| 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 |
|---|
| 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 sasview_model(model_definition, **pars): |
|---|
| 69 | """ |
|---|
| 70 | Load a sasview model given the model name. |
|---|
| 71 | """ |
|---|
| 72 | # convert model parameters from sasmodel form to sasview form |
|---|
| 73 | #print "old",sorted(pars.items()) |
|---|
| 74 | modelname, pars = revert_model(model_definition, pars) |
|---|
| 75 | #print "new",sorted(pars.items()) |
|---|
| 76 | sas = __import__('sas.models.'+modelname) |
|---|
| 77 | ModelClass = getattr(getattr(sas.models,modelname,None),modelname,None) |
|---|
| 78 | if ModelClass is None: |
|---|
| 79 | raise ValueError("could not find model %r in sas.models"%modelname) |
|---|
| 80 | model = ModelClass() |
|---|
| 81 | |
|---|
| 82 | for k,v in pars.items(): |
|---|
| 83 | if k.endswith("_pd"): |
|---|
| 84 | model.dispersion[k[:-3]]['width'] = v |
|---|
| 85 | elif k.endswith("_pd_n"): |
|---|
| 86 | model.dispersion[k[:-5]]['npts'] = v |
|---|
| 87 | elif k.endswith("_pd_nsigma"): |
|---|
| 88 | model.dispersion[k[:-10]]['nsigmas'] = v |
|---|
| 89 | elif k.endswith("_pd_type"): |
|---|
| 90 | model.dispersion[k[:-8]]['type'] = v |
|---|
| 91 | else: |
|---|
| 92 | model.setParam(k, v) |
|---|
| 93 | return model |
|---|
| 94 | |
|---|
| 95 | def randomize(p, v): |
|---|
| 96 | """ |
|---|
| 97 | Randomizing parameter. |
|---|
| 98 | |
|---|
| 99 | Guess the parameter type from name. |
|---|
| 100 | """ |
|---|
| 101 | if any(p.endswith(s) for s in ('_pd_n','_pd_nsigma','_pd_type')): |
|---|
| 102 | return v |
|---|
| 103 | elif any(s in p for s in ('theta','phi','psi')): |
|---|
| 104 | # orientation in [-180,180], orientation pd in [0,45] |
|---|
| 105 | if p.endswith('_pd'): |
|---|
| 106 | return 45*np.random.rand() |
|---|
| 107 | else: |
|---|
| 108 | return 360*np.random.rand() - 180 |
|---|
| 109 | elif 'sld' in p: |
|---|
| 110 | # sld in in [-0.5,10] |
|---|
| 111 | return 10.5*np.random.rand() - 0.5 |
|---|
| 112 | elif p.endswith('_pd'): |
|---|
| 113 | # length pd in [0,1] |
|---|
| 114 | return np.random.rand() |
|---|
| 115 | else: |
|---|
| 116 | # values from 0 to 2*x for all other parameters |
|---|
| 117 | return 2*np.random.rand()*(v if v != 0 else 1) |
|---|
| 118 | |
|---|
| 119 | def randomize_model(pars, seed=None): |
|---|
| 120 | if seed is None: |
|---|
| 121 | seed = np.random.randint(1e9) |
|---|
| 122 | np.random.seed(seed) |
|---|
| 123 | # Note: the sort guarantees order of calls to random number generator |
|---|
| 124 | pars = dict((p,randomize(p,v)) for p,v in sorted(pars.items())) |
|---|
| 125 | |
|---|
| 126 | return pars, seed |
|---|
| 127 | |
|---|
| 128 | def constrain_pars(model_definition, pars): |
|---|
| 129 | name = model_definition.name |
|---|
| 130 | if name == 'capped_cylinder' and pars['cap_radius'] < pars['radius']: |
|---|
| 131 | pars['radius'],pars['cap_radius'] = pars['cap_radius'],pars['radius'] |
|---|
| 132 | if name == 'barbell' and pars['bell_radius'] < pars['radius']: |
|---|
| 133 | pars['radius'],pars['bell_radius'] = pars['bell_radius'],pars['radius'] |
|---|
| 134 | |
|---|
| 135 | # Limit guinier to an Rg such that Iq > 1e-30 (single precision cutoff) |
|---|
| 136 | if name == 'guinier': |
|---|
| 137 | #q_max = 0.2 # mid q maximum |
|---|
| 138 | q_max = 1.0 # high q maximum |
|---|
| 139 | rg_max = np.sqrt(90*np.log(10) + 3*np.log(pars['scale']))/q_max |
|---|
| 140 | pars['rg'] = min(pars['rg'],rg_max) |
|---|
| 141 | |
|---|
| 142 | # These constraints are only needed for comparison to sasview |
|---|
| 143 | if name in ('teubner_strey','broad_peak'): |
|---|
| 144 | del pars['scale'] |
|---|
| 145 | if name in ('guinier',): |
|---|
| 146 | del pars['background'] |
|---|
| 147 | if getattr(model_definition, 'category', None) == 'structure-factor': |
|---|
| 148 | del pars['scale'], pars['background'] |
|---|
| 149 | |
|---|
| 150 | |
|---|
| 151 | def parlist(pars): |
|---|
| 152 | return "\n".join("%s: %s"%(p,v) for p,v in sorted(pars.items())) |
|---|
| 153 | |
|---|
| 154 | def suppress_pd(pars): |
|---|
| 155 | """ |
|---|
| 156 | Suppress theta_pd for now until the normalization is resolved. |
|---|
| 157 | |
|---|
| 158 | May also suppress complete polydispersity of the model to test |
|---|
| 159 | models more quickly. |
|---|
| 160 | """ |
|---|
| 161 | for p in pars: |
|---|
| 162 | if p.endswith("_pd"): pars[p] = 0 |
|---|
| 163 | |
|---|
| 164 | def eval_sasview(model_definition, pars, data, Nevals=1): |
|---|
| 165 | # importing sas here so that the error message will be that sas failed to |
|---|
| 166 | # import rather than the more obscure smear_selection not imported error |
|---|
| 167 | import sas |
|---|
| 168 | from sas.models.qsmearing import smear_selection |
|---|
| 169 | model = sasview_model(model_definition, **pars) |
|---|
| 170 | smearer = smear_selection(data, model=model) |
|---|
| 171 | value = None # silence the linter |
|---|
| 172 | toc = tic() |
|---|
| 173 | for _ in range(max(Nevals, 1)): # make sure there is at least one eval |
|---|
| 174 | if hasattr(data, 'qx_data'): |
|---|
| 175 | q = np.sqrt(data.qx_data**2 + data.qy_data**2) |
|---|
| 176 | index = ((~data.mask) & (~np.isnan(data.data)) |
|---|
| 177 | & (q >= data.qmin) & (q <= data.qmax)) |
|---|
| 178 | if smearer is not None: |
|---|
| 179 | smearer.model = model # because smear_selection has a bug |
|---|
| 180 | smearer.accuracy = data.accuracy |
|---|
| 181 | smearer.set_index(index) |
|---|
| 182 | value = smearer.get_value() |
|---|
| 183 | else: |
|---|
| 184 | value = model.evalDistribution([data.qx_data[index], data.qy_data[index]]) |
|---|
| 185 | else: |
|---|
| 186 | value = model.evalDistribution(data.x) |
|---|
| 187 | if smearer is not None: |
|---|
| 188 | value = smearer(value) |
|---|
| 189 | average_time = toc()*1000./Nevals |
|---|
| 190 | return value, average_time |
|---|
| 191 | |
|---|
| 192 | def eval_opencl(model_definition, pars, data, dtype='single', Nevals=1, cutoff=0.): |
|---|
| 193 | try: |
|---|
| 194 | model = core.load_model(model_definition, dtype=dtype, platform="ocl") |
|---|
| 195 | except Exception,exc: |
|---|
| 196 | print exc |
|---|
| 197 | print "... trying again with single precision" |
|---|
| 198 | model = core.load_model(model_definition, dtype='single', platform="ocl") |
|---|
| 199 | calculator = DirectModel(data, model, cutoff=cutoff) |
|---|
| 200 | value = None # silence the linter |
|---|
| 201 | toc = tic() |
|---|
| 202 | for _ in range(max(Nevals, 1)): # force at least one eval |
|---|
| 203 | value = calculator(**pars) |
|---|
| 204 | average_time = toc()*1000./Nevals |
|---|
| 205 | return value, average_time |
|---|
| 206 | |
|---|
| 207 | |
|---|
| 208 | def eval_ctypes(model_definition, pars, data, dtype='double', Nevals=1, cutoff=0.): |
|---|
| 209 | model = core.load_model(model_definition, dtype=dtype, platform="dll") |
|---|
| 210 | calculator = DirectModel(data, model, cutoff=cutoff) |
|---|
| 211 | value = None # silence the linter |
|---|
| 212 | toc = tic() |
|---|
| 213 | for _ in range(max(Nevals, 1)): # force at least one eval |
|---|
| 214 | value = calculator(**pars) |
|---|
| 215 | average_time = toc()*1000./Nevals |
|---|
| 216 | return value, average_time |
|---|
| 217 | |
|---|
| 218 | |
|---|
| 219 | def make_data(qmax, is2D, Nq=128, resolution=0.0, accuracy='Low', view='log'): |
|---|
| 220 | if is2D: |
|---|
| 221 | data = empty_data2D(np.linspace(-qmax, qmax, Nq), resolution=resolution) |
|---|
| 222 | data.accuracy = accuracy |
|---|
| 223 | set_beam_stop(data, 0.004) |
|---|
| 224 | index = ~data.mask |
|---|
| 225 | else: |
|---|
| 226 | if view == 'log': |
|---|
| 227 | qmax = math.log10(qmax) |
|---|
| 228 | q = np.logspace(qmax-3, qmax, Nq) |
|---|
| 229 | else: |
|---|
| 230 | q = np.linspace(0.001*qmax, qmax, Nq) |
|---|
| 231 | data = empty_data1D(q, resolution=resolution) |
|---|
| 232 | index = slice(None, None) |
|---|
| 233 | return data, index |
|---|
| 234 | |
|---|
| 235 | def compare(name, pars, Ncpu, Nocl, opts, set_pars): |
|---|
| 236 | model_definition = core.load_model_definition(name) |
|---|
| 237 | |
|---|
| 238 | view = ('linear' if '-linear' in opts |
|---|
| 239 | else 'log' if '-log' in opts |
|---|
| 240 | else 'q4' if '-q4' in opts |
|---|
| 241 | else 'log') |
|---|
| 242 | |
|---|
| 243 | opt_values = dict(split |
|---|
| 244 | for s in opts for split in ((s.split('='),)) |
|---|
| 245 | if len(split) == 2) |
|---|
| 246 | # Sort out data |
|---|
| 247 | qmax = (10.0 if '-exq' in opts |
|---|
| 248 | else 1.0 if '-highq' in opts |
|---|
| 249 | else 0.2 if '-midq' in opts |
|---|
| 250 | else 0.05) |
|---|
| 251 | Nq = int(opt_values.get('-Nq', '128')) |
|---|
| 252 | res = float(opt_values.get('-res', '0')) |
|---|
| 253 | accuracy = opt_values.get('-accuracy', 'Low') |
|---|
| 254 | is2D = "-2d" in opts |
|---|
| 255 | data, index = make_data(qmax, is2D, Nq, res, accuracy, view=view) |
|---|
| 256 | |
|---|
| 257 | |
|---|
| 258 | # modelling accuracy is determined by dtype and cutoff |
|---|
| 259 | dtype = ('longdouble' if '-quad' in opts |
|---|
| 260 | else 'double' if '-double' in opts |
|---|
| 261 | else 'single') |
|---|
| 262 | cutoff = float(opt_values.get('-cutoff','1e-5')) |
|---|
| 263 | |
|---|
| 264 | # randomize parameters |
|---|
| 265 | #pars.update(set_pars) # set value before random to control range |
|---|
| 266 | if '-random' in opts or '-random' in opt_values: |
|---|
| 267 | seed = int(opt_values['-random']) if '-random' in opt_values else None |
|---|
| 268 | pars, seed = randomize_model(pars, seed=seed) |
|---|
| 269 | print "Randomize using -random=%i"%seed |
|---|
| 270 | pars.update(set_pars) # set value after random to control value |
|---|
| 271 | constrain_pars(model_definition, pars) |
|---|
| 272 | |
|---|
| 273 | # parameter selection |
|---|
| 274 | if '-mono' in opts: |
|---|
| 275 | suppress_pd(pars) |
|---|
| 276 | if '-pars' in opts: |
|---|
| 277 | print "pars",parlist(pars) |
|---|
| 278 | |
|---|
| 279 | # OpenCl calculation |
|---|
| 280 | if Nocl > 0 and "-ctypes" in opts and "-sasview" in opts: |
|---|
| 281 | try: |
|---|
| 282 | ocl, ocl_time = eval_sasview(model_definition, pars, data, Ncpu) |
|---|
| 283 | base = "sasview" |
|---|
| 284 | #print "ocl/sasview", (ocl-pars['background'])/(cpu-pars['background']) |
|---|
| 285 | print "sasview t=%.1f ms, intensity=%.0f"%(ocl_time, sum(ocl)) |
|---|
| 286 | #print "sasview",cpu |
|---|
| 287 | except ImportError: |
|---|
| 288 | traceback.print_exc() |
|---|
| 289 | Ncpu = 0 |
|---|
| 290 | elif Nocl > 0: |
|---|
| 291 | ocl, ocl_time = eval_opencl(model_definition, pars, data, |
|---|
| 292 | dtype=dtype, cutoff=cutoff, Nevals=Nocl) |
|---|
| 293 | base = "ocl" |
|---|
| 294 | print "opencl t=%.1f ms, intensity=%.0f"%(ocl_time, sum(ocl)) |
|---|
| 295 | #print "ocl", ocl |
|---|
| 296 | #print max(ocl), min(ocl) |
|---|
| 297 | |
|---|
| 298 | # ctypes/sasview calculation |
|---|
| 299 | if Ncpu > 0 and "-ctypes" in opts: |
|---|
| 300 | cpu, cpu_time = eval_ctypes(model_definition, pars, data, |
|---|
| 301 | dtype=dtype, cutoff=cutoff, Nevals=Ncpu) |
|---|
| 302 | comp = "ctypes" |
|---|
| 303 | print "ctypes t=%.1f ms, intensity=%.0f"%(cpu_time, sum(cpu)) |
|---|
| 304 | elif Ncpu > 0: |
|---|
| 305 | try: |
|---|
| 306 | cpu, cpu_time = eval_sasview(model_definition, pars, data, Ncpu) |
|---|
| 307 | comp = "sasview" |
|---|
| 308 | #print "ocl/sasview", (ocl-pars['background'])/(cpu-pars['background']) |
|---|
| 309 | print "sasview t=%.1f ms, intensity=%.0f"%(cpu_time, sum(cpu)) |
|---|
| 310 | #print "sasview",cpu |
|---|
| 311 | except ImportError: |
|---|
| 312 | traceback.print_exc() |
|---|
| 313 | Ncpu = 0 |
|---|
| 314 | |
|---|
| 315 | # Compare, but only if computing both forms |
|---|
| 316 | if Nocl > 0 and Ncpu > 0: |
|---|
| 317 | #print "speedup %.2g"%(cpu_time/ocl_time) |
|---|
| 318 | #print "max |ocl/cpu|", max(abs(ocl/cpu)), "%.15g"%max(abs(ocl)), "%.15g"%max(abs(cpu)) |
|---|
| 319 | #cpu *= max(ocl/cpu) |
|---|
| 320 | resid = (ocl - cpu) |
|---|
| 321 | relerr = resid/cpu |
|---|
| 322 | #bad = (relerr>1e-4) |
|---|
| 323 | #print relerr[bad],cpu[bad],ocl[bad],data.qx_data[bad],data.qy_data[bad] |
|---|
| 324 | _print_stats("|%s-%s|"%(base,comp)+(" "*(3+len(comp))), resid) |
|---|
| 325 | _print_stats("|(%s-%s)/%s|"%(base,comp,comp), relerr) |
|---|
| 326 | |
|---|
| 327 | # Plot if requested |
|---|
| 328 | if '-noplot' in opts: return |
|---|
| 329 | import matplotlib.pyplot as plt |
|---|
| 330 | if Ncpu > 0: |
|---|
| 331 | if Nocl > 0: plt.subplot(131) |
|---|
| 332 | plot_theory(data, cpu, view=view, plot_data=False) |
|---|
| 333 | plt.title("%s t=%.1f ms"%(comp,cpu_time)) |
|---|
| 334 | #cbar_title = "log I" |
|---|
| 335 | if Nocl > 0: |
|---|
| 336 | if Ncpu > 0: plt.subplot(132) |
|---|
| 337 | plot_theory(data, ocl, view=view, plot_data=False) |
|---|
| 338 | plt.title("%s t=%.1f ms"%(base,ocl_time)) |
|---|
| 339 | #cbar_title = "log I" |
|---|
| 340 | if Ncpu > 0 and Nocl > 0: |
|---|
| 341 | plt.subplot(133) |
|---|
| 342 | if '-abs' in opts: |
|---|
| 343 | err,errstr,errview = resid, "abs err", "linear" |
|---|
| 344 | else: |
|---|
| 345 | err,errstr,errview = abs(relerr), "rel err", "log" |
|---|
| 346 | #err,errstr = ocl/cpu,"ratio" |
|---|
| 347 | plot_theory(data, None, resid=err, view=errview, plot_data=False) |
|---|
| 348 | plt.title("max %s = %.3g"%(errstr, max(abs(err)))) |
|---|
| 349 | #cbar_title = errstr if errview=="linear" else "log "+errstr |
|---|
| 350 | #if is2D: |
|---|
| 351 | # h = plt.colorbar() |
|---|
| 352 | # h.ax.set_title(cbar_title) |
|---|
| 353 | |
|---|
| 354 | if Ncpu > 0 and Nocl > 0 and '-hist' in opts: |
|---|
| 355 | plt.figure() |
|---|
| 356 | v = relerr |
|---|
| 357 | v[v==0] = 0.5*np.min(np.abs(v[v!=0])) |
|---|
| 358 | plt.hist(np.log10(np.abs(v)), normed=1, bins=50); |
|---|
| 359 | plt.xlabel('log10(err), err = | F(q) single - F(q) double| / | F(q) double |'); |
|---|
| 360 | plt.ylabel('P(err)') |
|---|
| 361 | plt.title('Comparison of single and double precision models for %s'%name) |
|---|
| 362 | |
|---|
| 363 | plt.show() |
|---|
| 364 | |
|---|
| 365 | def _print_stats(label, err): |
|---|
| 366 | sorted_err = np.sort(abs(err)) |
|---|
| 367 | p50 = int((len(err)-1)*0.50) |
|---|
| 368 | p98 = int((len(err)-1)*0.98) |
|---|
| 369 | data = [ |
|---|
| 370 | "max:%.3e"%sorted_err[-1], |
|---|
| 371 | "median:%.3e"%sorted_err[p50], |
|---|
| 372 | "98%%:%.3e"%sorted_err[p98], |
|---|
| 373 | "rms:%.3e"%np.sqrt(np.mean(err**2)), |
|---|
| 374 | "zero-offset:%+.3e"%np.mean(err), |
|---|
| 375 | ] |
|---|
| 376 | print label," ".join(data) |
|---|
| 377 | |
|---|
| 378 | |
|---|
| 379 | |
|---|
| 380 | # =========================================================================== |
|---|
| 381 | # |
|---|
| 382 | USAGE=""" |
|---|
| 383 | usage: compare.py model [Nopencl] [Nsasview] [options...] [key=val] |
|---|
| 384 | |
|---|
| 385 | Compare the speed and value for a model between the SasView original and the |
|---|
| 386 | OpenCL rewrite. |
|---|
| 387 | |
|---|
| 388 | model is the name of the model to compare (see below). |
|---|
| 389 | Nopencl is the number of times to run the OpenCL model (default=5) |
|---|
| 390 | Nsasview is the number of times to run the Sasview model (default=1) |
|---|
| 391 | |
|---|
| 392 | Options (* for default): |
|---|
| 393 | |
|---|
| 394 | -plot*/-noplot plots or suppress the plot of the model |
|---|
| 395 | -single*/-double/-quad use single/double/quad precision for comparison |
|---|
| 396 | -lowq*/-midq/-highq/-exq use q values up to 0.05, 0.2, 1.0, 10.0 |
|---|
| 397 | -Nq=128 sets the number of Q points in the data set |
|---|
| 398 | -1d*/-2d computes 1d or 2d data |
|---|
| 399 | -preset*/-random[=seed] preset or random parameters |
|---|
| 400 | -mono/-poly* force monodisperse/polydisperse |
|---|
| 401 | -ctypes/-sasview* selects gpu:cpu, gpu:sasview, or sasview:cpu if both |
|---|
| 402 | -cutoff=1e-5* cutoff value for including a point in polydispersity |
|---|
| 403 | -pars/-nopars* prints the parameter set or not |
|---|
| 404 | -abs/-rel* plot relative or absolute error |
|---|
| 405 | -linear/-log/-q4 intensity scaling |
|---|
| 406 | -hist/-nohist* plot histogram of relative error |
|---|
| 407 | -res=0 sets the resolution width dQ/Q if calculating with resolution |
|---|
| 408 | -accuracy=Low resolution accuracy Low, Mid, High, Xhigh |
|---|
| 409 | |
|---|
| 410 | Key=value pairs allow you to set specific values to any of the model |
|---|
| 411 | parameters. |
|---|
| 412 | |
|---|
| 413 | Available models: |
|---|
| 414 | """ |
|---|
| 415 | |
|---|
| 416 | |
|---|
| 417 | NAME_OPTIONS = set([ |
|---|
| 418 | 'plot','noplot', |
|---|
| 419 | 'single','double','quad', |
|---|
| 420 | 'lowq','midq','highq','exq', |
|---|
| 421 | '2d','1d', |
|---|
| 422 | 'preset','random', |
|---|
| 423 | 'poly','mono', |
|---|
| 424 | 'sasview','ctypes', |
|---|
| 425 | 'nopars','pars', |
|---|
| 426 | 'rel','abs', |
|---|
| 427 | 'linear', 'log', 'q4', |
|---|
| 428 | 'hist','nohist', |
|---|
| 429 | ]) |
|---|
| 430 | VALUE_OPTIONS = [ |
|---|
| 431 | # Note: random is both a name option and a value option |
|---|
| 432 | 'cutoff', 'random', 'Nq', 'res', 'accuracy', |
|---|
| 433 | ] |
|---|
| 434 | |
|---|
| 435 | def columnize(L, indent="", width=79): |
|---|
| 436 | column_width = max(len(w) for w in L) + 1 |
|---|
| 437 | num_columns = (width - len(indent)) // column_width |
|---|
| 438 | num_rows = len(L) // num_columns |
|---|
| 439 | L = L + [""] * (num_rows*num_columns - len(L)) |
|---|
| 440 | columns = [L[k*num_rows:(k+1)*num_rows] for k in range(num_columns)] |
|---|
| 441 | lines = [" ".join("%-*s"%(column_width, entry) for entry in row) |
|---|
| 442 | for row in zip(*columns)] |
|---|
| 443 | output = indent + ("\n"+indent).join(lines) |
|---|
| 444 | return output |
|---|
| 445 | |
|---|
| 446 | |
|---|
| 447 | def get_demo_pars(model_definition): |
|---|
| 448 | info = generate.make_info(model_definition) |
|---|
| 449 | pars = dict((p[0],p[2]) for p in info['parameters']) |
|---|
| 450 | pars.update(info['demo']) |
|---|
| 451 | return pars |
|---|
| 452 | |
|---|
| 453 | def main(): |
|---|
| 454 | opts = [arg for arg in sys.argv[1:] if arg.startswith('-')] |
|---|
| 455 | popts = [arg for arg in sys.argv[1:] if not arg.startswith('-') and '=' in arg] |
|---|
| 456 | args = [arg for arg in sys.argv[1:] if not arg.startswith('-') and '=' not in arg] |
|---|
| 457 | models = "\n ".join("%-15s"%v for v in MODELS) |
|---|
| 458 | if len(args) == 0: |
|---|
| 459 | print(USAGE) |
|---|
| 460 | print(columnize(MODELS, indent=" ")) |
|---|
| 461 | sys.exit(1) |
|---|
| 462 | if args[0] not in MODELS: |
|---|
| 463 | print "Model %r not available. Use one of:\n %s"%(args[0],models) |
|---|
| 464 | sys.exit(1) |
|---|
| 465 | if len(args) > 3: |
|---|
| 466 | print("expected parameters: model Nopencl Nsasview") |
|---|
| 467 | |
|---|
| 468 | invalid = [o[1:] for o in opts |
|---|
| 469 | if o[1:] not in NAME_OPTIONS |
|---|
| 470 | and not any(o.startswith('-%s='%t) for t in VALUE_OPTIONS)] |
|---|
| 471 | if invalid: |
|---|
| 472 | print "Invalid options: %s"%(", ".join(invalid)) |
|---|
| 473 | sys.exit(1) |
|---|
| 474 | |
|---|
| 475 | # Get demo parameters from model definition, or use default parameters |
|---|
| 476 | # if model does not define demo parameters |
|---|
| 477 | name = args[0] |
|---|
| 478 | model_definition = core.load_model_definition(name) |
|---|
| 479 | pars = get_demo_pars(model_definition) |
|---|
| 480 | |
|---|
| 481 | Nopencl = int(args[1]) if len(args) > 1 else 5 |
|---|
| 482 | Nsasview = int(args[2]) if len(args) > 2 else 1 |
|---|
| 483 | |
|---|
| 484 | # Fill in default polydispersity parameters |
|---|
| 485 | pds = set(p.split('_pd')[0] for p in pars if p.endswith('_pd')) |
|---|
| 486 | for p in pds: |
|---|
| 487 | if p+"_pd_nsigma" not in pars: pars[p+"_pd_nsigma"] = 3 |
|---|
| 488 | if p+"_pd_type" not in pars: pars[p+"_pd_type"] = "gaussian" |
|---|
| 489 | |
|---|
| 490 | # Fill in parameters given on the command line |
|---|
| 491 | set_pars = {} |
|---|
| 492 | for arg in popts: |
|---|
| 493 | k,v = arg.split('=',1) |
|---|
| 494 | if k not in pars: |
|---|
| 495 | # extract base name without distribution |
|---|
| 496 | s = set(p.split('_pd')[0] for p in pars) |
|---|
| 497 | print "%r invalid; parameters are: %s"%(k,", ".join(sorted(s))) |
|---|
| 498 | sys.exit(1) |
|---|
| 499 | set_pars[k] = float(v) if not v.endswith('type') else v |
|---|
| 500 | |
|---|
| 501 | compare(name, pars, Nsasview, Nopencl, opts, set_pars) |
|---|
| 502 | |
|---|
| 503 | if __name__ == "__main__": |
|---|
| 504 | main() |
|---|
