[6fe5100] | 1 | """ |
---|
| 2 | BumpsFitting module runs the bumps optimizer. |
---|
| 3 | """ |
---|
[3e6829d] | 4 | from __future__ import division |
---|
| 5 | |
---|
[249a7c6] | 6 | import os |
---|
[35086c3] | 7 | from datetime import timedelta, datetime |
---|
| 8 | |
---|
[9a5097c] | 9 | import numpy as np |
---|
[6fe5100] | 10 | |
---|
| 11 | from bumps import fitters |
---|
[7945367] | 12 | try: |
---|
| 13 | from bumps.options import FIT_CONFIG |
---|
| 14 | # Default bumps to use the Levenberg-Marquardt optimizer |
---|
| 15 | FIT_CONFIG.selected_id = fitters.LevenbergMarquardtFit.id |
---|
| 16 | def get_fitter(): |
---|
| 17 | return FIT_CONFIG.selected_fitter, FIT_CONFIG.selected_values |
---|
| 18 | except: |
---|
| 19 | # CRUFT: Bumps changed its handling of fit options around 0.7.5.6 |
---|
| 20 | # Default bumps to use the Levenberg-Marquardt optimizer |
---|
| 21 | fitters.FIT_DEFAULT = 'lm' |
---|
| 22 | def get_fitter(): |
---|
| 23 | fitopts = fitters.FIT_OPTIONS[fitters.FIT_DEFAULT] |
---|
| 24 | return fitopts.fitclass, fitopts.options.copy() |
---|
| 25 | |
---|
| 26 | |
---|
[249a7c6] | 27 | from bumps.mapper import SerialMapper, MPMapper |
---|
[e3efa6b3] | 28 | from bumps import parameter |
---|
| 29 | from bumps.fitproblem import FitProblem |
---|
[7945367] | 30 | |
---|
[6fe5100] | 31 | |
---|
[b699768] | 32 | from sas.sascalc.fit.AbstractFitEngine import FitEngine |
---|
| 33 | from sas.sascalc.fit.AbstractFitEngine import FResult |
---|
| 34 | from sas.sascalc.fit.expression import compile_constraints |
---|
[6fe5100] | 35 | |
---|
[35086c3] | 36 | class Progress(object): |
---|
| 37 | def __init__(self, history, max_step, pars, dof): |
---|
[3e6829d] | 38 | remaining_time = int(history.time[0]*(max_step/history.step[0]-1)) |
---|
[35086c3] | 39 | # Depending on the time remaining, either display the expected |
---|
| 40 | # time of completion, or the amount of time remaining. Use precision |
---|
| 41 | # appropriate for the duration. |
---|
| 42 | if remaining_time >= 1800: |
---|
| 43 | completion_time = datetime.now() + timedelta(seconds=remaining_time) |
---|
| 44 | if remaining_time >= 36000: |
---|
| 45 | time = completion_time.strftime('%Y-%m-%d %H:%M') |
---|
| 46 | else: |
---|
| 47 | time = completion_time.strftime('%H:%M') |
---|
| 48 | else: |
---|
| 49 | if remaining_time >= 3600: |
---|
| 50 | time = '%dh %dm'%(remaining_time//3600, (remaining_time%3600)//60) |
---|
| 51 | elif remaining_time >= 60: |
---|
| 52 | time = '%dm %ds'%(remaining_time//60, remaining_time%60) |
---|
| 53 | else: |
---|
| 54 | time = '%ds'%remaining_time |
---|
| 55 | chisq = "%.3g"%(2*history.value[0]/dof) |
---|
| 56 | step = "%d of %d"%(history.step[0], max_step) |
---|
| 57 | header = "=== Steps: %s chisq: %s ETA: %s\n"%(step, chisq, time) |
---|
| 58 | parameters = ["%15s: %-10.3g%s"%(k,v,("\n" if i%3==2 else " | ")) |
---|
| 59 | for i,(k,v) in enumerate(zip(pars,history.point[0]))] |
---|
| 60 | self.msg = "".join([header]+parameters) |
---|
| 61 | |
---|
| 62 | def __str__(self): |
---|
| 63 | return self.msg |
---|
| 64 | |
---|
| 65 | |
---|
[85f17f6] | 66 | class BumpsMonitor(object): |
---|
[35086c3] | 67 | def __init__(self, handler, max_step, pars, dof): |
---|
[85f17f6] | 68 | self.handler = handler |
---|
| 69 | self.max_step = max_step |
---|
[35086c3] | 70 | self.pars = pars |
---|
| 71 | self.dof = dof |
---|
[ed4aef2] | 72 | |
---|
[85f17f6] | 73 | def config_history(self, history): |
---|
| 74 | history.requires(time=1, value=2, point=1, step=1) |
---|
[ed4aef2] | 75 | |
---|
[85f17f6] | 76 | def __call__(self, history): |
---|
[e3efa6b3] | 77 | if self.handler is None: return |
---|
[35086c3] | 78 | self.handler.set_result(Progress(history, self.max_step, self.pars, self.dof)) |
---|
[85f17f6] | 79 | self.handler.progress(history.step[0], self.max_step) |
---|
| 80 | if len(history.step)>1 and history.step[1] > history.step[0]: |
---|
| 81 | self.handler.improvement() |
---|
| 82 | self.handler.update_fit() |
---|
| 83 | |
---|
[ed4aef2] | 84 | class ConvergenceMonitor(object): |
---|
| 85 | """ |
---|
| 86 | ConvergenceMonitor contains population summary statistics to show progress |
---|
| 87 | of the fit. This is a list [ (best, 0%, 25%, 50%, 75%, 100%) ] or |
---|
| 88 | just a list [ (best, ) ] if population size is 1. |
---|
| 89 | """ |
---|
| 90 | def __init__(self): |
---|
| 91 | self.convergence = [] |
---|
| 92 | |
---|
| 93 | def config_history(self, history): |
---|
| 94 | history.requires(value=1, population_values=1) |
---|
| 95 | |
---|
| 96 | def __call__(self, history): |
---|
| 97 | best = history.value[0] |
---|
| 98 | try: |
---|
| 99 | p = history.population_values[0] |
---|
[9a5097c] | 100 | n,p = len(p), np.sort(p) |
---|
[ed4aef2] | 101 | QI,Qmid, = int(0.2*n),int(0.5*n) |
---|
| 102 | self.convergence.append((best, p[0],p[QI],p[Qmid],p[-1-QI],p[-1])) |
---|
| 103 | except: |
---|
[e3efa6b3] | 104 | self.convergence.append((best, best,best,best,best,best)) |
---|
[ed4aef2] | 105 | |
---|
[e3efa6b3] | 106 | |
---|
[4e9f227] | 107 | # Note: currently using bumps parameters for each parameter object so that |
---|
| 108 | # a SasFitness can be used directly in bumps with the usual semantics. |
---|
| 109 | # The disadvantage of this technique is that we need to copy every parameter |
---|
| 110 | # back into the model each time the function is evaluated. We could instead |
---|
[fd5ac0d] | 111 | # define reference parameters for each sas parameter, but then we would not |
---|
[4e9f227] | 112 | # be able to express constraints using python expressions in the usual way |
---|
| 113 | # from bumps, and would instead need to use string expressions. |
---|
[e3efa6b3] | 114 | class SasFitness(object): |
---|
[6fe5100] | 115 | """ |
---|
[e3efa6b3] | 116 | Wrap SAS model as a bumps fitness object |
---|
[6fe5100] | 117 | """ |
---|
[5044543] | 118 | def __init__(self, model, data, fitted=[], constraints={}, |
---|
| 119 | initial_values=None, **kw): |
---|
[4e9f227] | 120 | self.name = model.name |
---|
| 121 | self.model = model.model |
---|
[6fe5100] | 122 | self.data = data |
---|
[9f7fbd9] | 123 | if self.data.smearer is not None: |
---|
| 124 | self.data.smearer.model = self.model |
---|
[e3efa6b3] | 125 | self._define_pars() |
---|
| 126 | self._init_pars(kw) |
---|
[5044543] | 127 | if initial_values is not None: |
---|
| 128 | self._reset_pars(fitted, initial_values) |
---|
[4e9f227] | 129 | self.constraints = dict(constraints) |
---|
[e3efa6b3] | 130 | self.set_fitted(fitted) |
---|
[5044543] | 131 | self.update() |
---|
| 132 | |
---|
| 133 | def _reset_pars(self, names, values): |
---|
| 134 | for k,v in zip(names, values): |
---|
| 135 | self._pars[k].value = v |
---|
[e3efa6b3] | 136 | |
---|
| 137 | def _define_pars(self): |
---|
| 138 | self._pars = {} |
---|
| 139 | for k in self.model.getParamList(): |
---|
| 140 | name = ".".join((self.name,k)) |
---|
| 141 | value = self.model.getParam(k) |
---|
| 142 | bounds = self.model.details.get(k,["",None,None])[1:3] |
---|
| 143 | self._pars[k] = parameter.Parameter(value=value, bounds=bounds, |
---|
| 144 | fixed=True, name=name) |
---|
[4e9f227] | 145 | #print parameter.summarize(self._pars.values()) |
---|
[e3efa6b3] | 146 | |
---|
| 147 | def _init_pars(self, kw): |
---|
| 148 | for k,v in kw.items(): |
---|
| 149 | # dispersion parameters initialized with _field instead of .field |
---|
| 150 | if k.endswith('_width'): k = k[:-6]+'.width' |
---|
| 151 | elif k.endswith('_npts'): k = k[:-5]+'.npts' |
---|
| 152 | elif k.endswith('_nsigmas'): k = k[:-7]+'.nsigmas' |
---|
| 153 | elif k.endswith('_type'): k = k[:-5]+'.type' |
---|
| 154 | if k not in self._pars: |
---|
| 155 | formatted_pars = ", ".join(sorted(self._pars.keys())) |
---|
| 156 | raise KeyError("invalid parameter %r for %s--use one of: %s" |
---|
| 157 | %(k, self.model, formatted_pars)) |
---|
| 158 | if '.' in k and not k.endswith('.width'): |
---|
| 159 | self.model.setParam(k, v) |
---|
| 160 | elif isinstance(v, parameter.BaseParameter): |
---|
| 161 | self._pars[k] = v |
---|
| 162 | elif isinstance(v, (tuple,list)): |
---|
| 163 | low, high = v |
---|
| 164 | self._pars[k].value = (low+high)/2 |
---|
| 165 | self._pars[k].range(low,high) |
---|
[95d58d3] | 166 | else: |
---|
[e3efa6b3] | 167 | self._pars[k].value = v |
---|
| 168 | |
---|
| 169 | def set_fitted(self, param_list): |
---|
[6fe5100] | 170 | """ |
---|
[e3efa6b3] | 171 | Flag a set of parameters as fitted parameters. |
---|
[6fe5100] | 172 | """ |
---|
[e3efa6b3] | 173 | for k,p in self._pars.items(): |
---|
[4e9f227] | 174 | p.fixed = (k not in param_list or k in self.constraints) |
---|
[4a0dc427] | 175 | self.fitted_par_names = [k for k in param_list if k not in self.constraints] |
---|
[bf5e985] | 176 | self.computed_par_names = [k for k in param_list if k in self.constraints] |
---|
| 177 | self.fitted_pars = [self._pars[k] for k in self.fitted_par_names] |
---|
| 178 | self.computed_pars = [self._pars[k] for k in self.computed_par_names] |
---|
[6fe5100] | 179 | |
---|
[e3efa6b3] | 180 | # ===== Fitness interface ==== |
---|
| 181 | def parameters(self): |
---|
| 182 | return self._pars |
---|
[6fe5100] | 183 | |
---|
[e3efa6b3] | 184 | def update(self): |
---|
| 185 | for k,v in self._pars.items(): |
---|
[4e9f227] | 186 | #print "updating",k,v,v.value |
---|
[e3efa6b3] | 187 | self.model.setParam(k,v.value) |
---|
| 188 | self._dirty = True |
---|
[6fe5100] | 189 | |
---|
[e3efa6b3] | 190 | def _recalculate(self): |
---|
| 191 | if self._dirty: |
---|
[9f7fbd9] | 192 | self._residuals, self._theory \ |
---|
| 193 | = self.data.residuals(self.model.evalDistribution) |
---|
[e3efa6b3] | 194 | self._dirty = False |
---|
[6fe5100] | 195 | |
---|
[e3efa6b3] | 196 | def numpoints(self): |
---|
[9a5097c] | 197 | return np.sum(self.data.idx) # number of fitted points |
---|
[6fe5100] | 198 | |
---|
[e3efa6b3] | 199 | def nllf(self): |
---|
[9a5097c] | 200 | return 0.5*np.sum(self.residuals()**2) |
---|
[e3efa6b3] | 201 | |
---|
| 202 | def theory(self): |
---|
| 203 | self._recalculate() |
---|
| 204 | return self._theory |
---|
| 205 | |
---|
| 206 | def residuals(self): |
---|
| 207 | self._recalculate() |
---|
| 208 | return self._residuals |
---|
| 209 | |
---|
| 210 | # Not implementing the data methods for now: |
---|
| 211 | # |
---|
| 212 | # resynth_data/restore_data/save/plot |
---|
[6fe5100] | 213 | |
---|
[191c648] | 214 | class ParameterExpressions(object): |
---|
| 215 | def __init__(self, models): |
---|
| 216 | self.models = models |
---|
| 217 | self._setup() |
---|
| 218 | |
---|
| 219 | def _setup(self): |
---|
| 220 | exprs = {} |
---|
| 221 | for M in self.models: |
---|
| 222 | exprs.update((".".join((M.name, k)), v) for k, v in M.constraints.items()) |
---|
| 223 | if exprs: |
---|
| 224 | symtab = dict((".".join((M.name, k)), p) |
---|
| 225 | for M in self.models |
---|
| 226 | for k,p in M.parameters().items()) |
---|
| 227 | self.update = compile_constraints(symtab, exprs) |
---|
| 228 | else: |
---|
| 229 | self.update = lambda: 0 |
---|
| 230 | |
---|
| 231 | def __call__(self): |
---|
| 232 | self.update() |
---|
| 233 | |
---|
| 234 | def __getstate__(self): |
---|
| 235 | return self.models |
---|
| 236 | |
---|
| 237 | def __setstate__(self, state): |
---|
| 238 | self.models = state |
---|
| 239 | self._setup() |
---|
| 240 | |
---|
[6fe5100] | 241 | class BumpsFit(FitEngine): |
---|
| 242 | """ |
---|
| 243 | Fit a model using bumps. |
---|
| 244 | """ |
---|
| 245 | def __init__(self): |
---|
| 246 | """ |
---|
| 247 | Creates a dictionary (self.fit_arrange_dict={})of FitArrange elements |
---|
| 248 | with Uid as keys |
---|
| 249 | """ |
---|
| 250 | FitEngine.__init__(self) |
---|
| 251 | self.curr_thread = None |
---|
| 252 | |
---|
| 253 | def fit(self, msg_q=None, |
---|
| 254 | q=None, handler=None, curr_thread=None, |
---|
| 255 | ftol=1.49012e-8, reset_flag=False): |
---|
[e3efa6b3] | 256 | # Build collection of bumps fitness calculators |
---|
[bf5e985] | 257 | models = [SasFitness(model=M.get_model(), |
---|
| 258 | data=M.get_data(), |
---|
| 259 | constraints=M.constraints, |
---|
[5044543] | 260 | fitted=M.pars, |
---|
| 261 | initial_values=M.vals if reset_flag else None) |
---|
[bf5e985] | 262 | for M in self.fit_arrange_dict.values() |
---|
| 263 | if M.get_to_fit()] |
---|
[233c121] | 264 | if len(models) == 0: |
---|
| 265 | raise RuntimeError("Nothing to fit") |
---|
[e3efa6b3] | 266 | problem = FitProblem(models) |
---|
| 267 | |
---|
[191c648] | 268 | # TODO: need better handling of parameter expressions and bounds constraints |
---|
| 269 | # so that they are applied during polydispersity calculations. This |
---|
| 270 | # will remove the immediate need for the setp_hook in bumps, though |
---|
| 271 | # bumps may still need something similar, such as a sane class structure |
---|
| 272 | # which allows a subclass to override setp. |
---|
| 273 | problem.setp_hook = ParameterExpressions(models) |
---|
[4e9f227] | 274 | |
---|
[e3efa6b3] | 275 | # Run the fit |
---|
| 276 | result = run_bumps(problem, handler, curr_thread) |
---|
[6fe5100] | 277 | if handler is not None: |
---|
| 278 | handler.update_fit(last=True) |
---|
[e3efa6b3] | 279 | |
---|
[eff93b8] | 280 | # TODO: shouldn't reference internal parameters of fit problem |
---|
[e3efa6b3] | 281 | varying = problem._parameters |
---|
| 282 | # collect the results |
---|
| 283 | all_results = [] |
---|
| 284 | for M in problem.models: |
---|
| 285 | fitness = M.fitness |
---|
| 286 | fitted_index = [varying.index(p) for p in fitness.fitted_pars] |
---|
[e1442d4] | 287 | param_list = fitness.fitted_par_names + fitness.computed_par_names |
---|
[e3efa6b3] | 288 | R = FResult(model=fitness.model, data=fitness.data, |
---|
[e1442d4] | 289 | param_list=param_list) |
---|
[e3efa6b3] | 290 | R.theory = fitness.theory() |
---|
| 291 | R.residuals = fitness.residuals() |
---|
[5044543] | 292 | R.index = fitness.data.idx |
---|
[e3efa6b3] | 293 | R.fitter_id = self.fitter_id |
---|
[eff93b8] | 294 | # TODO: should scale stderr by sqrt(chisq/DOF) if dy is unknown |
---|
[e3efa6b3] | 295 | R.success = result['success'] |
---|
[e1442d4] | 296 | if R.success: |
---|
[9a5097c] | 297 | R.stderr = np.hstack((result['stderr'][fitted_index], |
---|
| 298 | np.NaN*np.ones(len(fitness.computed_pars)))) |
---|
| 299 | R.pvec = np.hstack((result['value'][fitted_index], |
---|
[e1442d4] | 300 | [p.value for p in fitness.computed_pars])) |
---|
[9a5097c] | 301 | R.fitness = np.sum(R.residuals**2)/(fitness.numpoints() - len(fitted_index)) |
---|
[e1442d4] | 302 | else: |
---|
[9a5097c] | 303 | R.stderr = np.NaN*np.ones(len(param_list)) |
---|
| 304 | R.pvec = np.asarray( [p.value for p in fitness.fitted_pars+fitness.computed_pars]) |
---|
| 305 | R.fitness = np.NaN |
---|
[e3efa6b3] | 306 | R.convergence = result['convergence'] |
---|
| 307 | if result['uncertainty'] is not None: |
---|
| 308 | R.uncertainty_state = result['uncertainty'] |
---|
| 309 | all_results.append(R) |
---|
| 310 | |
---|
[6fe5100] | 311 | if q is not None: |
---|
[e3efa6b3] | 312 | q.put(all_results) |
---|
[6fe5100] | 313 | return q |
---|
[e3efa6b3] | 314 | else: |
---|
| 315 | return all_results |
---|
[6fe5100] | 316 | |
---|
[e3efa6b3] | 317 | def run_bumps(problem, handler, curr_thread): |
---|
[85f17f6] | 318 | def abort_test(): |
---|
| 319 | if curr_thread is None: return False |
---|
| 320 | try: curr_thread.isquit() |
---|
| 321 | except KeyboardInterrupt: |
---|
| 322 | if handler is not None: |
---|
| 323 | handler.stop("Fitting: Terminated!!!") |
---|
| 324 | return True |
---|
| 325 | return False |
---|
| 326 | |
---|
[7945367] | 327 | fitclass, options = get_fitter() |
---|
| 328 | steps = options.get('steps', 0) |
---|
| 329 | if steps == 0: |
---|
| 330 | pop = options.get('pop',0)*len(problem._parameters) |
---|
| 331 | samples = options.get('samples', 0) |
---|
| 332 | steps = (samples+pop-1)/pop if pop != 0 else samples |
---|
| 333 | max_step = steps + options.get('burn', 0) |
---|
[35086c3] | 334 | pars = [p.name for p in problem._parameters] |
---|
[9a5097c] | 335 | #x0 = np.asarray([p.value for p in problem._parameters]) |
---|
[e3efa6b3] | 336 | options['monitors'] = [ |
---|
[35086c3] | 337 | BumpsMonitor(handler, max_step, pars, problem.dof), |
---|
[e3efa6b3] | 338 | ConvergenceMonitor(), |
---|
| 339 | ] |
---|
[95d58d3] | 340 | fitdriver = fitters.FitDriver(fitclass, problem=problem, |
---|
[042f065] | 341 | abort_test=abort_test, **options) |
---|
[249a7c6] | 342 | omp_threads = int(os.environ.get('OMP_NUM_THREADS','0')) |
---|
[e1442d4] | 343 | mapper = MPMapper if omp_threads == 1 else SerialMapper |
---|
[6fe5100] | 344 | fitdriver.mapper = mapper.start_mapper(problem, None) |
---|
[233c121] | 345 | #import time; T0 = time.time() |
---|
[6fe5100] | 346 | try: |
---|
| 347 | best, fbest = fitdriver.fit() |
---|
| 348 | except: |
---|
| 349 | import traceback; traceback.print_exc() |
---|
| 350 | raise |
---|
[95d58d3] | 351 | finally: |
---|
| 352 | mapper.stop_mapper(fitdriver.mapper) |
---|
[e3efa6b3] | 353 | |
---|
| 354 | |
---|
| 355 | convergence_list = options['monitors'][-1].convergence |
---|
[9a5097c] | 356 | convergence = (2*np.asarray(convergence_list)/problem.dof |
---|
| 357 | if convergence_list else np.empty((0,1),'d')) |
---|
[e1442d4] | 358 | |
---|
| 359 | success = best is not None |
---|
[e3efa6b3] | 360 | return { |
---|
[e1442d4] | 361 | 'value': best if success else None, |
---|
| 362 | 'stderr': fitdriver.stderr() if success else None, |
---|
| 363 | 'success': success, |
---|
[e3efa6b3] | 364 | 'convergence': convergence, |
---|
| 365 | 'uncertainty': getattr(fitdriver.fitter, 'state', None), |
---|
| 366 | } |
---|
[6fe5100] | 367 | |
---|