[a10364b] | 1 | """ |
---|
| 2 | ScipyFitting module contains FitArrange , ScipyFit, |
---|
| 3 | Parameter classes.All listed classes work together to perform a |
---|
| 4 | simple fit with scipy optimizer. |
---|
| 5 | """ |
---|
| 6 | import sys |
---|
| 7 | import copy |
---|
| 8 | |
---|
| 9 | import numpy |
---|
| 10 | |
---|
| 11 | from sas.fit.AbstractFitEngine import FitEngine |
---|
| 12 | from sas.fit.AbstractFitEngine import FResult |
---|
| 13 | |
---|
| 14 | _SMALLVALUE = 1.0e-10 |
---|
| 15 | |
---|
| 16 | class SasAssembly: |
---|
| 17 | """ |
---|
| 18 | Sas Assembly class a class wrapper to be call in optimizer.leastsq method |
---|
| 19 | """ |
---|
| 20 | def __init__(self, paramlist, model=None, data=None, fitresult=None, |
---|
| 21 | handler=None, curr_thread=None, msg_q=None): |
---|
| 22 | """ |
---|
| 23 | :param Model: the model wrapper fro sas -model |
---|
| 24 | :param Data: the data wrapper for sas data |
---|
| 25 | |
---|
| 26 | """ |
---|
| 27 | self.model = model |
---|
| 28 | self.data = data |
---|
| 29 | self.paramlist = paramlist |
---|
| 30 | self.msg_q = msg_q |
---|
| 31 | self.curr_thread = curr_thread |
---|
| 32 | self.handler = handler |
---|
| 33 | self.fitresult = fitresult |
---|
| 34 | self.res = [] |
---|
| 35 | self.true_res = [] |
---|
| 36 | self.func_name = "Functor" |
---|
| 37 | self.theory = None |
---|
| 38 | |
---|
| 39 | def chisq(self): |
---|
| 40 | """ |
---|
| 41 | Calculates chi^2 |
---|
| 42 | |
---|
| 43 | :param params: list of parameter values |
---|
| 44 | |
---|
| 45 | :return: chi^2 |
---|
| 46 | |
---|
| 47 | """ |
---|
| 48 | total = 0 |
---|
| 49 | for item in self.true_res: |
---|
| 50 | total += item * item |
---|
| 51 | if len(self.true_res) == 0: |
---|
| 52 | return None |
---|
| 53 | return total / (len(self.true_res) - len(self.paramlist)) |
---|
| 54 | |
---|
| 55 | def __call__(self, params): |
---|
| 56 | """ |
---|
| 57 | Compute residuals |
---|
| 58 | :param params: value of parameters to fit |
---|
| 59 | """ |
---|
| 60 | #import thread |
---|
| 61 | self.model.set_params(self.paramlist, params) |
---|
| 62 | #print "params", params |
---|
| 63 | self.true_res, theory = self.data.residuals(self.model.eval) |
---|
| 64 | self.theory = copy.deepcopy(theory) |
---|
| 65 | # check parameters range |
---|
| 66 | if self.check_param_range(): |
---|
| 67 | # if the param value is outside of the bound |
---|
| 68 | # just silent return res = inf |
---|
| 69 | return self.res |
---|
| 70 | self.res = self.true_res |
---|
| 71 | |
---|
| 72 | if self.fitresult is not None: |
---|
| 73 | self.fitresult.set_model(model=self.model) |
---|
| 74 | self.fitresult.residuals = self.true_res |
---|
| 75 | self.fitresult.iterations += 1 |
---|
| 76 | self.fitresult.theory = theory |
---|
| 77 | |
---|
| 78 | #fitness = self.chisq(params=params) |
---|
| 79 | fitness = self.chisq() |
---|
| 80 | self.fitresult.pvec = params |
---|
| 81 | self.fitresult.set_fitness(fitness=fitness) |
---|
| 82 | if self.msg_q is not None: |
---|
| 83 | self.msg_q.put(self.fitresult) |
---|
| 84 | |
---|
| 85 | if self.handler is not None: |
---|
| 86 | self.handler.set_result(result=self.fitresult) |
---|
| 87 | self.handler.update_fit() |
---|
| 88 | |
---|
| 89 | if self.curr_thread != None: |
---|
| 90 | try: |
---|
| 91 | self.curr_thread.isquit() |
---|
| 92 | except: |
---|
| 93 | #msg = "Fitting: Terminated... Note: Forcing to stop " |
---|
| 94 | #msg += "fitting may cause a 'Functor error message' " |
---|
| 95 | #msg += "being recorded in the log file....." |
---|
| 96 | #self.handler.stop(msg) |
---|
| 97 | raise |
---|
| 98 | |
---|
| 99 | return self.res |
---|
| 100 | |
---|
| 101 | def check_param_range(self): |
---|
| 102 | """ |
---|
| 103 | Check the lower and upper bound of the parameter value |
---|
| 104 | and set res to the inf if the value is outside of the |
---|
| 105 | range |
---|
| 106 | :limitation: the initial values must be within range. |
---|
| 107 | """ |
---|
| 108 | |
---|
| 109 | #time.sleep(0.01) |
---|
| 110 | is_outofbound = False |
---|
| 111 | # loop through the fit parameters |
---|
| 112 | model = self.model.model |
---|
| 113 | for p in self.paramlist: |
---|
| 114 | value = model.getParam(p) |
---|
| 115 | low,high = model.details[p][1:3] |
---|
| 116 | if low is not None and numpy.isfinite(low): |
---|
| 117 | if value == 0: |
---|
| 118 | # This value works on Scipy |
---|
| 119 | # Do not change numbers below |
---|
| 120 | value = _SMALLVALUE |
---|
| 121 | # For leastsq, it needs a bit step back from the boundary |
---|
| 122 | val = low - value * _SMALLVALUE |
---|
| 123 | if value < val: |
---|
| 124 | self.res *= 1e+6 |
---|
| 125 | is_outofbound = True |
---|
| 126 | break |
---|
| 127 | if high is not None and numpy.isfinite(high): |
---|
| 128 | # This value works on Scipy |
---|
| 129 | # Do not change numbers below |
---|
| 130 | if value == 0: |
---|
| 131 | value = _SMALLVALUE |
---|
| 132 | # For leastsq, it needs a bit step back from the boundary |
---|
| 133 | val = high + value * _SMALLVALUE |
---|
| 134 | if value > val: |
---|
| 135 | self.res *= 1e+6 |
---|
| 136 | is_outofbound = True |
---|
| 137 | break |
---|
| 138 | |
---|
| 139 | return is_outofbound |
---|
| 140 | |
---|
| 141 | class ScipyFit(FitEngine): |
---|
| 142 | """ |
---|
| 143 | ScipyFit performs the Fit.This class can be used as follow: |
---|
| 144 | #Do the fit SCIPY |
---|
| 145 | create an engine: engine = ScipyFit() |
---|
| 146 | Use data must be of type plottable |
---|
| 147 | Use a sas model |
---|
| 148 | |
---|
| 149 | Add data with a dictionnary of FitArrangeDict where Uid is a key and data |
---|
| 150 | is saved in FitArrange object. |
---|
| 151 | engine.set_data(data,Uid) |
---|
| 152 | |
---|
| 153 | Set model parameter "M1"= model.name add {model.parameter.name:value}. |
---|
| 154 | |
---|
| 155 | :note: Set_param() if used must always preceded set_model() |
---|
| 156 | for the fit to be performed.In case of Scipyfit set_param is called in |
---|
| 157 | fit () automatically. |
---|
| 158 | |
---|
| 159 | engine.set_param( model,"M1", {'A':2,'B':4}) |
---|
| 160 | |
---|
| 161 | Add model with a dictionnary of FitArrangeDict{} where Uid is a key and model |
---|
| 162 | is save in FitArrange object. |
---|
| 163 | engine.set_model(model,Uid) |
---|
| 164 | |
---|
| 165 | engine.fit return chisqr,[model.parameter 1,2,..],[[err1....][..err2...]] |
---|
| 166 | chisqr1, out1, cov1=engine.fit({model.parameter.name:value},qmin,qmax) |
---|
| 167 | """ |
---|
| 168 | def __init__(self): |
---|
| 169 | """ |
---|
| 170 | Creates a dictionary (self.fit_arrange_dict={})of FitArrange elements |
---|
| 171 | with Uid as keys |
---|
| 172 | """ |
---|
| 173 | FitEngine.__init__(self) |
---|
| 174 | self.curr_thread = None |
---|
| 175 | #def fit(self, *args, **kw): |
---|
| 176 | # return profile(self._fit, *args, **kw) |
---|
| 177 | |
---|
| 178 | def fit(self, msg_q=None, |
---|
| 179 | q=None, handler=None, curr_thread=None, |
---|
| 180 | ftol=1.49012e-8, reset_flag=False): |
---|
| 181 | """ |
---|
| 182 | """ |
---|
| 183 | fitproblem = [] |
---|
| 184 | for fproblem in self.fit_arrange_dict.itervalues(): |
---|
| 185 | if fproblem.get_to_fit() == 1: |
---|
| 186 | fitproblem.append(fproblem) |
---|
| 187 | if len(fitproblem) > 1 : |
---|
| 188 | msg = "Scipy can't fit more than a single fit problem at a time." |
---|
| 189 | raise RuntimeError, msg |
---|
| 190 | elif len(fitproblem) == 0 : |
---|
| 191 | raise RuntimeError, "No Assembly scheduled for Scipy fitting." |
---|
| 192 | model = fitproblem[0].get_model() |
---|
| 193 | pars = fitproblem[0].pars |
---|
| 194 | if reset_flag: |
---|
| 195 | # reset the initial value; useful for batch |
---|
| 196 | for name in fitproblem[0].pars: |
---|
| 197 | ind = fitproblem[0].pars.index(name) |
---|
| 198 | model.model.setParam(name, fitproblem[0].vals[ind]) |
---|
| 199 | listdata = [] |
---|
| 200 | listdata = fitproblem[0].get_data() |
---|
| 201 | # Concatenate dList set (contains one or more data)before fitting |
---|
| 202 | data = listdata |
---|
| 203 | |
---|
| 204 | self.curr_thread = curr_thread |
---|
| 205 | ftol = ftol |
---|
| 206 | |
---|
| 207 | # Check the initial value if it is within range |
---|
| 208 | _check_param_range(model.model, pars) |
---|
| 209 | |
---|
| 210 | result = FResult(model=model.model, data=data, param_list=pars) |
---|
| 211 | result.fitter_id = self.fitter_id |
---|
| 212 | if handler is not None: |
---|
| 213 | handler.set_result(result=result) |
---|
| 214 | functor = SasAssembly(paramlist=pars, |
---|
| 215 | model=model, |
---|
| 216 | data=data, |
---|
| 217 | handler=handler, |
---|
| 218 | fitresult=result, |
---|
| 219 | curr_thread=curr_thread, |
---|
| 220 | msg_q=msg_q) |
---|
| 221 | try: |
---|
| 222 | # This import must be here; otherwise it will be confused when more |
---|
| 223 | # than one thread exist. |
---|
| 224 | from scipy import optimize |
---|
| 225 | |
---|
| 226 | out, cov_x, _, mesg, success = optimize.leastsq(functor, |
---|
| 227 | model.get_params(pars), |
---|
| 228 | ftol=ftol, |
---|
| 229 | full_output=1) |
---|
| 230 | except: |
---|
| 231 | if hasattr(sys, 'last_type') and sys.last_type == KeyboardInterrupt: |
---|
| 232 | if handler is not None: |
---|
| 233 | msg = "Fitting: Terminated!!!" |
---|
| 234 | handler.stop(msg) |
---|
| 235 | raise KeyboardInterrupt, msg |
---|
| 236 | else: |
---|
| 237 | raise |
---|
| 238 | chisqr = functor.chisq() |
---|
| 239 | |
---|
| 240 | if cov_x is not None and numpy.isfinite(cov_x).all(): |
---|
| 241 | stderr = numpy.sqrt(numpy.diag(cov_x)) |
---|
| 242 | else: |
---|
| 243 | stderr = [] |
---|
| 244 | |
---|
| 245 | result.index = data.idx |
---|
| 246 | result.fitness = chisqr |
---|
| 247 | result.stderr = stderr |
---|
| 248 | result.pvec = out |
---|
| 249 | result.success = success |
---|
| 250 | result.theory = functor.theory |
---|
| 251 | if handler is not None: |
---|
| 252 | handler.set_result(result=result) |
---|
| 253 | handler.update_fit(last=True) |
---|
| 254 | if q is not None: |
---|
| 255 | q.put(result) |
---|
| 256 | return q |
---|
| 257 | if success < 1 or success > 5: |
---|
| 258 | result.fitness = None |
---|
| 259 | return [result] |
---|
| 260 | |
---|
| 261 | |
---|
| 262 | def _check_param_range(model, pars): |
---|
| 263 | """ |
---|
| 264 | Check parameter range and set the initial value inside |
---|
| 265 | if it is out of range. |
---|
| 266 | |
---|
| 267 | : model: park model object |
---|
| 268 | """ |
---|
| 269 | # loop through parameterset |
---|
| 270 | for p in pars: |
---|
| 271 | value = model.getParam(p) |
---|
| 272 | low,high = model.details.setdefault(p,["",None,None])[1:3] |
---|
| 273 | # if the range was defined, check the range |
---|
| 274 | if low is not None and value <= low: |
---|
| 275 | value = low + _get_zero_shift(low) |
---|
| 276 | if high is not None and value > high: |
---|
| 277 | value = high - _get_zero_shift(high) |
---|
| 278 | # Check one more time if the new value goes below |
---|
| 279 | # the low bound, If so, re-evaluate the value |
---|
| 280 | # with the mean of the range. |
---|
| 281 | if low is not None and value < low: |
---|
| 282 | value = 0.5 * (low+high) |
---|
| 283 | model.setParam(p, value) |
---|
| 284 | |
---|
| 285 | def _get_zero_shift(limit): |
---|
| 286 | """ |
---|
| 287 | Get 10% shift of the param value = 0 based on the range value |
---|
| 288 | |
---|
| 289 | : param range: min or max value of the bounds |
---|
| 290 | """ |
---|
| 291 | return 0.1 * (limit if limit != 0.0 else 1.0) |
---|
| 292 | |
---|
| 293 | |
---|
| 294 | #def profile(fn, *args, **kw): |
---|
| 295 | # import cProfile, pstats, os |
---|
| 296 | # global call_result |
---|
| 297 | # def call(): |
---|
| 298 | # global call_result |
---|
| 299 | # call_result = fn(*args, **kw) |
---|
| 300 | # cProfile.runctx('call()', dict(call=call), {}, 'profile.out') |
---|
| 301 | # stats = pstats.Stats('profile.out') |
---|
| 302 | # stats.sort_stats('time') |
---|
| 303 | # stats.sort_stats('calls') |
---|
| 304 | # stats.print_stats() |
---|
| 305 | # os.unlink('profile.out') |
---|
| 306 | # return call_result |
---|
| 307 | |
---|
[48882d1] | 308 | |
---|