| 1 | """ |
|---|
| 2 | BumpsFitting module runs the bumps optimizer. |
|---|
| 3 | """ |
|---|
| 4 | import sys |
|---|
| 5 | |
|---|
| 6 | import numpy |
|---|
| 7 | |
|---|
| 8 | from bumps import fitters |
|---|
| 9 | from bumps.mapper import SerialMapper |
|---|
| 10 | |
|---|
| 11 | from sans.fit.AbstractFitEngine import FitEngine |
|---|
| 12 | from sans.fit.AbstractFitEngine import FResult |
|---|
| 13 | |
|---|
| 14 | class SasProblem(object): |
|---|
| 15 | """ |
|---|
| 16 | Wrap the SAS model in a form that can be understood by bumps. |
|---|
| 17 | """ |
|---|
| 18 | def __init__(self, param_list, model=None, data=None, fitresult=None, |
|---|
| 19 | handler=None, curr_thread=None, msg_q=None): |
|---|
| 20 | """ |
|---|
| 21 | :param Model: the model wrapper fro sans -model |
|---|
| 22 | :param Data: the data wrapper for sans data |
|---|
| 23 | """ |
|---|
| 24 | self.model = model |
|---|
| 25 | self.data = data |
|---|
| 26 | self.param_list = param_list |
|---|
| 27 | self.msg_q = msg_q |
|---|
| 28 | self.curr_thread = curr_thread |
|---|
| 29 | self.handler = handler |
|---|
| 30 | self.fitresult = fitresult |
|---|
| 31 | self.res = [] |
|---|
| 32 | self.func_name = "Functor" |
|---|
| 33 | self.theory = None |
|---|
| 34 | self.name = "Fill in proper name!" |
|---|
| 35 | |
|---|
| 36 | @property |
|---|
| 37 | def dof(self): |
|---|
| 38 | return self.data.num_points - len(self.param_list) |
|---|
| 39 | |
|---|
| 40 | def summarize(self): |
|---|
| 41 | """ |
|---|
| 42 | Return a stylized list of parameter names and values with range bars |
|---|
| 43 | suitable for printing. |
|---|
| 44 | """ |
|---|
| 45 | output = [] |
|---|
| 46 | bounds = self.bounds() |
|---|
| 47 | for i,p in enumerate(self.getp()): |
|---|
| 48 | name = self.param_list[i] |
|---|
| 49 | low,high = bounds[:,i] |
|---|
| 50 | range = ",".join((("[%g"%low if numpy.isfinite(low) else "(-inf"), |
|---|
| 51 | ("%g]"%high if numpy.isfinite(high) else "inf)"))) |
|---|
| 52 | if not numpy.isfinite(p): |
|---|
| 53 | bar = "*invalid* " |
|---|
| 54 | else: |
|---|
| 55 | bar = ['.']*10 |
|---|
| 56 | if numpy.isfinite(high-low): |
|---|
| 57 | position = int(9.999999999 * float(p-low)/float(high-low)) |
|---|
| 58 | if position < 0: bar[0] = '<' |
|---|
| 59 | elif position > 9: bar[9] = '>' |
|---|
| 60 | else: bar[position] = '|' |
|---|
| 61 | bar = "".join(bar) |
|---|
| 62 | output.append("%40s %s %10g in %s"%(name,bar,p,range)) |
|---|
| 63 | return "\n".join(output) |
|---|
| 64 | |
|---|
| 65 | def nllf(self, p=None): |
|---|
| 66 | residuals = self.residuals(p) |
|---|
| 67 | return 0.5*numpy.sum(residuals**2) |
|---|
| 68 | |
|---|
| 69 | def setp(self, p): |
|---|
| 70 | for k,v in zip(self.param_list, p): |
|---|
| 71 | self.model.setParam(k,v) |
|---|
| 72 | #self.model.set_params(self.param_list, params) |
|---|
| 73 | |
|---|
| 74 | def getp(self): |
|---|
| 75 | return numpy.array([self.model.getParam(k) for k in self.param_list]) |
|---|
| 76 | #return numpy.asarray(self.model.get_params(self.param_list)) |
|---|
| 77 | |
|---|
| 78 | def bounds(self): |
|---|
| 79 | return numpy.array([self._getrange(p) for p in self.param_list]).T |
|---|
| 80 | |
|---|
| 81 | def labels(self): |
|---|
| 82 | return self.param_list |
|---|
| 83 | |
|---|
| 84 | def _getrange(self, p): |
|---|
| 85 | """ |
|---|
| 86 | Override _getrange of park parameter |
|---|
| 87 | return the range of parameter |
|---|
| 88 | """ |
|---|
| 89 | lo, hi = self.model.details[p][1:3] |
|---|
| 90 | if lo is None: lo = -numpy.inf |
|---|
| 91 | if hi is None: hi = numpy.inf |
|---|
| 92 | return lo, hi |
|---|
| 93 | |
|---|
| 94 | def randomize(self, n): |
|---|
| 95 | p = self.getp() |
|---|
| 96 | # since randn is symmetric and random, doesn't matter |
|---|
| 97 | # point value is negative. |
|---|
| 98 | # TODO: throw in bounds checking! |
|---|
| 99 | return numpy.random.randn(n, len(self.param_list))*p + p |
|---|
| 100 | |
|---|
| 101 | def chisq(self): |
|---|
| 102 | """ |
|---|
| 103 | Calculates chi^2 |
|---|
| 104 | |
|---|
| 105 | :param params: list of parameter values |
|---|
| 106 | |
|---|
| 107 | :return: chi^2 |
|---|
| 108 | |
|---|
| 109 | """ |
|---|
| 110 | return numpy.sum(self.res**2)/self.dof |
|---|
| 111 | |
|---|
| 112 | def residuals(self, params=None): |
|---|
| 113 | """ |
|---|
| 114 | Compute residuals |
|---|
| 115 | :param params: value of parameters to fit |
|---|
| 116 | """ |
|---|
| 117 | if params is not None: self.setp(params) |
|---|
| 118 | #import thread |
|---|
| 119 | #print "params", params |
|---|
| 120 | self.res, self.theory = self.data.residuals(self.model.evalDistribution) |
|---|
| 121 | |
|---|
| 122 | # TODO: this belongs in monitor not residuals calculation |
|---|
| 123 | if False: # self.fitresult is not None: |
|---|
| 124 | #self.fitresult.set_model(model=self.model) |
|---|
| 125 | self.fitresult.residuals = self.res+0 |
|---|
| 126 | self.fitresult.iterations += 1 |
|---|
| 127 | self.fitresult.theory = self.theory+0 |
|---|
| 128 | |
|---|
| 129 | #fitness = self.chisq(params=params) |
|---|
| 130 | fitness = self.chisq() |
|---|
| 131 | self.fitresult.p = params |
|---|
| 132 | self.fitresult.set_fitness(fitness=fitness) |
|---|
| 133 | if self.msg_q is not None: |
|---|
| 134 | self.msg_q.put(self.fitresult) |
|---|
| 135 | |
|---|
| 136 | if self.handler is not None: |
|---|
| 137 | self.handler.set_result(result=self.fitresult) |
|---|
| 138 | self.handler.update_fit() |
|---|
| 139 | |
|---|
| 140 | if self.curr_thread != None: |
|---|
| 141 | try: |
|---|
| 142 | self.curr_thread.isquit() |
|---|
| 143 | except: |
|---|
| 144 | #msg = "Fitting: Terminated... Note: Forcing to stop " |
|---|
| 145 | #msg += "fitting may cause a 'Functor error message' " |
|---|
| 146 | #msg += "being recorded in the log file....." |
|---|
| 147 | #self.handler.stop(msg) |
|---|
| 148 | raise |
|---|
| 149 | |
|---|
| 150 | return self.res |
|---|
| 151 | __call__ = residuals |
|---|
| 152 | |
|---|
| 153 | def _DEAD_check_param_range(self): |
|---|
| 154 | """ |
|---|
| 155 | Check the lower and upper bound of the parameter value |
|---|
| 156 | and set res to the inf if the value is outside of the |
|---|
| 157 | range |
|---|
| 158 | :limitation: the initial values must be within range. |
|---|
| 159 | """ |
|---|
| 160 | |
|---|
| 161 | #time.sleep(0.01) |
|---|
| 162 | is_outofbound = False |
|---|
| 163 | # loop through the fit parameters |
|---|
| 164 | model = self.model |
|---|
| 165 | for p in self.param_list: |
|---|
| 166 | value = model.getParam(p) |
|---|
| 167 | low,high = model.details[p][1:3] |
|---|
| 168 | if low is not None and numpy.isfinite(low): |
|---|
| 169 | if p.value == 0: |
|---|
| 170 | # This value works on Scipy |
|---|
| 171 | # Do not change numbers below |
|---|
| 172 | value = _SMALLVALUE |
|---|
| 173 | # For leastsq, it needs a bit step back from the boundary |
|---|
| 174 | val = low - value * _SMALLVALUE |
|---|
| 175 | if value < val: |
|---|
| 176 | self.res *= 1e+6 |
|---|
| 177 | is_outofbound = True |
|---|
| 178 | break |
|---|
| 179 | if high is not None and numpy.isfinite(high): |
|---|
| 180 | # This value works on Scipy |
|---|
| 181 | # Do not change numbers below |
|---|
| 182 | if value == 0: |
|---|
| 183 | value = _SMALLVALUE |
|---|
| 184 | # For leastsq, it needs a bit step back from the boundary |
|---|
| 185 | val = high + value * _SMALLVALUE |
|---|
| 186 | if value > val: |
|---|
| 187 | self.res *= 1e+6 |
|---|
| 188 | is_outofbound = True |
|---|
| 189 | break |
|---|
| 190 | |
|---|
| 191 | return is_outofbound |
|---|
| 192 | |
|---|
| 193 | class BumpsFit(FitEngine): |
|---|
| 194 | """ |
|---|
| 195 | Fit a model using bumps. |
|---|
| 196 | """ |
|---|
| 197 | def __init__(self): |
|---|
| 198 | """ |
|---|
| 199 | Creates a dictionary (self.fit_arrange_dict={})of FitArrange elements |
|---|
| 200 | with Uid as keys |
|---|
| 201 | """ |
|---|
| 202 | FitEngine.__init__(self) |
|---|
| 203 | self.curr_thread = None |
|---|
| 204 | |
|---|
| 205 | def fit(self, msg_q=None, |
|---|
| 206 | q=None, handler=None, curr_thread=None, |
|---|
| 207 | ftol=1.49012e-8, reset_flag=False): |
|---|
| 208 | """ |
|---|
| 209 | """ |
|---|
| 210 | fitproblem = [] |
|---|
| 211 | for fproblem in self.fit_arrange_dict.itervalues(): |
|---|
| 212 | if fproblem.get_to_fit() == 1: |
|---|
| 213 | fitproblem.append(fproblem) |
|---|
| 214 | if len(fitproblem) > 1 : |
|---|
| 215 | msg = "Bumps can't fit more than a single fit problem at a time." |
|---|
| 216 | raise RuntimeError, msg |
|---|
| 217 | elif len(fitproblem) == 0 : |
|---|
| 218 | raise RuntimeError, "No problem scheduled for fitting." |
|---|
| 219 | model = fitproblem[0].get_model() |
|---|
| 220 | if reset_flag: |
|---|
| 221 | # reset the initial value; useful for batch |
|---|
| 222 | for name in fitproblem[0].pars: |
|---|
| 223 | ind = fitproblem[0].pars.index(name) |
|---|
| 224 | model.setParam(name, fitproblem[0].vals[ind]) |
|---|
| 225 | listdata = fitproblem[0].get_data() |
|---|
| 226 | # Concatenate dList set (contains one or more data)before fitting |
|---|
| 227 | data = listdata |
|---|
| 228 | |
|---|
| 229 | self.curr_thread = curr_thread |
|---|
| 230 | |
|---|
| 231 | result = FResult(model=model, data=data, param_list=self.param_list) |
|---|
| 232 | result.pars = fitproblem[0].pars |
|---|
| 233 | result.fitter_id = self.fitter_id |
|---|
| 234 | result.index = data.idx |
|---|
| 235 | if handler is not None: |
|---|
| 236 | handler.set_result(result=result) |
|---|
| 237 | problem = SasProblem(param_list=self.param_list, |
|---|
| 238 | model=model.model, |
|---|
| 239 | data=data, |
|---|
| 240 | handler=handler, |
|---|
| 241 | fitresult=result, |
|---|
| 242 | curr_thread=curr_thread, |
|---|
| 243 | msg_q=msg_q) |
|---|
| 244 | try: |
|---|
| 245 | run_bumps(problem, result, ftol) |
|---|
| 246 | #run_scipy(problem, result, ftol) |
|---|
| 247 | except: |
|---|
| 248 | if hasattr(sys, 'last_type') and sys.last_type == KeyboardInterrupt: |
|---|
| 249 | if handler is not None: |
|---|
| 250 | msg = "Fitting: Terminated!!!" |
|---|
| 251 | handler.stop(msg) |
|---|
| 252 | raise KeyboardInterrupt, msg |
|---|
| 253 | else: |
|---|
| 254 | raise |
|---|
| 255 | |
|---|
| 256 | if handler is not None: |
|---|
| 257 | handler.set_result(result=result) |
|---|
| 258 | handler.update_fit(last=True) |
|---|
| 259 | if q is not None: |
|---|
| 260 | q.put(result) |
|---|
| 261 | return q |
|---|
| 262 | #if success < 1 or success > 5: |
|---|
| 263 | # result.fitness = None |
|---|
| 264 | return [result] |
|---|
| 265 | |
|---|
| 266 | def run_bumps(problem, result, ftol): |
|---|
| 267 | fitopts = fitters.FIT_OPTIONS[fitters.FIT_DEFAULT] |
|---|
| 268 | fitclass = fitopts.fitclass |
|---|
| 269 | options = fitopts.options.copy() |
|---|
| 270 | options['ftol'] = ftol |
|---|
| 271 | fitdriver = fitters.FitDriver(fitclass, problem=problem, |
|---|
| 272 | abort_test=lambda: False, **options) |
|---|
| 273 | mapper = SerialMapper |
|---|
| 274 | fitdriver.mapper = mapper.start_mapper(problem, None) |
|---|
| 275 | try: |
|---|
| 276 | best, fbest = fitdriver.fit() |
|---|
| 277 | except: |
|---|
| 278 | import traceback; traceback.print_exc() |
|---|
| 279 | raise |
|---|
| 280 | finally: |
|---|
| 281 | mapper.stop_mapper(fitdriver.mapper) |
|---|
| 282 | #print "best,fbest",best,fbest,problem.dof |
|---|
| 283 | result.fitness = 2*fbest/problem.dof |
|---|
| 284 | #print "fitness",result.fitness |
|---|
| 285 | result.stderr = fitdriver.stderr() |
|---|
| 286 | result.pvec = best |
|---|
| 287 | # TODO: track success better |
|---|
| 288 | result.success = True |
|---|
| 289 | result.theory = problem.theory |
|---|
| 290 | |
|---|
| 291 | def run_scipy(model, result, ftol): |
|---|
| 292 | # This import must be here; otherwise it will be confused when more |
|---|
| 293 | # than one thread exist. |
|---|
| 294 | from scipy import optimize |
|---|
| 295 | |
|---|
| 296 | out, cov_x, _, mesg, success = optimize.leastsq(model.residuals, |
|---|
| 297 | model.getp(), |
|---|
| 298 | ftol=ftol, |
|---|
| 299 | full_output=1) |
|---|
| 300 | if cov_x is not None and numpy.isfinite(cov_x).all(): |
|---|
| 301 | stderr = numpy.sqrt(numpy.diag(cov_x)) |
|---|
| 302 | else: |
|---|
| 303 | stderr = [] |
|---|
| 304 | result.fitness = model.chisq() |
|---|
| 305 | result.stderr = stderr |
|---|
| 306 | result.pvec = out |
|---|
| 307 | result.success = success |
|---|
| 308 | result.theory = model.theory |
|---|
| 309 | |
|---|
