def fit(func, coords, data, p0, bounds=[], args=None, ptol=1e-4, seed=None, optimizer="apache_nm"): ''' Arguments: func -- function (or list of functions) coords -- coordinate dataset(s) data -- data to fit p0 -- list of initial parameters bounds -- list of parameter bounds, bounds are tuples of lower and upper values (any can be None) args -- extra arguments ptol -- parameter fit tolerance seed -- seed value for genetic algorithm-based optimiser optimizer -- description of the optimizer to use, e.g. ['local','global','simplex','genetic', 'gradient','apache_nm','apache_md','apache_cg'] local and global are general settings, which point the one of the specific methods If any global methods are picked, the bounds argument must also be filled in. Returns: fitresult object ''' fnlist = [] func = _toList(func) p0 = _toList(p0) if not isinstance(bounds, list): bounds = [bounds] else: bounds = list(bounds) # make a copy n_bounds = len(bounds) mixed = False for f in func: if isinstance(f, tuple): print "parameter count is no longer required" f = f[0] if function.isjclass(f): # create bound function object np = function.nparams(f) pl = _createparams(np, p0, bounds) fnlist.append(f(pl)) elif not _inspect.isfunction(f): # instantiated Java function # np = f.getNoOfParameters() fnlist.append(f) else: np = len(_inspect.getargspec(f)[0]) - 1 if np < 1: raise ValueError, "Function needs more than one argument (i.e. at least one parameter)" pl = _createparams(np, p0, bounds) fnlist.append(fitfunc(f, f.__name__, pl, args)) mixed = True if not mixed: # no jython functions cfunc = _compfn() else: cfunc = cfitfunc() for f in fnlist: if function.isjmethod(f): # unwrap f = f._jfunc() cfunc.addFunction(f) coords = list(_asIterable(coords)) for i in range(len(coords)): # check and slice coordinates to match data c = coords[i] if c.shape != data.shape: ns = [slice(d) for d in data.shape] coords[i] = c[ns] if seed: _fitter.seed = int(seed) import time start = -time.time() jcoords = [__cvt_jobj(c, copy=False, force=True) for c in coords] jdata = data._jdataset() # use the appropriate fitter for the task if optimizer == 'local': _fitter.simplexFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'global': if n_bounds == 0: print "Using a global optimizer with no bounds is unlikely to work, please use the bounds argument to narrow the search space" _fitter.geneticFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'simplex': _fitter.simplexFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'gradient': _fitter.GDFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'apache_nm': _fitter.ApacheNelderMeadFit(jcoords, jdata, cfunc) elif optimizer == 'apache_md': _fitter.ApacheMultiDirectionFit(jcoords, jdata, cfunc) elif optimizer == 'apache_cg': _fitter.ApacheConjugateGradientFit(jcoords, jdata, cfunc) elif optimizer == 'genetic': if n_bounds == 0: print "Using a global optimizer with no bounds is unlikely to work, please use the bounds argument to narrow the search space" _fitter.geneticFit(ptol, jcoords, jdata, cfunc) start += time.time() print "Fit took %fs" % start return fitresult(cfunc, coords, data)
def fit(func, coords, data, p0, bounds=[], args=None, ptol=1e-4, seed=None, optimizer="apache_nm"): ''' Arguments: func -- function (or list of functions) coords -- coordinate dataset(s) data -- data to fit p0 -- list of initial parameters bounds -- list of parameter bounds, bounds are tuples of lower and upper values (any can be None) args -- extra arguments ptol -- parameter fit tolerance seed -- seed value for genetic algorithm-based optimiser optimizer -- description of the optimizer to use, e.g. ['local','global','simplex','genetic', 'gradient','apache_nm','apache_md','apache_cg'] local and global are general settings, which point the one of the specific methods If any global methods are picked, the bounds argument must also be filled in. Returns: fitresult object ''' fnlist = [] func = _toList(func) p0 = _toList(p0) if not isinstance(bounds, list): bounds = [bounds] else: bounds = list(bounds) # make a copy n_bounds = len(bounds) mixed = False for f in func: if isinstance(f, tuple): print("parameter count is no longer required") f = f[0] if function.isjclass(f): # create bound function object np = function.nparams(f) pl = _createparams(np, p0, bounds) fnlist.append(f(pl)) elif not _inspect.isfunction(f): # instantiated Java function # np = f.getNoOfParameters() fnlist.append(f) else: np = len(_inspect.getargspec(f)[0]) - 1 if np < 1: raise ValueError("Function needs more than one argument (i.e. at least one parameter)") pl = _createparams(np, p0, bounds) fnlist.append(fitfunc(f, f.__name__, pl, args)) mixed = True if not mixed: # no jython functions cfunc = _compfn() else: cfunc = cfitfunc() for f in fnlist: if function.isjmethod(f): # unwrap f = f._jfunc() cfunc.addFunction(f) coords = list(_asIterable(coords)) for i in range(len(coords)): # check and slice coordinates to match data c = coords[i] if c.shape != data.shape: ns = [slice(d) for d in data.shape] coords[i] = c[ns] if seed: _fitter.seed = int(seed) jcoords = [ __cvt_jobj(c, copy=False, force=True) for c in coords ] jdata = data._jdataset() # use the appropriate fitter for the task if optimizer == 'local': _fitter.ApacheNelderMeadFit(jcoords, jdata, cfunc) elif optimizer == 'global': if n_bounds == 0: print("Using a global optimizer with no bounds is unlikely to work, please use the bounds argument to narrow the search space") _fitter.geneticFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'simplex': _fitter.simplexFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'gradient': _fitter.GDFit(ptol, jcoords, jdata, cfunc) elif optimizer == 'apache_nm': _fitter.ApacheNelderMeadFit(jcoords, jdata, cfunc) elif optimizer == 'apache_md': _fitter.ApacheMultiDirectionFit(jcoords, jdata, cfunc) elif optimizer == 'apache_cg': _fitter.ApacheConjugateGradientFit(jcoords, jdata, cfunc) elif optimizer == 'genetic': if n_bounds == 0: print("Using a global optimizer with no bounds is unlikely to work, please use the bounds argument to narrow the search space") _fitter.geneticFit(ptol, jcoords, jdata, cfunc) return fitresult(cfunc, coords, data)