| 1 | #!/usr/bin/env python |
|---|
| 2 | # -*- coding: utf-8 -*- |
|---|
| 3 | |
|---|
| 4 | import sys |
|---|
| 5 | import math |
|---|
| 6 | from os.path import basename, dirname, join as joinpath |
|---|
| 7 | import glob |
|---|
| 8 | import datetime |
|---|
| 9 | import traceback |
|---|
| 10 | |
|---|
| 11 | import numpy as np |
|---|
| 12 | |
|---|
| 13 | ROOT = dirname(__file__) |
|---|
| 14 | sys.path.insert(0, ROOT) # Make sure sasmodels is first on the path |
|---|
| 15 | |
|---|
| 16 | |
|---|
| 17 | from . import core |
|---|
| 18 | from . import kerneldll |
|---|
| 19 | from . import generate |
|---|
| 20 | from .data import plot_theory, empty_data1D, empty_data2D |
|---|
| 21 | from .direct_model import DirectModel |
|---|
| 22 | from .convert import revert_model, constrain_new_to_old |
|---|
| 23 | kerneldll.ALLOW_SINGLE_PRECISION_DLLS = True |
|---|
| 24 | |
|---|
| 25 | # List of available models |
|---|
| 26 | MODELS = [basename(f)[:-3] |
|---|
| 27 | for f in sorted(glob.glob(joinpath(ROOT,"models","[a-zA-Z]*.py")))] |
|---|
| 28 | |
|---|
| 29 | # CRUFT python 2.6 |
|---|
| 30 | if not hasattr(datetime.timedelta, 'total_seconds'): |
|---|
| 31 | def delay(dt): |
|---|
| 32 | """Return number date-time delta as number seconds""" |
|---|
| 33 | return dt.days * 86400 + dt.seconds + 1e-6 * dt.microseconds |
|---|
| 34 | else: |
|---|
| 35 | def delay(dt): |
|---|
| 36 | """Return number date-time delta as number seconds""" |
|---|
| 37 | return dt.total_seconds() |
|---|
| 38 | |
|---|
| 39 | |
|---|
| 40 | def tic(): |
|---|
| 41 | """ |
|---|
| 42 | Timer function. |
|---|
| 43 | |
|---|
| 44 | Use "toc=tic()" to start the clock and "toc()" to measure |
|---|
| 45 | a time interval. |
|---|
| 46 | """ |
|---|
| 47 | then = datetime.datetime.now() |
|---|
| 48 | return lambda: delay(datetime.datetime.now() - then) |
|---|
| 49 | |
|---|
| 50 | |
|---|
| 51 | def set_beam_stop(data, radius, outer=None): |
|---|
| 52 | """ |
|---|
| 53 | Add a beam stop of the given *radius*. If *outer*, make an annulus. |
|---|
| 54 | |
|---|
| 55 | Note: this function does not use the sasview package |
|---|
| 56 | """ |
|---|
| 57 | if hasattr(data, 'qx_data'): |
|---|
| 58 | q = np.sqrt(data.qx_data**2 + data.qy_data**2) |
|---|
| 59 | data.mask = (q < radius) |
|---|
| 60 | if outer is not None: |
|---|
| 61 | data.mask |= (q >= outer) |
|---|
| 62 | else: |
|---|
| 63 | data.mask = (data.x < radius) |
|---|
| 64 | if outer is not None: |
|---|
| 65 | data.mask |= (data.x >= outer) |
|---|
| 66 | |
|---|
| 67 | |
|---|
| 68 | def parameter_range(p, v): |
|---|
| 69 | """ |
|---|
| 70 | Choose a parameter range based on parameter name and initial value. |
|---|
| 71 | """ |
|---|
| 72 | if p.endswith('_pd_n'): |
|---|
| 73 | return [0, 100] |
|---|
| 74 | elif p.endswith('_pd_nsigma'): |
|---|
| 75 | return [0, 5] |
|---|
| 76 | elif p.endswith('_pd_type'): |
|---|
| 77 | return v |
|---|
| 78 | elif any(s in p for s in ('theta','phi','psi')): |
|---|
| 79 | # orientation in [-180,180], orientation pd in [0,45] |
|---|
| 80 | if p.endswith('_pd'): |
|---|
| 81 | return [0,45] |
|---|
| 82 | else: |
|---|
| 83 | return [-180, 180] |
|---|
| 84 | elif 'sld' in p: |
|---|
| 85 | return [-0.5, 10] |
|---|
| 86 | elif p.endswith('_pd'): |
|---|
| 87 | return [0, 1] |
|---|
| 88 | elif p == 'background': |
|---|
| 89 | return [0, 10] |
|---|
| 90 | elif p == 'scale': |
|---|
| 91 | return [0, 1e3] |
|---|
| 92 | elif p == 'case_num': |
|---|
| 93 | # RPA hack |
|---|
| 94 | return [0, 10] |
|---|
| 95 | elif v < 0: |
|---|
| 96 | # Kxy parameters in rpa model can be negative |
|---|
| 97 | return [2*v, -2*v] |
|---|
| 98 | else: |
|---|
| 99 | return [0, (2*v if v>0 else 1)] |
|---|
| 100 | |
|---|
| 101 | def _randomize_one(p, v): |
|---|
| 102 | """ |
|---|
| 103 | Randomizing parameter. |
|---|
| 104 | """ |
|---|
| 105 | if any(p.endswith(s) for s in ('_pd_n','_pd_nsigma','_pd_type')): |
|---|
| 106 | return v |
|---|
| 107 | else: |
|---|
| 108 | return np.random.uniform(*parameter_range(p, v)) |
|---|
| 109 | |
|---|
| 110 | def randomize_pars(pars, seed=None): |
|---|
| 111 | np.random.seed(seed) |
|---|
| 112 | # Note: the sort guarantees order `of calls to random number generator |
|---|
| 113 | pars = dict((p,_randomize_one(p,v)) |
|---|
| 114 | for p,v in sorted(pars.items())) |
|---|
| 115 | return pars |
|---|
| 116 | |
|---|
| 117 | def constrain_pars(model_definition, pars): |
|---|
| 118 | """ |
|---|
| 119 | Restrict parameters to valid values. |
|---|
| 120 | """ |
|---|
| 121 | name = model_definition.name |
|---|
| 122 | if name == 'capped_cylinder' and pars['cap_radius'] < pars['radius']: |
|---|
| 123 | pars['radius'],pars['cap_radius'] = pars['cap_radius'],pars['radius'] |
|---|
| 124 | if name == 'barbell' and pars['bell_radius'] < pars['radius']: |
|---|
| 125 | pars['radius'],pars['bell_radius'] = pars['bell_radius'],pars['radius'] |
|---|
| 126 | |
|---|
| 127 | # Limit guinier to an Rg such that Iq > 1e-30 (single precision cutoff) |
|---|
| 128 | if name == 'guinier': |
|---|
| 129 | #q_max = 0.2 # mid q maximum |
|---|
| 130 | q_max = 1.0 # high q maximum |
|---|
| 131 | rg_max = np.sqrt(90*np.log(10) + 3*np.log(pars['scale']))/q_max |
|---|
| 132 | pars['rg'] = min(pars['rg'],rg_max) |
|---|
| 133 | |
|---|
| 134 | if name == 'rpa': |
|---|
| 135 | # Make sure phi sums to 1.0 |
|---|
| 136 | if pars['case_num'] < 2: |
|---|
| 137 | pars['Phia'] = 0. |
|---|
| 138 | pars['Phib'] = 0. |
|---|
| 139 | elif pars['case_num'] < 5: |
|---|
| 140 | pars['Phia'] = 0. |
|---|
| 141 | total = sum(pars['Phi'+c] for c in 'abcd') |
|---|
| 142 | for c in 'abcd': |
|---|
| 143 | pars['Phi'+c] /= total |
|---|
| 144 | |
|---|
| 145 | def parlist(pars): |
|---|
| 146 | return "\n".join("%s: %s"%(p,v) for p,v in sorted(pars.items())) |
|---|
| 147 | |
|---|
| 148 | def suppress_pd(pars): |
|---|
| 149 | """ |
|---|
| 150 | Suppress theta_pd for now until the normalization is resolved. |
|---|
| 151 | |
|---|
| 152 | May also suppress complete polydispersity of the model to test |
|---|
| 153 | models more quickly. |
|---|
| 154 | """ |
|---|
| 155 | pars = pars.copy() |
|---|
| 156 | for p in pars: |
|---|
| 157 | if p.endswith("_pd_n"): pars[p] = 0 |
|---|
| 158 | return pars |
|---|
| 159 | |
|---|
| 160 | def eval_sasview(model_definition, data): |
|---|
| 161 | # importing sas here so that the error message will be that sas failed to |
|---|
| 162 | # import rather than the more obscure smear_selection not imported error |
|---|
| 163 | import sas |
|---|
| 164 | from sas.models.qsmearing import smear_selection |
|---|
| 165 | |
|---|
| 166 | # convert model parameters from sasmodel form to sasview form |
|---|
| 167 | #print("old",sorted(pars.items())) |
|---|
| 168 | modelname, pars = revert_model(model_definition, {}) |
|---|
| 169 | #print("new",sorted(pars.items())) |
|---|
| 170 | sas = __import__('sas.models.'+modelname) |
|---|
| 171 | ModelClass = getattr(getattr(sas.models,modelname,None),modelname,None) |
|---|
| 172 | if ModelClass is None: |
|---|
| 173 | raise ValueError("could not find model %r in sas.models"%modelname) |
|---|
| 174 | model = ModelClass() |
|---|
| 175 | smearer = smear_selection(data, model=model) |
|---|
| 176 | |
|---|
| 177 | if hasattr(data, 'qx_data'): |
|---|
| 178 | q = np.sqrt(data.qx_data**2 + data.qy_data**2) |
|---|
| 179 | index = ((~data.mask) & (~np.isnan(data.data)) |
|---|
| 180 | & (q >= data.qmin) & (q <= data.qmax)) |
|---|
| 181 | if smearer is not None: |
|---|
| 182 | smearer.model = model # because smear_selection has a bug |
|---|
| 183 | smearer.accuracy = data.accuracy |
|---|
| 184 | smearer.set_index(index) |
|---|
| 185 | theory = lambda: smearer.get_value() |
|---|
| 186 | else: |
|---|
| 187 | theory = lambda: model.evalDistribution([data.qx_data[index], data.qy_data[index]]) |
|---|
| 188 | elif smearer is not None: |
|---|
| 189 | theory = lambda: smearer(model.evalDistribution(data.x)) |
|---|
| 190 | else: |
|---|
| 191 | theory = lambda: model.evalDistribution(data.x) |
|---|
| 192 | |
|---|
| 193 | def calculator(**pars): |
|---|
| 194 | # paying for parameter conversion each time to keep life simple, if not fast |
|---|
| 195 | _, pars = revert_model(model_definition, pars) |
|---|
| 196 | for k,v in pars.items(): |
|---|
| 197 | parts = k.split('.') # polydispersity components |
|---|
| 198 | if len(parts) == 2: |
|---|
| 199 | model.dispersion[parts[0]][parts[1]] = v |
|---|
| 200 | else: |
|---|
| 201 | model.setParam(k, v) |
|---|
| 202 | return theory() |
|---|
| 203 | |
|---|
| 204 | calculator.engine = "sasview" |
|---|
| 205 | return calculator |
|---|
| 206 | |
|---|
| 207 | DTYPE_MAP = { |
|---|
| 208 | 'half': '16', |
|---|
| 209 | 'fast': 'fast', |
|---|
| 210 | 'single': '32', |
|---|
| 211 | 'double': '64', |
|---|
| 212 | 'quad': '128', |
|---|
| 213 | 'f16': '16', |
|---|
| 214 | 'f32': '32', |
|---|
| 215 | 'f64': '64', |
|---|
| 216 | 'longdouble': '128', |
|---|
| 217 | } |
|---|
| 218 | def eval_opencl(model_definition, data, dtype='single', cutoff=0.): |
|---|
| 219 | try: |
|---|
| 220 | model = core.load_model(model_definition, dtype=dtype, platform="ocl") |
|---|
| 221 | except Exception as exc: |
|---|
| 222 | print(exc) |
|---|
| 223 | print("... trying again with single precision") |
|---|
| 224 | dtype = 'single' |
|---|
| 225 | model = core.load_model(model_definition, dtype=dtype, platform="ocl") |
|---|
| 226 | calculator = DirectModel(data, model, cutoff=cutoff) |
|---|
| 227 | calculator.engine = "OCL%s"%DTYPE_MAP[dtype] |
|---|
| 228 | return calculator |
|---|
| 229 | |
|---|
| 230 | def eval_ctypes(model_definition, data, dtype='double', cutoff=0.): |
|---|
| 231 | if dtype=='quad': |
|---|
| 232 | dtype = 'longdouble' |
|---|
| 233 | model = core.load_model(model_definition, dtype=dtype, platform="dll") |
|---|
| 234 | calculator = DirectModel(data, model, cutoff=cutoff) |
|---|
| 235 | calculator.engine = "OMP%s"%DTYPE_MAP[dtype] |
|---|
| 236 | return calculator |
|---|
| 237 | |
|---|
| 238 | def time_calculation(calculator, pars, Nevals=1): |
|---|
| 239 | # initialize the code so time is more accurate |
|---|
| 240 | value = calculator(**suppress_pd(pars)) |
|---|
| 241 | toc = tic() |
|---|
| 242 | for _ in range(max(Nevals, 1)): # make sure there is at least one eval |
|---|
| 243 | value = calculator(**pars) |
|---|
| 244 | average_time = toc()*1000./Nevals |
|---|
| 245 | return value, average_time |
|---|
| 246 | |
|---|
| 247 | def make_data(opts): |
|---|
| 248 | qmax, nq, res = opts['qmax'], opts['nq'], opts['res'] |
|---|
| 249 | if opts['is2d']: |
|---|
| 250 | data = empty_data2D(np.linspace(-qmax, qmax, nq), resolution=res) |
|---|
| 251 | data.accuracy = opts['accuracy'] |
|---|
| 252 | set_beam_stop(data, 0.004) |
|---|
| 253 | index = ~data.mask |
|---|
| 254 | else: |
|---|
| 255 | if opts['view'] == 'log': |
|---|
| 256 | qmax = math.log10(qmax) |
|---|
| 257 | q = np.logspace(qmax-3, qmax, nq) |
|---|
| 258 | else: |
|---|
| 259 | q = np.linspace(0.001*qmax, qmax, nq) |
|---|
| 260 | data = empty_data1D(q, resolution=res) |
|---|
| 261 | index = slice(None, None) |
|---|
| 262 | return data, index |
|---|
| 263 | |
|---|
| 264 | def make_engine(model_definition, data, dtype, cutoff): |
|---|
| 265 | if dtype == 'sasview': |
|---|
| 266 | return eval_sasview(model_definition, data) |
|---|
| 267 | elif dtype.endswith('!'): |
|---|
| 268 | return eval_ctypes(model_definition, data, dtype=dtype[:-1], |
|---|
| 269 | cutoff=cutoff) |
|---|
| 270 | else: |
|---|
| 271 | return eval_opencl(model_definition, data, dtype=dtype, |
|---|
| 272 | cutoff=cutoff) |
|---|
| 273 | |
|---|
| 274 | def compare(opts, limits=None): |
|---|
| 275 | Nbase, Ncomp = opts['N1'], opts['N2'] |
|---|
| 276 | pars = opts['pars'] |
|---|
| 277 | data = opts['data'] |
|---|
| 278 | |
|---|
| 279 | # Base calculation |
|---|
| 280 | if Nbase > 0: |
|---|
| 281 | base = opts['engines'][0] |
|---|
| 282 | try: |
|---|
| 283 | base_value, base_time = time_calculation(base, pars, Nbase) |
|---|
| 284 | print("%s t=%.1f ms, intensity=%.0f"%(base.engine, base_time, sum(base_value))) |
|---|
| 285 | except ImportError: |
|---|
| 286 | traceback.print_exc() |
|---|
| 287 | Nbase = 0 |
|---|
| 288 | |
|---|
| 289 | # Comparison calculation |
|---|
| 290 | if Ncomp > 0: |
|---|
| 291 | comp = opts['engines'][1] |
|---|
| 292 | try: |
|---|
| 293 | comp_value, comp_time = time_calculation(comp, pars, Ncomp) |
|---|
| 294 | print("%s t=%.1f ms, intensity=%.0f"%(comp.engine, comp_time, sum(comp_value))) |
|---|
| 295 | except ImportError: |
|---|
| 296 | traceback.print_exc() |
|---|
| 297 | Ncomp = 0 |
|---|
| 298 | |
|---|
| 299 | # Compare, but only if computing both forms |
|---|
| 300 | if Nbase > 0 and Ncomp > 0: |
|---|
| 301 | #print("speedup %.2g"%(comp_time/base_time)) |
|---|
| 302 | #print("max |base/comp|", max(abs(base_value/comp_value)), "%.15g"%max(abs(base_value)), "%.15g"%max(abs(comp_value))) |
|---|
| 303 | #comp *= max(base_value/comp_value) |
|---|
| 304 | resid = (base_value - comp_value) |
|---|
| 305 | relerr = resid/comp_value |
|---|
| 306 | _print_stats("|%s - %s|"%(base.engine,comp.engine)+(" "*(3+len(comp.engine))), resid) |
|---|
| 307 | _print_stats("|(%s - %s) / %s|"%(base.engine,comp.engine,comp.engine), relerr) |
|---|
| 308 | |
|---|
| 309 | # Plot if requested |
|---|
| 310 | if not opts['plot'] and not opts['explore']: return |
|---|
| 311 | view = opts['view'] |
|---|
| 312 | import matplotlib.pyplot as plt |
|---|
| 313 | if limits is None: |
|---|
| 314 | vmin, vmax = np.Inf, -np.Inf |
|---|
| 315 | if Nbase > 0: |
|---|
| 316 | vmin = min(vmin, min(base_value)) |
|---|
| 317 | vmax = max(vmax, max(base_value)) |
|---|
| 318 | if Ncomp > 0: |
|---|
| 319 | vmin = min(vmin, min(comp_value)) |
|---|
| 320 | vmax = max(vmax, max(comp_value)) |
|---|
| 321 | limits = vmin, vmax |
|---|
| 322 | |
|---|
| 323 | if Nbase > 0: |
|---|
| 324 | if Ncomp > 0: plt.subplot(131) |
|---|
| 325 | plot_theory(data, base_value, view=view, plot_data=False, limits=limits) |
|---|
| 326 | plt.title("%s t=%.1f ms"%(base.engine, base_time)) |
|---|
| 327 | #cbar_title = "log I" |
|---|
| 328 | if Ncomp > 0: |
|---|
| 329 | if Nbase > 0: plt.subplot(132) |
|---|
| 330 | plot_theory(data, comp_value, view=view, plot_data=False, limits=limits) |
|---|
| 331 | plt.title("%s t=%.1f ms"%(comp.engine,comp_time)) |
|---|
| 332 | #cbar_title = "log I" |
|---|
| 333 | if Ncomp > 0 and Nbase > 0: |
|---|
| 334 | plt.subplot(133) |
|---|
| 335 | if '-abs' in opts: |
|---|
| 336 | err,errstr,errview = resid, "abs err", "linear" |
|---|
| 337 | else: |
|---|
| 338 | err,errstr,errview = abs(relerr), "rel err", "log" |
|---|
| 339 | #err,errstr = base/comp,"ratio" |
|---|
| 340 | plot_theory(data, None, resid=err, view=errview, plot_data=False) |
|---|
| 341 | plt.title("max %s = %.3g"%(errstr, max(abs(err)))) |
|---|
| 342 | #cbar_title = errstr if errview=="linear" else "log "+errstr |
|---|
| 343 | #if is2D: |
|---|
| 344 | # h = plt.colorbar() |
|---|
| 345 | # h.ax.set_title(cbar_title) |
|---|
| 346 | |
|---|
| 347 | if Ncomp > 0 and Nbase > 0 and '-hist' in opts: |
|---|
| 348 | plt.figure() |
|---|
| 349 | v = relerr |
|---|
| 350 | v[v==0] = 0.5*np.min(np.abs(v[v!=0])) |
|---|
| 351 | plt.hist(np.log10(np.abs(v)), normed=1, bins=50); |
|---|
| 352 | plt.xlabel('log10(err), err = |(%s - %s) / %s|'%(base.engine, comp.engine, comp.engine)); |
|---|
| 353 | plt.ylabel('P(err)') |
|---|
| 354 | plt.title('Distribution of relative error between calculation engines') |
|---|
| 355 | |
|---|
| 356 | if not opts['explore']: |
|---|
| 357 | plt.show() |
|---|
| 358 | |
|---|
| 359 | return limits |
|---|
| 360 | |
|---|
| 361 | def _print_stats(label, err): |
|---|
| 362 | sorted_err = np.sort(abs(err)) |
|---|
| 363 | p50 = int((len(err)-1)*0.50) |
|---|
| 364 | p98 = int((len(err)-1)*0.98) |
|---|
| 365 | data = [ |
|---|
| 366 | "max:%.3e"%sorted_err[-1], |
|---|
| 367 | "median:%.3e"%sorted_err[p50], |
|---|
| 368 | "98%%:%.3e"%sorted_err[p98], |
|---|
| 369 | "rms:%.3e"%np.sqrt(np.mean(err**2)), |
|---|
| 370 | "zero-offset:%+.3e"%np.mean(err), |
|---|
| 371 | ] |
|---|
| 372 | print(label+" ".join(data)) |
|---|
| 373 | |
|---|
| 374 | |
|---|
| 375 | |
|---|
| 376 | # =========================================================================== |
|---|
| 377 | # |
|---|
| 378 | USAGE=""" |
|---|
| 379 | usage: compare.py model N1 N2 [options...] [key=val] |
|---|
| 380 | |
|---|
| 381 | Compare the speed and value for a model between the SasView original and the |
|---|
| 382 | sasmodels rewrite. |
|---|
| 383 | |
|---|
| 384 | model is the name of the model to compare (see below). |
|---|
| 385 | N1 is the number of times to run sasmodels (default=1). |
|---|
| 386 | N2 is the number times to run sasview (default=1). |
|---|
| 387 | |
|---|
| 388 | Options (* for default): |
|---|
| 389 | |
|---|
| 390 | -plot*/-noplot plots or suppress the plot of the model |
|---|
| 391 | -lowq*/-midq/-highq/-exq use q values up to 0.05, 0.2, 1.0, 10.0 |
|---|
| 392 | -nq=128 sets the number of Q points in the data set |
|---|
| 393 | -1d*/-2d computes 1d or 2d data |
|---|
| 394 | -preset*/-random[=seed] preset or random parameters |
|---|
| 395 | -mono/-poly* force monodisperse/polydisperse |
|---|
| 396 | -cutoff=1e-5* cutoff value for including a point in polydispersity |
|---|
| 397 | -pars/-nopars* prints the parameter set or not |
|---|
| 398 | -abs/-rel* plot relative or absolute error |
|---|
| 399 | -linear/-log*/-q4 intensity scaling |
|---|
| 400 | -hist/-nohist* plot histogram of relative error |
|---|
| 401 | -res=0 sets the resolution width dQ/Q if calculating with resolution |
|---|
| 402 | -accuracy=Low accuracy of the resolution calculation Low, Mid, High, Xhigh |
|---|
| 403 | -edit starts the parameter explorer |
|---|
| 404 | |
|---|
| 405 | Any two calculation engines can be selected for comparison: |
|---|
| 406 | |
|---|
| 407 | -single/-double/-half/-fast sets an OpenCL calculation engine |
|---|
| 408 | -single!/-double!/-quad! sets an OpenMP calculation engine |
|---|
| 409 | -sasview sets the sasview calculation engine |
|---|
| 410 | |
|---|
| 411 | The default is -single -sasview. Note that the interpretation of quad |
|---|
| 412 | precision depends on architecture, and may vary from 64-bit to 128-bit, |
|---|
| 413 | with 80-bit floats being common (1e-19 precision). |
|---|
| 414 | |
|---|
| 415 | Key=value pairs allow you to set specific values for the model parameters. |
|---|
| 416 | |
|---|
| 417 | Available models: |
|---|
| 418 | """ |
|---|
| 419 | |
|---|
| 420 | |
|---|
| 421 | NAME_OPTIONS = set([ |
|---|
| 422 | 'plot', 'noplot', |
|---|
| 423 | 'half', 'fast', 'single', 'double', |
|---|
| 424 | 'single!', 'double!', 'quad!', 'sasview', |
|---|
| 425 | 'lowq', 'midq', 'highq', 'exq', |
|---|
| 426 | '2d', '1d', |
|---|
| 427 | 'preset', 'random', |
|---|
| 428 | 'poly', 'mono', |
|---|
| 429 | 'nopars', 'pars', |
|---|
| 430 | 'rel', 'abs', |
|---|
| 431 | 'linear', 'log', 'q4', |
|---|
| 432 | 'hist', 'nohist', |
|---|
| 433 | 'edit', |
|---|
| 434 | ]) |
|---|
| 435 | VALUE_OPTIONS = [ |
|---|
| 436 | # Note: random is both a name option and a value option |
|---|
| 437 | 'cutoff', 'random', 'nq', 'res', 'accuracy', |
|---|
| 438 | ] |
|---|
| 439 | |
|---|
| 440 | def columnize(L, indent="", width=79): |
|---|
| 441 | column_width = max(len(w) for w in L) + 1 |
|---|
| 442 | num_columns = (width - len(indent)) // column_width |
|---|
| 443 | num_rows = len(L) // num_columns |
|---|
| 444 | L = L + [""] * (num_rows*num_columns - len(L)) |
|---|
| 445 | columns = [L[k*num_rows:(k+1)*num_rows] for k in range(num_columns)] |
|---|
| 446 | lines = [" ".join("%-*s"%(column_width, entry) for entry in row) |
|---|
| 447 | for row in zip(*columns)] |
|---|
| 448 | output = indent + ("\n"+indent).join(lines) |
|---|
| 449 | return output |
|---|
| 450 | |
|---|
| 451 | |
|---|
| 452 | def get_demo_pars(model_definition): |
|---|
| 453 | info = generate.make_info(model_definition) |
|---|
| 454 | # Get the default values for the parameters |
|---|
| 455 | pars = dict((p[0],p[2]) for p in info['parameters']) |
|---|
| 456 | |
|---|
| 457 | # Fill in default values for the polydispersity parameters |
|---|
| 458 | for p in info['parameters']: |
|---|
| 459 | if p[4] in ('volume', 'orientation'): |
|---|
| 460 | pars[p[0]+'_pd'] = 0.0 |
|---|
| 461 | pars[p[0]+'_pd_n'] = 0 |
|---|
| 462 | pars[p[0]+'_pd_nsigma'] = 3.0 |
|---|
| 463 | pars[p[0]+'_pd_type'] = "gaussian" |
|---|
| 464 | |
|---|
| 465 | # Plug in values given in demo |
|---|
| 466 | pars.update(info['demo']) |
|---|
| 467 | return pars |
|---|
| 468 | |
|---|
| 469 | def parse_opts(): |
|---|
| 470 | flags = [arg for arg in sys.argv[1:] if arg.startswith('-')] |
|---|
| 471 | values = [arg for arg in sys.argv[1:] if not arg.startswith('-') and '=' in arg] |
|---|
| 472 | args = [arg for arg in sys.argv[1:] if not arg.startswith('-') and '=' not in arg] |
|---|
| 473 | models = "\n ".join("%-15s"%v for v in MODELS) |
|---|
| 474 | if len(args) == 0: |
|---|
| 475 | print(USAGE) |
|---|
| 476 | print(columnize(MODELS, indent=" ")) |
|---|
| 477 | sys.exit(1) |
|---|
| 478 | if args[0] not in MODELS: |
|---|
| 479 | print("Model %r not available. Use one of:\n %s"%(args[0],models)) |
|---|
| 480 | sys.exit(1) |
|---|
| 481 | if len(args) > 3: |
|---|
| 482 | print("expected parameters: model Nopencl Nsasview") |
|---|
| 483 | |
|---|
| 484 | invalid = [o[1:] for o in flags |
|---|
| 485 | if o[1:] not in NAME_OPTIONS |
|---|
| 486 | and not any(o.startswith('-%s='%t) for t in VALUE_OPTIONS)] |
|---|
| 487 | if invalid: |
|---|
| 488 | print("Invalid options: %s"%(", ".join(invalid))) |
|---|
| 489 | sys.exit(1) |
|---|
| 490 | |
|---|
| 491 | |
|---|
| 492 | # Interpret the flags |
|---|
| 493 | opts = { |
|---|
| 494 | 'plot' : True, |
|---|
| 495 | 'view' : 'log', |
|---|
| 496 | 'is2d' : False, |
|---|
| 497 | 'qmax' : 0.05, |
|---|
| 498 | 'nq' : 128, |
|---|
| 499 | 'res' : 0.0, |
|---|
| 500 | 'accuracy' : 'Low', |
|---|
| 501 | 'cutoff' : 1e-5, |
|---|
| 502 | 'seed' : -1, # default to preset |
|---|
| 503 | 'mono' : False, |
|---|
| 504 | 'show_pars' : False, |
|---|
| 505 | 'show_hist' : False, |
|---|
| 506 | 'rel_err' : True, |
|---|
| 507 | 'explore' : False, |
|---|
| 508 | } |
|---|
| 509 | engines = [] |
|---|
| 510 | for arg in flags: |
|---|
| 511 | if arg == '-noplot': opts['plot'] = False |
|---|
| 512 | elif arg == '-plot': opts['plot'] = True |
|---|
| 513 | elif arg == '-linear': opts['view'] = 'linear' |
|---|
| 514 | elif arg == '-log': opts['view'] = 'log' |
|---|
| 515 | elif arg == '-q4': opts['view'] = 'q4' |
|---|
| 516 | elif arg == '-1d': opts['is2d'] = False |
|---|
| 517 | elif arg == '-2d': opts['is2d'] = True |
|---|
| 518 | elif arg == '-exq': opts['qmax'] = 10.0 |
|---|
| 519 | elif arg == '-highq': opts['qmax'] = 1.0 |
|---|
| 520 | elif arg == '-midq': opts['qmax'] = 0.2 |
|---|
| 521 | elif arg == '-loq': opts['qmax'] = 0.05 |
|---|
| 522 | elif arg.startswith('-nq='): opts['nq'] = int(arg[4:]) |
|---|
| 523 | elif arg.startswith('-res='): opts['res'] = float(arg[5:]) |
|---|
| 524 | elif arg.startswith('-accuracy='): opts['accuracy'] = arg[10:] |
|---|
| 525 | elif arg.startswith('-cutoff='): opts['cutoff'] = float(arg[8:]) |
|---|
| 526 | elif arg.startswith('-random='): opts['seed'] = int(arg[8:]) |
|---|
| 527 | elif arg == '-random': opts['seed'] = np.random.randint(1e6) |
|---|
| 528 | elif arg == '-preset': opts['seed'] = -1 |
|---|
| 529 | elif arg == '-mono': opts['mono'] = True |
|---|
| 530 | elif arg == '-poly': opts['mono'] = False |
|---|
| 531 | elif arg == '-pars': opts['show_pars'] = True |
|---|
| 532 | elif arg == '-nopars': opts['show_pars'] = False |
|---|
| 533 | elif arg == '-hist': opts['show_hist'] = True |
|---|
| 534 | elif arg == '-nohist': opts['show_hist'] = False |
|---|
| 535 | elif arg == '-rel': opts['rel_err'] = True |
|---|
| 536 | elif arg == '-abs': opts['rel_err'] = False |
|---|
| 537 | elif arg == '-half': engines.append(arg[1:]) |
|---|
| 538 | elif arg == '-fast': engines.append(arg[1:]) |
|---|
| 539 | elif arg == '-single': engines.append(arg[1:]) |
|---|
| 540 | elif arg == '-double': engines.append(arg[1:]) |
|---|
| 541 | elif arg == '-single!': engines.append(arg[1:]) |
|---|
| 542 | elif arg == '-double!': engines.append(arg[1:]) |
|---|
| 543 | elif arg == '-quad!': engines.append(arg[1:]) |
|---|
| 544 | elif arg == '-sasview': engines.append(arg[1:]) |
|---|
| 545 | elif arg == '-edit': opts['explore'] = True |
|---|
| 546 | |
|---|
| 547 | if len(engines) == 0: |
|---|
| 548 | engines.extend(['single','sasview']) |
|---|
| 549 | elif len(engines) == 1: |
|---|
| 550 | if engines[0][0] != 'sasview': |
|---|
| 551 | engines.append('sasview') |
|---|
| 552 | else: |
|---|
| 553 | engines.append('single') |
|---|
| 554 | elif len(engines) > 2: |
|---|
| 555 | del engines[2:] |
|---|
| 556 | |
|---|
| 557 | name = args[0] |
|---|
| 558 | model_definition = core.load_model_definition(name) |
|---|
| 559 | |
|---|
| 560 | N1 = int(args[1]) if len(args) > 1 else 1 |
|---|
| 561 | N2 = int(args[2]) if len(args) > 2 else 1 |
|---|
| 562 | |
|---|
| 563 | # Get demo parameters from model definition, or use default parameters |
|---|
| 564 | # if model does not define demo parameters |
|---|
| 565 | pars = get_demo_pars(model_definition) |
|---|
| 566 | |
|---|
| 567 | # Fill in parameters given on the command line |
|---|
| 568 | presets = {} |
|---|
| 569 | for arg in values: |
|---|
| 570 | k,v = arg.split('=',1) |
|---|
| 571 | if k not in pars: |
|---|
| 572 | # extract base name without polydispersity info |
|---|
| 573 | s = set(p.split('_pd')[0] for p in pars) |
|---|
| 574 | print("%r invalid; parameters are: %s"%(k,", ".join(sorted(s)))) |
|---|
| 575 | sys.exit(1) |
|---|
| 576 | presets[k] = float(v) if not k.endswith('type') else v |
|---|
| 577 | |
|---|
| 578 | # randomize parameters |
|---|
| 579 | #pars.update(set_pars) # set value before random to control range |
|---|
| 580 | if opts['seed'] > -1: |
|---|
| 581 | pars = randomize_pars(pars, seed=opts['seed']) |
|---|
| 582 | print("Randomize using -random=%i"%opts['seed']) |
|---|
| 583 | if opts['mono']: |
|---|
| 584 | pars = suppress_pd(pars) |
|---|
| 585 | pars.update(presets) # set value after random to control value |
|---|
| 586 | constrain_pars(model_definition, pars) |
|---|
| 587 | constrain_new_to_old(model_definition, pars) |
|---|
| 588 | if opts['show_pars']: |
|---|
| 589 | print("pars " + str(parlist(pars))) |
|---|
| 590 | |
|---|
| 591 | # Create the computational engines |
|---|
| 592 | data, _index = make_data(opts) |
|---|
| 593 | if N1: |
|---|
| 594 | base = make_engine(model_definition, data, engines[0], opts['cutoff']) |
|---|
| 595 | else: |
|---|
| 596 | base = None |
|---|
| 597 | if N2: |
|---|
| 598 | comp = make_engine(model_definition, data, engines[1], opts['cutoff']) |
|---|
| 599 | else: |
|---|
| 600 | comp = None |
|---|
| 601 | |
|---|
| 602 | # Remember it all |
|---|
| 603 | opts.update({ |
|---|
| 604 | 'name' : name, |
|---|
| 605 | 'def' : model_definition, |
|---|
| 606 | 'N1' : N1, |
|---|
| 607 | 'N2' : N2, |
|---|
| 608 | 'presets' : presets, |
|---|
| 609 | 'pars' : pars, |
|---|
| 610 | 'data' : data, |
|---|
| 611 | 'engines' : [base, comp], |
|---|
| 612 | }) |
|---|
| 613 | |
|---|
| 614 | return opts |
|---|
| 615 | |
|---|
| 616 | def main(): |
|---|
| 617 | opts = parse_opts() |
|---|
| 618 | if opts['explore']: |
|---|
| 619 | explore(opts) |
|---|
| 620 | else: |
|---|
| 621 | compare(opts) |
|---|
| 622 | |
|---|
| 623 | def explore(opts): |
|---|
| 624 | import wx |
|---|
| 625 | from bumps.names import FitProblem |
|---|
| 626 | from bumps.gui.app_frame import AppFrame |
|---|
| 627 | |
|---|
| 628 | problem = FitProblem(Explore(opts)) |
|---|
| 629 | isMac = "cocoa" in wx.version() |
|---|
| 630 | app = wx.App() |
|---|
| 631 | frame = AppFrame(parent=None, title="explore") |
|---|
| 632 | if not isMac: frame.Show() |
|---|
| 633 | frame.panel.set_model(model=problem) |
|---|
| 634 | frame.panel.Layout() |
|---|
| 635 | frame.panel.aui.Split(0, wx.TOP) |
|---|
| 636 | if isMac: frame.Show() |
|---|
| 637 | app.MainLoop() |
|---|
| 638 | |
|---|
| 639 | class Explore(object): |
|---|
| 640 | """ |
|---|
| 641 | Return a bumps wrapper for a SAS model comparison. |
|---|
| 642 | """ |
|---|
| 643 | def __init__(self, opts): |
|---|
| 644 | from bumps.cli import config_matplotlib |
|---|
| 645 | import bumps_model |
|---|
| 646 | config_matplotlib() |
|---|
| 647 | self.opts = opts |
|---|
| 648 | info = generate.make_info(opts['def']) |
|---|
| 649 | pars, pd_types = bumps_model.create_parameters(info, **opts['pars']) |
|---|
| 650 | if not opts['is2d']: |
|---|
| 651 | active = [base + ext |
|---|
| 652 | for base in info['partype']['pd-1d'] |
|---|
| 653 | for ext in ['','_pd','_pd_n','_pd_nsigma']] |
|---|
| 654 | active.extend(info['partype']['fixed-1d']) |
|---|
| 655 | for k in active: |
|---|
| 656 | v = pars[k] |
|---|
| 657 | v.range(*parameter_range(k, v.value)) |
|---|
| 658 | else: |
|---|
| 659 | for k, v in pars.items(): |
|---|
| 660 | v.range(*parameter_range(k, v.value)) |
|---|
| 661 | |
|---|
| 662 | self.pars = pars |
|---|
| 663 | self.pd_types = pd_types |
|---|
| 664 | self.limits = None |
|---|
| 665 | |
|---|
| 666 | def numpoints(self): |
|---|
| 667 | """ |
|---|
| 668 | Return the number of points |
|---|
| 669 | """ |
|---|
| 670 | return len(self.pars) + 1 # so dof is 1 |
|---|
| 671 | |
|---|
| 672 | def parameters(self): |
|---|
| 673 | """ |
|---|
| 674 | Return a dictionary of parameters |
|---|
| 675 | """ |
|---|
| 676 | return self.pars |
|---|
| 677 | |
|---|
| 678 | def nllf(self): |
|---|
| 679 | return 0. # No nllf |
|---|
| 680 | |
|---|
| 681 | def plot(self, view='log'): |
|---|
| 682 | """ |
|---|
| 683 | Plot the data and residuals. |
|---|
| 684 | """ |
|---|
| 685 | pars = dict((k, v.value) for k,v in self.pars.items()) |
|---|
| 686 | pars.update(self.pd_types) |
|---|
| 687 | self.opts['pars'] = pars |
|---|
| 688 | limits = compare(self.opts, limits=self.limits) |
|---|
| 689 | if self.limits is None: |
|---|
| 690 | vmin, vmax = limits |
|---|
| 691 | vmax = 1.3*vmax |
|---|
| 692 | vmin = vmax*1e-7 |
|---|
| 693 | self.limits = vmin, vmax |
|---|
| 694 | |
|---|
| 695 | |
|---|
| 696 | if __name__ == "__main__": |
|---|
| 697 | main() |
|---|
