)
[6fe5100]311            return q
[e3efa6b3]312        else:
313            return all_results
[6fe5100]314
[e3efa6b3]315def run_bumps(problem, handler, curr_thread):
[85f17f6]316    def abort_test():
317        if curr_thread is None: return False
318        try: curr_thread.isquit()
319        except KeyboardInterrupt:
320            if handler is not None:
321                handler.stop("Fitting: Terminated!!!")
322            return True
323        return False
324
[7945367]325    fitclass, options = get_fitter()
326    steps = options.get('steps', 0)
327    if steps == 0:
328        pop = options.get('pop',0)*len(problem._parameters)
329        samples = options.get('samples', 0)
330        steps = (samples+pop-1)/pop if pop != 0 else samples
331    max_step = steps + options.get('burn', 0)
[35086c3]332    pars = [p.name for p in problem._parameters]
[9a5097c]333    #x0 = np.asarray([p.value for p in problem._parameters])
[e3efa6b3]334    options['monitors'] = [
[35086c3]335        BumpsMonitor(handler, max_step, pars, problem.dof),
[e3efa6b3]336        ConvergenceMonitor(),
337        ]
[95d58d3]338    fitdriver = fitters.FitDriver(fitclass, problem=problem,
[042f065]339                                  abort_test=abort_test, **options)
[249a7c6]340    omp_threads = int(os.environ.get('OMP_NUM_THREADS','0'))
[e1442d4]341    mapper = MPMapper if omp_threads == 1 else SerialMapper
[6fe5100]342    fitdriver.mapper = mapper.start_mapper(problem, None)
[233c121]343    #import time; T0 = time.time()
[6fe5100]344    try:
345        best, fbest = fitdriver.fit()
346    except:
347        import traceback; traceback.print_exc()
348        raise
[95d58d3]349    finally:
350        mapper.stop_mapper(fitdriver.mapper)
[e3efa6b3]351
352
353    convergence_list = options['monitors'][-1].convergence
[9a5097c]354    convergence = (2*np.asarray(convergence_list)/problem.dof
355                   if convergence_list else np.empty((0,1),'d'))
[e1442d4]356
357    success = best is not None
[e3efa6b3]358    return {
[e1442d4]359        'value': best if success else None,
360        'stderr': fitdriver.stderr() if success else None,
361        'success': success,
[e3efa6b3]362        'convergence': convergence,
363        'uncertainty': getattr(fitdriver.fitter, 'state', None),
364        }
[6fe5100]365
Note: See TracBrowser for help on using the repository browser.