1 | |
---|
2 | |
---|
3 | |
---|
4 | """ |
---|
5 | ParkFitting module contains SansParameter,Model,Data |
---|
6 | FitArrange, ParkFit,Parameter classes.All listed classes work together |
---|
7 | to perform a simple fit with park optimizer. |
---|
8 | """ |
---|
9 | #import time |
---|
10 | import numpy |
---|
11 | import math |
---|
12 | from numpy.linalg.linalg import LinAlgError |
---|
13 | #import park |
---|
14 | from park import fit |
---|
15 | from park import fitresult |
---|
16 | from park.fitresult import FitParameter |
---|
17 | import park.simplex |
---|
18 | from park.assembly import Assembly |
---|
19 | from park.assembly import Part |
---|
20 | from park.fitmc import FitSimplex |
---|
21 | import park.fitmc |
---|
22 | from park.fitmc import FitMC |
---|
23 | from park.fit import Fitter |
---|
24 | from park.formatnum import format_uncertainty |
---|
25 | #from Loader import Load |
---|
26 | from sans.fit.AbstractFitEngine import FitEngine |
---|
27 | from sans.fit.AbstractFitEngine import FResult |
---|
28 | |
---|
29 | class SansFitResult(fitresult.FitResult): |
---|
30 | def __init__(self, *args, **kwrds): |
---|
31 | fitresult.FitResult.__init__(self, *args, **kwrds) |
---|
32 | self.theory = None |
---|
33 | self.inputs = [] |
---|
34 | |
---|
35 | class SansFitSimplex(FitSimplex): |
---|
36 | """ |
---|
37 | Local minimizer using Nelder-Mead simplex algorithm. |
---|
38 | |
---|
39 | Simplex is robust and derivative free, though not very efficient. |
---|
40 | |
---|
41 | This class wraps the bounds contrained Nelder-Mead simplex |
---|
42 | implementation for `park.simplex.simplex`. |
---|
43 | """ |
---|
44 | radius = 0.05 |
---|
45 | """Size of the initial simplex; this is a portion between 0 and 1""" |
---|
46 | xtol = 1 |
---|
47 | #xtol = 1e-4 |
---|
48 | """Stop when simplex vertices are within xtol of each other""" |
---|
49 | ftol = 5e-5 |
---|
50 | """Stop when vertex values are within ftol of each other""" |
---|
51 | maxiter = None |
---|
52 | """Maximum number of iterations before fit terminates""" |
---|
53 | def __init__(self, ftol=5e-5): |
---|
54 | self.ftol = ftol |
---|
55 | |
---|
56 | def fit(self, fitness, x0): |
---|
57 | """Run the fit""" |
---|
58 | self.cancel = False |
---|
59 | pars = fitness.fit_parameters() |
---|
60 | bounds = numpy.array([p.range for p in pars]).T |
---|
61 | result = park.simplex.simplex(fitness, x0, bounds=bounds, |
---|
62 | radius=self.radius, xtol=self.xtol, |
---|
63 | ftol=self.ftol, maxiter=self.maxiter, |
---|
64 | abort_test=self._iscancelled) |
---|
65 | #print "calls:",result.calls |
---|
66 | #print "simplex returned",result.x,result.fx |
---|
67 | # Need to make our own copy of the fit results so that the |
---|
68 | # values don't get stomped on by the next fit iteration. |
---|
69 | fitpars = [SansFitParameter(pars[i].name,pars[i].range,v, pars[i].model, pars[i].data) |
---|
70 | for i,v in enumerate(result.x)] |
---|
71 | res = SansFitResult(fitpars, result.calls, result.fx) |
---|
72 | res.inputs = [(pars[i].model, pars[i].data) for i,v in enumerate(result.x)] |
---|
73 | # Compute the parameter uncertainties from the jacobian |
---|
74 | res.calc_cov(fitness) |
---|
75 | return res |
---|
76 | |
---|
77 | class SansFitter(Fitter): |
---|
78 | """ |
---|
79 | """ |
---|
80 | def fit(self, fitness, handler): |
---|
81 | """ |
---|
82 | Global optimizer. |
---|
83 | |
---|
84 | This function should return immediately |
---|
85 | """ |
---|
86 | # Determine initial value and bounds |
---|
87 | pars = fitness.fit_parameters() |
---|
88 | bounds = numpy.array([p.range for p in pars]).T |
---|
89 | x0 = [p.value for p in pars] |
---|
90 | |
---|
91 | # Initialize the monitor and results. |
---|
92 | # Need to make our own copy of the fit results so that the |
---|
93 | # values don't get stomped on by the next fit iteration. |
---|
94 | handler.done = False |
---|
95 | self.handler = handler |
---|
96 | fitpars = [SansFitParameter(pars[i].name, pars[i].range, v, |
---|
97 | pars[i].model, pars[i].data) |
---|
98 | for i,v in enumerate(x0)] |
---|
99 | handler.result = fitresult.FitResult(fitpars, 0, numpy.NaN) |
---|
100 | |
---|
101 | # Run the fit (fit should perform _progress and _improvement updates) |
---|
102 | # This function may return before the fit is complete. |
---|
103 | self._fit(fitness, x0, bounds) |
---|
104 | |
---|
105 | class SansFitMC(SansFitter): |
---|
106 | """ |
---|
107 | Monte Carlo optimizer. |
---|
108 | |
---|
109 | This implements `park.fit.Fitter`. |
---|
110 | """ |
---|
111 | localfit = SansFitSimplex() |
---|
112 | start_points = 10 |
---|
113 | def __init__(self, localfit, start_points=10): |
---|
114 | self.localfit = localfit |
---|
115 | self.start_points = start_points |
---|
116 | |
---|
117 | def _fit(self, objective, x0, bounds): |
---|
118 | """ |
---|
119 | Run a monte carlo fit. |
---|
120 | |
---|
121 | This procedure maps a local optimizer across a set of initial points. |
---|
122 | """ |
---|
123 | try: |
---|
124 | park.fitmc.fitmc(objective, x0, bounds, self.localfit, |
---|
125 | self.start_points, self.handler) |
---|
126 | except: |
---|
127 | raise ValueError, "Fit did not converge.\n" |
---|
128 | |
---|
129 | class SansPart(Part): |
---|
130 | """ |
---|
131 | Part of a fitting assembly. Part holds the model itself and |
---|
132 | associated data. The part can be initialized with a fitness |
---|
133 | object or with a pair (model,data) for the default fitness function. |
---|
134 | |
---|
135 | fitness (Fitness) |
---|
136 | object implementing the `park.assembly.Fitness` interface. In |
---|
137 | particular, fitness should provide a parameterset attribute |
---|
138 | containing a ParameterSet and a residuals method returning a vector |
---|
139 | of residuals. |
---|
140 | weight (dimensionless) |
---|
141 | weight for the model. See comments in assembly.py for details. |
---|
142 | isfitted (boolean) |
---|
143 | True if the model residuals should be included in the fit. |
---|
144 | The model parameters may still be used in parameter |
---|
145 | expressions, but there will be no comparison to the data. |
---|
146 | residuals (vector) |
---|
147 | Residuals for the model if they have been calculated, or None |
---|
148 | degrees_of_freedom |
---|
149 | Number of residuals minus number of fitted parameters. |
---|
150 | Degrees of freedom for individual models does not make |
---|
151 | sense in the presence of expressions combining models, |
---|
152 | particularly in the case where a model has many parameters |
---|
153 | but no data or many computed parameters. The degrees of |
---|
154 | freedom for the model is set to be at least one. |
---|
155 | chisq |
---|
156 | sum(residuals**2); use chisq/degrees_of_freedom to |
---|
157 | get the reduced chisq value. |
---|
158 | |
---|
159 | Get/set the weight on the given model. |
---|
160 | |
---|
161 | assembly.weight(3) returns the weight on model 3 (0-origin) |
---|
162 | assembly.weight(3,0.5) sets the weight on model 3 (0-origin) |
---|
163 | """ |
---|
164 | |
---|
165 | def __init__(self, fitness, weight=1., isfitted=True): |
---|
166 | Part.__init__(self, fitness=fitness, weight=weight, |
---|
167 | isfitted=isfitted) |
---|
168 | |
---|
169 | self.model, self.data = fitness[0], fitness[1] |
---|
170 | |
---|
171 | class SansFitParameter(FitParameter): |
---|
172 | """ |
---|
173 | Fit result for an individual parameter. |
---|
174 | """ |
---|
175 | def __init__(self, name, range, value, model, data): |
---|
176 | FitParameter.__init__(self, name, range, value) |
---|
177 | self.model = model |
---|
178 | self.data = data |
---|
179 | |
---|
180 | def summarize(self): |
---|
181 | """ |
---|
182 | Return parameter range string. |
---|
183 | |
---|
184 | E.g., " Gold .....|.... 5.2043 in [2,7]" |
---|
185 | """ |
---|
186 | bar = ['.']*10 |
---|
187 | lo,hi = self.range |
---|
188 | if numpy.isfinite(lo)and numpy.isfinite(hi): |
---|
189 | portion = (self.value-lo)/(hi-lo) |
---|
190 | if portion < 0: portion = 0. |
---|
191 | elif portion >= 1: portion = 0.99999999 |
---|
192 | barpos = int(math.floor(portion*len(bar))) |
---|
193 | bar[barpos] = '|' |
---|
194 | bar = "".join(bar) |
---|
195 | lostr = "[%g"%lo if numpy.isfinite(lo) else "(-inf" |
---|
196 | histr = "%g]"%hi if numpy.isfinite(hi) else "inf)" |
---|
197 | valstr = format_uncertainty(self.value, self.stderr) |
---|
198 | model_name = str(None) |
---|
199 | if self.model is not None: |
---|
200 | model_name = self.model.name |
---|
201 | data_name = str(None) |
---|
202 | if self.data is not None: |
---|
203 | data_name = self.data.name |
---|
204 | |
---|
205 | return "%25s %s %s in %s,%s, %s, %s" % (self.name,bar,valstr,lostr,histr, |
---|
206 | model_name, data_name) |
---|
207 | def __repr__(self): |
---|
208 | #return "FitParameter('%s')"%self.name |
---|
209 | return str(self.__class__) |
---|
210 | |
---|
211 | class MyAssembly(Assembly): |
---|
212 | def __init__(self, models, curr_thread=None): |
---|
213 | """Build an assembly from a list of models.""" |
---|
214 | self.parts = [] |
---|
215 | for m in models: |
---|
216 | self.parts.append(SansPart(m)) |
---|
217 | self.curr_thread = curr_thread |
---|
218 | self.chisq = None |
---|
219 | self._cancel = False |
---|
220 | self.theory = None |
---|
221 | self._reset() |
---|
222 | |
---|
223 | def fit_parameters(self): |
---|
224 | """ |
---|
225 | Return an alphabetical list of the fitting parameters. |
---|
226 | |
---|
227 | This function is called once at the beginning of a fit, |
---|
228 | and serves as a convenient place to precalculate what |
---|
229 | can be precalculated such as the set of fitting parameters |
---|
230 | and the parameter expressions evaluator. |
---|
231 | """ |
---|
232 | self.parameterset.setprefix() |
---|
233 | self._fitparameters = self.parameterset.fitted |
---|
234 | self._restraints = self.parameterset.restrained |
---|
235 | pars = self.parameterset.flatten() |
---|
236 | context = self.parameterset.gather_context() |
---|
237 | self._fitexpression = park.expression.build_eval(pars,context) |
---|
238 | #print "constraints",self._fitexpression.__doc__ |
---|
239 | |
---|
240 | self._fitparameters.sort(lambda a,b: cmp(a.path,b.path)) |
---|
241 | # Convert to fitparameter a object |
---|
242 | |
---|
243 | fitpars = [SansFitParameter(p.path,p.range,p.value, p.model, p.data) |
---|
244 | for p in self._fitparameters] |
---|
245 | #print "fitpars", fitpars |
---|
246 | return fitpars |
---|
247 | |
---|
248 | def all_results(self, result): |
---|
249 | """ |
---|
250 | Extend result from the fit with the calculated parameters. |
---|
251 | """ |
---|
252 | calcpars = [SansFitParameter(p.path,p.range,p.value, p.model, p.data) |
---|
253 | for p in self.parameterset.computed] |
---|
254 | result.parameters += calcpars |
---|
255 | result.theory = self.theory |
---|
256 | |
---|
257 | def eval(self): |
---|
258 | """ |
---|
259 | Recalculate the theory functions, and from them, the |
---|
260 | residuals and chisq. |
---|
261 | |
---|
262 | :note: Call this after the parameters have been updated. |
---|
263 | """ |
---|
264 | # Handle abort from a separate thread. |
---|
265 | self._cancel = False |
---|
266 | if self.curr_thread != None: |
---|
267 | try: |
---|
268 | self.curr_thread.isquit() |
---|
269 | except: |
---|
270 | self._cancel = True |
---|
271 | |
---|
272 | # Evaluate the computed parameters |
---|
273 | try: |
---|
274 | self._fitexpression() |
---|
275 | except NameError: |
---|
276 | pass |
---|
277 | |
---|
278 | # Check that the resulting parameters are in a feasible region. |
---|
279 | if not self.isfeasible(): return numpy.inf |
---|
280 | |
---|
281 | resid = [] |
---|
282 | k = len(self._fitparameters) |
---|
283 | for m in self.parts: |
---|
284 | # In order to support abort, need to be able to propagate an |
---|
285 | # external abort signal from self.abort() into an abort signal |
---|
286 | # for the particular model. Can't see a way to do this which |
---|
287 | # doesn't involve setting a state variable. |
---|
288 | self._current_model = m |
---|
289 | if self._cancel: return numpy.inf |
---|
290 | if m.isfitted and m.weight != 0: |
---|
291 | m.residuals, self.theory = m.fitness.residuals() |
---|
292 | N = len(m.residuals) |
---|
293 | m.degrees_of_freedom = N-k if N>k else 1 |
---|
294 | # dividing residuals by N in order to be consistent with Scipy |
---|
295 | m.chisq = numpy.sum(m.residuals**2/N) |
---|
296 | resid.append(m.weight*m.residuals/math.sqrt(N)) |
---|
297 | self.residuals = numpy.hstack(resid) |
---|
298 | N = len(self.residuals) |
---|
299 | self.degrees_of_freedom = N-k if N>k else 1 |
---|
300 | self.chisq = numpy.sum(self.residuals**2) |
---|
301 | return self.chisq |
---|
302 | |
---|
303 | class ParkFit(FitEngine): |
---|
304 | """ |
---|
305 | ParkFit performs the Fit.This class can be used as follow: |
---|
306 | #Do the fit Park |
---|
307 | create an engine: engine = ParkFit() |
---|
308 | Use data must be of type plottable |
---|
309 | Use a sans model |
---|
310 | |
---|
311 | Add data with a dictionnary of FitArrangeList where Uid is a key and data |
---|
312 | is saved in FitArrange object. |
---|
313 | engine.set_data(data,Uid) |
---|
314 | |
---|
315 | Set model parameter "M1"= model.name add {model.parameter.name:value}. |
---|
316 | |
---|
317 | :note: Set_param() if used must always preceded set_model() |
---|
318 | for the fit to be performed. |
---|
319 | engine.set_param( model,"M1", {'A':2,'B':4}) |
---|
320 | |
---|
321 | Add model with a dictionnary of FitArrangeList{} where Uid is a key |
---|
322 | and model |
---|
323 | is save in FitArrange object. |
---|
324 | engine.set_model(model,Uid) |
---|
325 | |
---|
326 | engine.fit return chisqr,[model.parameter 1,2,..],[[err1....][..err2...]] |
---|
327 | chisqr1, out1, cov1=engine.fit({model.parameter.name:value},qmin,qmax) |
---|
328 | |
---|
329 | :note: {model.parameter.name:value} is ignored in fit function since |
---|
330 | the user should make sure to call set_param himself. |
---|
331 | |
---|
332 | """ |
---|
333 | def __init__(self): |
---|
334 | """ |
---|
335 | Creates a dictionary (self.fitArrangeList={})of FitArrange elements |
---|
336 | with Uid as keys |
---|
337 | """ |
---|
338 | FitEngine.__init__(self) |
---|
339 | self.fit_arrange_dict = {} |
---|
340 | self.param_list = [] |
---|
341 | |
---|
342 | def create_assembly(self, curr_thread, reset_flag=False): |
---|
343 | """ |
---|
344 | Extract sansmodel and sansdata from |
---|
345 | self.FitArrangelist ={Uid:FitArrange} |
---|
346 | Create parkmodel and park data ,form a list couple of parkmodel |
---|
347 | and parkdata |
---|
348 | create an assembly self.problem= park.Assembly([(parkmodel,parkdata)]) |
---|
349 | """ |
---|
350 | mylist = [] |
---|
351 | #listmodel = [] |
---|
352 | #i = 0 |
---|
353 | fitproblems = [] |
---|
354 | for fproblem in self.fit_arrange_dict.itervalues(): |
---|
355 | if fproblem.get_to_fit() == 1: |
---|
356 | fitproblems.append(fproblem) |
---|
357 | if len(fitproblems) == 0: |
---|
358 | raise RuntimeError, "No Assembly scheduled for Park fitting." |
---|
359 | return |
---|
360 | for item in fitproblems: |
---|
361 | parkmodel = item.get_model() |
---|
362 | if reset_flag: |
---|
363 | # reset the initial value; useful for batch |
---|
364 | for name in item.pars: |
---|
365 | ind = item.pars.index(name) |
---|
366 | parkmodel.model.setParam(name, item.vals[ind]) |
---|
367 | |
---|
368 | for p in parkmodel.parameterset: |
---|
369 | ## does not allow status change for constraint parameters |
---|
370 | if p.status != 'computed': |
---|
371 | if p.get_name()in item.pars: |
---|
372 | ## make parameters selected for |
---|
373 | #fit will be between boundaries |
---|
374 | p.set(p.range) |
---|
375 | else: |
---|
376 | p.status = 'fixed' |
---|
377 | data_list = item.get_data() |
---|
378 | parkdata = data_list |
---|
379 | fitness = (parkmodel, parkdata) |
---|
380 | mylist.append(fitness) |
---|
381 | self.problem = MyAssembly(models=mylist, curr_thread=curr_thread) |
---|
382 | |
---|
383 | |
---|
384 | def fit(self, msg_q=None, |
---|
385 | q=None, handler=None, curr_thread=None, |
---|
386 | ftol=1.49012e-8, reset_flag=False): |
---|
387 | """ |
---|
388 | Performs fit with park.fit module.It can perform fit with one model |
---|
389 | and a set of data, more than two fit of one model and sets of data or |
---|
390 | fit with more than two model associated with their set of data and |
---|
391 | constraints |
---|
392 | |
---|
393 | :param pars: Dictionary of parameter names for the model and their |
---|
394 | values. |
---|
395 | :param qmin: The minimum value of data's range to be fit |
---|
396 | :param qmax: The maximum value of data's range to be fit |
---|
397 | |
---|
398 | :note: all parameter are ignored most of the time.Are just there |
---|
399 | to keep ScipyFit and ParkFit interface the same. |
---|
400 | |
---|
401 | :return: result.fitness Value of the goodness of fit metric |
---|
402 | :return: result.pvec list of parameter with the best value |
---|
403 | found during fitting |
---|
404 | :return: result.cov Covariance matrix |
---|
405 | |
---|
406 | """ |
---|
407 | self.create_assembly(curr_thread=curr_thread, reset_flag=reset_flag) |
---|
408 | localfit = SansFitSimplex() |
---|
409 | localfit.ftol = ftol |
---|
410 | |
---|
411 | # See `park.fitresult.FitHandler` for details. |
---|
412 | fitter = SansFitMC(localfit=localfit, start_points=1) |
---|
413 | if handler == None: |
---|
414 | handler = fitresult.ConsoleUpdate(improvement_delta=0.1) |
---|
415 | |
---|
416 | result_list = [] |
---|
417 | try: |
---|
418 | result = fit.fit(self.problem, fitter=fitter, handler=handler) |
---|
419 | self.problem.all_results(result) |
---|
420 | |
---|
421 | except LinAlgError: |
---|
422 | raise ValueError, "SVD did not converge" |
---|
423 | |
---|
424 | for m in self.problem.parts: |
---|
425 | residuals, theory = m.fitness.residuals() |
---|
426 | small_result = FResult(model=m.model, data=m.data.sans_data) |
---|
427 | small_result.fitter_id = self.fitter_id |
---|
428 | small_result.theory = theory |
---|
429 | small_result.residuals = residuals |
---|
430 | small_result.pvec = [] |
---|
431 | small_result.cov = [] |
---|
432 | small_result.stderr = [] |
---|
433 | small_result.param_list = [] |
---|
434 | small_result.residuals = m.residuals |
---|
435 | if result is not None: |
---|
436 | for p in result.parameters: |
---|
437 | if p.data.name == small_result.data.name and \ |
---|
438 | p.model.name == small_result.model.name: |
---|
439 | small_result.index = m.data.idx |
---|
440 | small_result.fitness = result.fitness |
---|
441 | small_result.pvec.append(p.value) |
---|
442 | small_result.stderr.append(p.stderr) |
---|
443 | name_split = p.name.split('.') |
---|
444 | name = name_split[1].strip() |
---|
445 | if len(name_split) > 2: |
---|
446 | name += '.' + name_split[2].strip() |
---|
447 | small_result.param_list.append(name) |
---|
448 | result_list.append(small_result) |
---|
449 | if q != None: |
---|
450 | q.put(result_list) |
---|
451 | return q |
---|
452 | return result_list |
---|
453 | |
---|