1 | #!/usr/bin/env python |
---|
2 | # -*- coding: utf-8 -*- |
---|
3 | """ |
---|
4 | Program to compare models using different compute engines. |
---|
5 | |
---|
6 | This program lets you compare results between OpenCL and DLL versions |
---|
7 | of the code and between precision (half, fast, single, double, quad), |
---|
8 | where fast precision is single precision using native functions for |
---|
9 | trig, etc., and may not be completely IEEE 754 compliant. This lets |
---|
10 | make sure that the model calculations are stable, or if you need to |
---|
11 | tag the model as double precision only. |
---|
12 | |
---|
13 | Run using ./compare.sh (Linux, Mac) or compare.bat (Windows) in the |
---|
14 | sasmodels root to see the command line options. |
---|
15 | |
---|
16 | Note that there is no way within sasmodels to select between an |
---|
17 | OpenCL CPU device and a GPU device, but you can do so by setting the |
---|
18 | PYOPENCL_CTX environment variable ahead of time. Start a python |
---|
19 | interpreter and enter:: |
---|
20 | |
---|
21 | import pyopencl as cl |
---|
22 | cl.create_some_context() |
---|
23 | |
---|
24 | This will prompt you to select from the available OpenCL devices |
---|
25 | and tell you which string to use for the PYOPENCL_CTX variable. |
---|
26 | On Windows you will need to remove the quotes. |
---|
27 | """ |
---|
28 | |
---|
29 | from __future__ import print_function |
---|
30 | |
---|
31 | import sys |
---|
32 | import math |
---|
33 | import datetime |
---|
34 | import traceback |
---|
35 | |
---|
36 | import numpy as np # type: ignore |
---|
37 | |
---|
38 | from . import core |
---|
39 | from . import kerneldll |
---|
40 | from .data import plot_theory, empty_data1D, empty_data2D |
---|
41 | from .direct_model import DirectModel |
---|
42 | from .convert import revert_name, revert_pars, constrain_new_to_old |
---|
43 | |
---|
44 | try: |
---|
45 | from typing import Optional, Dict, Any, Callable, Tuple |
---|
46 | except: |
---|
47 | pass |
---|
48 | else: |
---|
49 | from .modelinfo import ModelInfo, Parameter, ParameterSet |
---|
50 | from .data import Data |
---|
51 | Calculator = Callable[[float], np.ndarray] |
---|
52 | |
---|
53 | USAGE = """ |
---|
54 | usage: compare.py model N1 N2 [options...] [key=val] |
---|
55 | |
---|
56 | Compare the speed and value for a model between the SasView original and the |
---|
57 | sasmodels rewrite. |
---|
58 | |
---|
59 | model is the name of the model to compare (see below). |
---|
60 | N1 is the number of times to run sasmodels (default=1). |
---|
61 | N2 is the number times to run sasview (default=1). |
---|
62 | |
---|
63 | Options (* for default): |
---|
64 | |
---|
65 | -plot*/-noplot plots or suppress the plot of the model |
---|
66 | -lowq*/-midq/-highq/-exq use q values up to 0.05, 0.2, 1.0, 10.0 |
---|
67 | -nq=128 sets the number of Q points in the data set |
---|
68 | -zero indicates that q=0 should be included |
---|
69 | -1d*/-2d computes 1d or 2d data |
---|
70 | -preset*/-random[=seed] preset or random parameters |
---|
71 | -mono/-poly* force monodisperse/polydisperse |
---|
72 | -cutoff=1e-5* cutoff value for including a point in polydispersity |
---|
73 | -pars/-nopars* prints the parameter set or not |
---|
74 | -abs/-rel* plot relative or absolute error |
---|
75 | -linear/-log*/-q4 intensity scaling |
---|
76 | -hist/-nohist* plot histogram of relative error |
---|
77 | -res=0 sets the resolution width dQ/Q if calculating with resolution |
---|
78 | -accuracy=Low accuracy of the resolution calculation Low, Mid, High, Xhigh |
---|
79 | -edit starts the parameter explorer |
---|
80 | -default/-demo* use demo vs default parameters |
---|
81 | |
---|
82 | Any two calculation engines can be selected for comparison: |
---|
83 | |
---|
84 | -single/-double/-half/-fast sets an OpenCL calculation engine |
---|
85 | -single!/-double!/-quad! sets an OpenMP calculation engine |
---|
86 | -sasview sets the sasview calculation engine |
---|
87 | |
---|
88 | The default is -single -sasview. Note that the interpretation of quad |
---|
89 | precision depends on architecture, and may vary from 64-bit to 128-bit, |
---|
90 | with 80-bit floats being common (1e-19 precision). |
---|
91 | |
---|
92 | Key=value pairs allow you to set specific values for the model parameters. |
---|
93 | """ |
---|
94 | |
---|
95 | # Update docs with command line usage string. This is separate from the usual |
---|
96 | # doc string so that we can display it at run time if there is an error. |
---|
97 | # lin |
---|
98 | __doc__ = (__doc__ # pylint: disable=redefined-builtin |
---|
99 | + """ |
---|
100 | Program description |
---|
101 | ------------------- |
---|
102 | |
---|
103 | """ |
---|
104 | + USAGE) |
---|
105 | |
---|
106 | kerneldll.ALLOW_SINGLE_PRECISION_DLLS = True |
---|
107 | |
---|
108 | # CRUFT python 2.6 |
---|
109 | if not hasattr(datetime.timedelta, 'total_seconds'): |
---|
110 | def delay(dt): |
---|
111 | """Return number date-time delta as number seconds""" |
---|
112 | return dt.days * 86400 + dt.seconds + 1e-6 * dt.microseconds |
---|
113 | else: |
---|
114 | def delay(dt): |
---|
115 | """Return number date-time delta as number seconds""" |
---|
116 | return dt.total_seconds() |
---|
117 | |
---|
118 | |
---|
119 | class push_seed(object): |
---|
120 | """ |
---|
121 | Set the seed value for the random number generator. |
---|
122 | |
---|
123 | When used in a with statement, the random number generator state is |
---|
124 | restored after the with statement is complete. |
---|
125 | |
---|
126 | :Parameters: |
---|
127 | |
---|
128 | *seed* : int or array_like, optional |
---|
129 | Seed for RandomState |
---|
130 | |
---|
131 | :Example: |
---|
132 | |
---|
133 | Seed can be used directly to set the seed:: |
---|
134 | |
---|
135 | >>> from numpy.random import randint |
---|
136 | >>> push_seed(24) |
---|
137 | <...push_seed object at...> |
---|
138 | >>> print(randint(0,1000000,3)) |
---|
139 | [242082 899 211136] |
---|
140 | |
---|
141 | Seed can also be used in a with statement, which sets the random |
---|
142 | number generator state for the enclosed computations and restores |
---|
143 | it to the previous state on completion:: |
---|
144 | |
---|
145 | >>> with push_seed(24): |
---|
146 | ... print(randint(0,1000000,3)) |
---|
147 | [242082 899 211136] |
---|
148 | |
---|
149 | Using nested contexts, we can demonstrate that state is indeed |
---|
150 | restored after the block completes:: |
---|
151 | |
---|
152 | >>> with push_seed(24): |
---|
153 | ... print(randint(0,1000000)) |
---|
154 | ... with push_seed(24): |
---|
155 | ... print(randint(0,1000000,3)) |
---|
156 | ... print(randint(0,1000000)) |
---|
157 | 242082 |
---|
158 | [242082 899 211136] |
---|
159 | 899 |
---|
160 | |
---|
161 | The restore step is protected against exceptions in the block:: |
---|
162 | |
---|
163 | >>> with push_seed(24): |
---|
164 | ... print(randint(0,1000000)) |
---|
165 | ... try: |
---|
166 | ... with push_seed(24): |
---|
167 | ... print(randint(0,1000000,3)) |
---|
168 | ... raise Exception() |
---|
169 | ... except Exception: |
---|
170 | ... print("Exception raised") |
---|
171 | ... print(randint(0,1000000)) |
---|
172 | 242082 |
---|
173 | [242082 899 211136] |
---|
174 | Exception raised |
---|
175 | 899 |
---|
176 | """ |
---|
177 | def __init__(self, seed=None): |
---|
178 | # type: (Optional[int]) -> None |
---|
179 | self._state = np.random.get_state() |
---|
180 | np.random.seed(seed) |
---|
181 | |
---|
182 | def __enter__(self): |
---|
183 | # type: () -> None |
---|
184 | pass |
---|
185 | |
---|
186 | def __exit__(self, type, value, traceback): |
---|
187 | # type: (Any, BaseException, Any) -> None |
---|
188 | # TODO: better typing for __exit__ method |
---|
189 | np.random.set_state(self._state) |
---|
190 | |
---|
191 | def tic(): |
---|
192 | # type: () -> Callable[[], float] |
---|
193 | """ |
---|
194 | Timer function. |
---|
195 | |
---|
196 | Use "toc=tic()" to start the clock and "toc()" to measure |
---|
197 | a time interval. |
---|
198 | """ |
---|
199 | then = datetime.datetime.now() |
---|
200 | return lambda: delay(datetime.datetime.now() - then) |
---|
201 | |
---|
202 | |
---|
203 | def set_beam_stop(data, radius, outer=None): |
---|
204 | # type: (Data, float, float) -> None |
---|
205 | """ |
---|
206 | Add a beam stop of the given *radius*. If *outer*, make an annulus. |
---|
207 | |
---|
208 | Note: this function does not require sasview |
---|
209 | """ |
---|
210 | if hasattr(data, 'qx_data'): |
---|
211 | q = np.sqrt(data.qx_data**2 + data.qy_data**2) |
---|
212 | data.mask = (q < radius) |
---|
213 | if outer is not None: |
---|
214 | data.mask |= (q >= outer) |
---|
215 | else: |
---|
216 | data.mask = (data.x < radius) |
---|
217 | if outer is not None: |
---|
218 | data.mask |= (data.x >= outer) |
---|
219 | |
---|
220 | |
---|
221 | def parameter_range(p, v): |
---|
222 | # type: (str, float) -> Tuple[float, float] |
---|
223 | """ |
---|
224 | Choose a parameter range based on parameter name and initial value. |
---|
225 | """ |
---|
226 | # process the polydispersity options |
---|
227 | if p.endswith('_pd_n'): |
---|
228 | return 0., 100. |
---|
229 | elif p.endswith('_pd_nsigma'): |
---|
230 | return 0., 5. |
---|
231 | elif p.endswith('_pd_type'): |
---|
232 | raise ValueError("Cannot return a range for a string value") |
---|
233 | elif any(s in p for s in ('theta', 'phi', 'psi')): |
---|
234 | # orientation in [-180,180], orientation pd in [0,45] |
---|
235 | if p.endswith('_pd'): |
---|
236 | return 0., 45. |
---|
237 | else: |
---|
238 | return -180., 180. |
---|
239 | elif p.endswith('_pd'): |
---|
240 | return 0., 1. |
---|
241 | elif 'sld' in p: |
---|
242 | return -0.5, 10. |
---|
243 | elif p == 'background': |
---|
244 | return 0., 10. |
---|
245 | elif p == 'scale': |
---|
246 | return 0., 1.e3 |
---|
247 | elif v < 0.: |
---|
248 | return 2.*v, -2.*v |
---|
249 | else: |
---|
250 | return 0., (2.*v if v > 0. else 1.) |
---|
251 | |
---|
252 | |
---|
253 | def _randomize_one(model_info, p, v): |
---|
254 | # type: (ModelInfo, str, float) -> float |
---|
255 | # type: (ModelInfo, str, str) -> str |
---|
256 | """ |
---|
257 | Randomize a single parameter. |
---|
258 | """ |
---|
259 | if any(p.endswith(s) for s in ('_pd_n', '_pd_nsigma', '_pd_type')): |
---|
260 | return v |
---|
261 | |
---|
262 | # Find the parameter definition |
---|
263 | for par in model_info.parameters.call_parameters: |
---|
264 | if par.name == p: |
---|
265 | break |
---|
266 | else: |
---|
267 | raise ValueError("unknown parameter %r"%p) |
---|
268 | |
---|
269 | if len(par.limits) > 2: # choice list |
---|
270 | return np.random.randint(len(par.limits)) |
---|
271 | |
---|
272 | limits = par.limits |
---|
273 | if not np.isfinite(limits).all(): |
---|
274 | low, high = parameter_range(p, v) |
---|
275 | limits = (max(limits[0], low), min(limits[1], high)) |
---|
276 | |
---|
277 | return np.random.uniform(*limits) |
---|
278 | |
---|
279 | |
---|
280 | def randomize_pars(model_info, pars, seed=None): |
---|
281 | # type: (ModelInfo, ParameterSet, int) -> ParameterSet |
---|
282 | """ |
---|
283 | Generate random values for all of the parameters. |
---|
284 | |
---|
285 | Valid ranges for the random number generator are guessed from the name of |
---|
286 | the parameter; this will not account for constraints such as cap radius |
---|
287 | greater than cylinder radius in the capped_cylinder model, so |
---|
288 | :func:`constrain_pars` needs to be called afterward.. |
---|
289 | """ |
---|
290 | with push_seed(seed): |
---|
291 | # Note: the sort guarantees order `of calls to random number generator |
---|
292 | random_pars = dict((p, _randomize_one(model_info, p, v)) |
---|
293 | for p, v in sorted(pars.items())) |
---|
294 | return random_pars |
---|
295 | |
---|
296 | def constrain_pars(model_info, pars): |
---|
297 | # type: (ModelInfo, ParameterSet) -> None |
---|
298 | """ |
---|
299 | Restrict parameters to valid values. |
---|
300 | |
---|
301 | This includes model specific code for models such as capped_cylinder |
---|
302 | which need to support within model constraints (cap radius more than |
---|
303 | cylinder radius in this case). |
---|
304 | |
---|
305 | Warning: this updates the *pars* dictionary in place. |
---|
306 | """ |
---|
307 | name = model_info.id |
---|
308 | # if it is a product model, then just look at the form factor since |
---|
309 | # none of the structure factors need any constraints. |
---|
310 | if '*' in name: |
---|
311 | name = name.split('*')[0] |
---|
312 | |
---|
313 | if name == 'capped_cylinder' and pars['cap_radius'] < pars['radius']: |
---|
314 | pars['radius'], pars['cap_radius'] = pars['cap_radius'], pars['radius'] |
---|
315 | if name == 'barbell' and pars['bell_radius'] < pars['radius']: |
---|
316 | pars['radius'], pars['bell_radius'] = pars['bell_radius'], pars['radius'] |
---|
317 | |
---|
318 | # Limit guinier to an Rg such that Iq > 1e-30 (single precision cutoff) |
---|
319 | if name == 'guinier': |
---|
320 | #q_max = 0.2 # mid q maximum |
---|
321 | q_max = 1.0 # high q maximum |
---|
322 | rg_max = np.sqrt(90*np.log(10) + 3*np.log(pars['scale']))/q_max |
---|
323 | pars['rg'] = min(pars['rg'], rg_max) |
---|
324 | |
---|
325 | if name == 'rpa': |
---|
326 | # Make sure phi sums to 1.0 |
---|
327 | if pars['case_num'] < 2: |
---|
328 | pars['Phi1'] = 0. |
---|
329 | pars['Phi2'] = 0. |
---|
330 | elif pars['case_num'] < 5: |
---|
331 | pars['Phi1'] = 0. |
---|
332 | total = sum(pars['Phi'+c] for c in '1234') |
---|
333 | for c in '1234': |
---|
334 | pars['Phi'+c] /= total |
---|
335 | |
---|
336 | def parlist(model_info, pars, is2d): |
---|
337 | # type: (ModelInfo, ParameterSet, bool) -> str |
---|
338 | """ |
---|
339 | Format the parameter list for printing. |
---|
340 | """ |
---|
341 | lines = [] |
---|
342 | parameters = model_info.parameters |
---|
343 | for p in parameters.user_parameters(pars, is2d): |
---|
344 | fields = dict( |
---|
345 | value=pars.get(p.id, p.default), |
---|
346 | pd=pars.get(p.id+"_pd", 0.), |
---|
347 | n=int(pars.get(p.id+"_pd_n", 0)), |
---|
348 | nsigma=pars.get(p.id+"_pd_nsgima", 3.), |
---|
349 | pdtype=pars.get(p.id+"_pd_type", 'gaussian'), |
---|
350 | ) |
---|
351 | lines.append(_format_par(p.name, **fields)) |
---|
352 | return "\n".join(lines) |
---|
353 | |
---|
354 | #return "\n".join("%s: %s"%(p, v) for p, v in sorted(pars.items())) |
---|
355 | |
---|
356 | def _format_par(name, value=0., pd=0., n=0, nsigma=3., pdtype='gaussian'): |
---|
357 | # type: (str, float, float, int, float, str) -> str |
---|
358 | line = "%s: %g"%(name, value) |
---|
359 | if pd != 0. and n != 0: |
---|
360 | line += " +/- %g (%d points in [-%g,%g] sigma %s)"\ |
---|
361 | % (pd, n, nsigma, nsigma, pdtype) |
---|
362 | return line |
---|
363 | |
---|
364 | def suppress_pd(pars): |
---|
365 | # type: (ParameterSet) -> ParameterSet |
---|
366 | """ |
---|
367 | Suppress theta_pd for now until the normalization is resolved. |
---|
368 | |
---|
369 | May also suppress complete polydispersity of the model to test |
---|
370 | models more quickly. |
---|
371 | """ |
---|
372 | pars = pars.copy() |
---|
373 | for p in pars: |
---|
374 | if p.endswith("_pd_n"): pars[p] = 0 |
---|
375 | return pars |
---|
376 | |
---|
377 | def eval_sasview(model_info, data): |
---|
378 | # type: (Modelinfo, Data) -> Calculator |
---|
379 | """ |
---|
380 | Return a model calculator using the pre-4.0 SasView models. |
---|
381 | """ |
---|
382 | # importing sas here so that the error message will be that sas failed to |
---|
383 | # import rather than the more obscure smear_selection not imported error |
---|
384 | import sas |
---|
385 | import sas.models |
---|
386 | from sas.models.qsmearing import smear_selection |
---|
387 | from sas.models.MultiplicationModel import MultiplicationModel |
---|
388 | |
---|
389 | def get_model(name): |
---|
390 | # type: (str) -> "sas.models.BaseComponent" |
---|
391 | #print("new",sorted(_pars.items())) |
---|
392 | __import__('sas.models.' + name) |
---|
393 | ModelClass = getattr(getattr(sas.models, name, None), name, None) |
---|
394 | if ModelClass is None: |
---|
395 | raise ValueError("could not find model %r in sas.models"%name) |
---|
396 | return ModelClass() |
---|
397 | |
---|
398 | # grab the sasview model, or create it if it is a product model |
---|
399 | if model_info.composition: |
---|
400 | composition_type, parts = model_info.composition |
---|
401 | if composition_type == 'product': |
---|
402 | P, S = [get_model(revert_name(p)) for p in parts] |
---|
403 | model = MultiplicationModel(P, S) |
---|
404 | else: |
---|
405 | raise ValueError("sasview mixture models not supported by compare") |
---|
406 | else: |
---|
407 | model = get_model(revert_name(model_info)) |
---|
408 | |
---|
409 | # build a smearer with which to call the model, if necessary |
---|
410 | smearer = smear_selection(data, model=model) |
---|
411 | if hasattr(data, 'qx_data'): |
---|
412 | q = np.sqrt(data.qx_data**2 + data.qy_data**2) |
---|
413 | index = ((~data.mask) & (~np.isnan(data.data)) |
---|
414 | & (q >= data.qmin) & (q <= data.qmax)) |
---|
415 | if smearer is not None: |
---|
416 | smearer.model = model # because smear_selection has a bug |
---|
417 | smearer.accuracy = data.accuracy |
---|
418 | smearer.set_index(index) |
---|
419 | theory = lambda: smearer.get_value() |
---|
420 | else: |
---|
421 | theory = lambda: model.evalDistribution([data.qx_data[index], |
---|
422 | data.qy_data[index]]) |
---|
423 | elif smearer is not None: |
---|
424 | theory = lambda: smearer(model.evalDistribution(data.x)) |
---|
425 | else: |
---|
426 | theory = lambda: model.evalDistribution(data.x) |
---|
427 | |
---|
428 | def calculator(**pars): |
---|
429 | # type: (float, ...) -> np.ndarray |
---|
430 | """ |
---|
431 | Sasview calculator for model. |
---|
432 | """ |
---|
433 | # paying for parameter conversion each time to keep life simple, if not fast |
---|
434 | pars = revert_pars(model_info, pars) |
---|
435 | for k, v in pars.items(): |
---|
436 | name_attr = k.split('.') # polydispersity components |
---|
437 | if len(name_attr) == 2: |
---|
438 | model.dispersion[name_attr[0]][name_attr[1]] = v |
---|
439 | else: |
---|
440 | model.setParam(k, v) |
---|
441 | return theory() |
---|
442 | |
---|
443 | calculator.engine = "sasview" |
---|
444 | return calculator |
---|
445 | |
---|
446 | DTYPE_MAP = { |
---|
447 | 'half': '16', |
---|
448 | 'fast': 'fast', |
---|
449 | 'single': '32', |
---|
450 | 'double': '64', |
---|
451 | 'quad': '128', |
---|
452 | 'f16': '16', |
---|
453 | 'f32': '32', |
---|
454 | 'f64': '64', |
---|
455 | 'longdouble': '128', |
---|
456 | } |
---|
457 | def eval_opencl(model_info, data, dtype='single', cutoff=0.): |
---|
458 | # type: (ModelInfo, Data, str, float) -> Calculator |
---|
459 | """ |
---|
460 | Return a model calculator using the OpenCL calculation engine. |
---|
461 | """ |
---|
462 | try: |
---|
463 | model = core.build_model(model_info, dtype=dtype, platform="ocl") |
---|
464 | except Exception as exc: |
---|
465 | print(exc) |
---|
466 | print("... trying again with single precision") |
---|
467 | model = core.build_model(model_info, dtype='single', platform="ocl") |
---|
468 | calculator = DirectModel(data, model, cutoff=cutoff) |
---|
469 | calculator.engine = "OCL%s"%DTYPE_MAP[dtype] |
---|
470 | return calculator |
---|
471 | |
---|
472 | def eval_ctypes(model_info, data, dtype='double', cutoff=0.): |
---|
473 | # type: (ModelInfo, Data, str, float) -> Calculator |
---|
474 | """ |
---|
475 | Return a model calculator using the DLL calculation engine. |
---|
476 | """ |
---|
477 | model = core.build_model(model_info, dtype=dtype, platform="dll") |
---|
478 | calculator = DirectModel(data, model, cutoff=cutoff) |
---|
479 | calculator.engine = "OMP%s"%DTYPE_MAP[dtype] |
---|
480 | return calculator |
---|
481 | |
---|
482 | def time_calculation(calculator, pars, Nevals=1): |
---|
483 | # type: (Calculator, ParameterSet, int) -> Tuple[np.ndarray, float] |
---|
484 | """ |
---|
485 | Compute the average calculation time over N evaluations. |
---|
486 | |
---|
487 | An additional call is generated without polydispersity in order to |
---|
488 | initialize the calculation engine, and make the average more stable. |
---|
489 | """ |
---|
490 | # initialize the code so time is more accurate |
---|
491 | if Nevals > 1: |
---|
492 | calculator(**suppress_pd(pars)) |
---|
493 | toc = tic() |
---|
494 | # make sure there is at least one eval |
---|
495 | value = calculator(**pars) |
---|
496 | for _ in range(Nevals-1): |
---|
497 | value = calculator(**pars) |
---|
498 | average_time = toc()*1000./Nevals |
---|
499 | #print("I(q)",value) |
---|
500 | return value, average_time |
---|
501 | |
---|
502 | def make_data(opts): |
---|
503 | # type: (Dict[str, Any]) -> Tuple[Data, np.ndarray] |
---|
504 | """ |
---|
505 | Generate an empty dataset, used with the model to set Q points |
---|
506 | and resolution. |
---|
507 | |
---|
508 | *opts* contains the options, with 'qmax', 'nq', 'res', |
---|
509 | 'accuracy', 'is2d' and 'view' parsed from the command line. |
---|
510 | """ |
---|
511 | qmax, nq, res = opts['qmax'], opts['nq'], opts['res'] |
---|
512 | if opts['is2d']: |
---|
513 | q = np.linspace(-qmax, qmax, nq) # type: np.ndarray |
---|
514 | data = empty_data2D(q, resolution=res) |
---|
515 | data.accuracy = opts['accuracy'] |
---|
516 | set_beam_stop(data, 0.0004) |
---|
517 | index = ~data.mask |
---|
518 | else: |
---|
519 | if opts['view'] == 'log' and not opts['zero']: |
---|
520 | qmax = math.log10(qmax) |
---|
521 | q = np.logspace(qmax-3, qmax, nq) |
---|
522 | else: |
---|
523 | q = np.linspace(0.001*qmax, qmax, nq) |
---|
524 | if opts['zero']: |
---|
525 | q = np.hstack((0, q)) |
---|
526 | data = empty_data1D(q, resolution=res) |
---|
527 | index = slice(None, None) |
---|
528 | return data, index |
---|
529 | |
---|
530 | def make_engine(model_info, data, dtype, cutoff): |
---|
531 | # type: (ModelInfo, Data, str, float) -> Calculator |
---|
532 | """ |
---|
533 | Generate the appropriate calculation engine for the given datatype. |
---|
534 | |
---|
535 | Datatypes with '!' appended are evaluated using external C DLLs rather |
---|
536 | than OpenCL. |
---|
537 | """ |
---|
538 | if dtype == 'sasview': |
---|
539 | return eval_sasview(model_info, data) |
---|
540 | elif dtype.endswith('!'): |
---|
541 | return eval_ctypes(model_info, data, dtype=dtype[:-1], cutoff=cutoff) |
---|
542 | else: |
---|
543 | return eval_opencl(model_info, data, dtype=dtype, cutoff=cutoff) |
---|
544 | |
---|
545 | def _show_invalid(data, theory): |
---|
546 | # type: (Data, np.ma.ndarray) -> None |
---|
547 | """ |
---|
548 | Display a list of the non-finite values in theory. |
---|
549 | """ |
---|
550 | if not theory.mask.any(): |
---|
551 | return |
---|
552 | |
---|
553 | if hasattr(data, 'x'): |
---|
554 | bad = zip(data.x[theory.mask], theory[theory.mask]) |
---|
555 | print(" *** ", ", ".join("I(%g)=%g"%(x, y) for x, y in bad)) |
---|
556 | |
---|
557 | |
---|
558 | def compare(opts, limits=None): |
---|
559 | # type: (Dict[str, Any], Optional[Tuple[float, float]]) -> Tuple[float, float] |
---|
560 | """ |
---|
561 | Preform a comparison using options from the command line. |
---|
562 | |
---|
563 | *limits* are the limits on the values to use, either to set the y-axis |
---|
564 | for 1D or to set the colormap scale for 2D. If None, then they are |
---|
565 | inferred from the data and returned. When exploring using Bumps, |
---|
566 | the limits are set when the model is initially called, and maintained |
---|
567 | as the values are adjusted, making it easier to see the effects of the |
---|
568 | parameters. |
---|
569 | """ |
---|
570 | Nbase, Ncomp = opts['n1'], opts['n2'] |
---|
571 | pars = opts['pars'] |
---|
572 | data = opts['data'] |
---|
573 | |
---|
574 | # silence the linter |
---|
575 | base = opts['engines'][0] if Nbase else None |
---|
576 | comp = opts['engines'][1] if Ncomp else None |
---|
577 | base_time = comp_time = None |
---|
578 | base_value = comp_value = resid = relerr = None |
---|
579 | |
---|
580 | # Base calculation |
---|
581 | if Nbase > 0: |
---|
582 | try: |
---|
583 | base_raw, base_time = time_calculation(base, pars, Nbase) |
---|
584 | base_value = np.ma.masked_invalid(base_raw) |
---|
585 | print("%s t=%.2f ms, intensity=%.0f" |
---|
586 | % (base.engine, base_time, base_value.sum())) |
---|
587 | _show_invalid(data, base_value) |
---|
588 | except ImportError: |
---|
589 | traceback.print_exc() |
---|
590 | Nbase = 0 |
---|
591 | |
---|
592 | # Comparison calculation |
---|
593 | if Ncomp > 0: |
---|
594 | try: |
---|
595 | comp_raw, comp_time = time_calculation(comp, pars, Ncomp) |
---|
596 | comp_value = np.ma.masked_invalid(comp_raw) |
---|
597 | print("%s t=%.2f ms, intensity=%.0f" |
---|
598 | % (comp.engine, comp_time, comp_value.sum())) |
---|
599 | _show_invalid(data, comp_value) |
---|
600 | except ImportError: |
---|
601 | traceback.print_exc() |
---|
602 | Ncomp = 0 |
---|
603 | |
---|
604 | # Compare, but only if computing both forms |
---|
605 | if Nbase > 0 and Ncomp > 0: |
---|
606 | resid = (base_value - comp_value) |
---|
607 | relerr = resid/np.where(comp_value!=0., abs(comp_value), 1.0) |
---|
608 | _print_stats("|%s-%s|" |
---|
609 | % (base.engine, comp.engine) + (" "*(3+len(comp.engine))), |
---|
610 | resid) |
---|
611 | _print_stats("|(%s-%s)/%s|" |
---|
612 | % (base.engine, comp.engine, comp.engine), |
---|
613 | relerr) |
---|
614 | |
---|
615 | # Plot if requested |
---|
616 | if not opts['plot'] and not opts['explore']: return |
---|
617 | view = opts['view'] |
---|
618 | import matplotlib.pyplot as plt |
---|
619 | if limits is None: |
---|
620 | vmin, vmax = np.Inf, -np.Inf |
---|
621 | if Nbase > 0: |
---|
622 | vmin = min(vmin, base_value.min()) |
---|
623 | vmax = max(vmax, base_value.max()) |
---|
624 | if Ncomp > 0: |
---|
625 | vmin = min(vmin, comp_value.min()) |
---|
626 | vmax = max(vmax, comp_value.max()) |
---|
627 | limits = vmin, vmax |
---|
628 | |
---|
629 | if Nbase > 0: |
---|
630 | if Ncomp > 0: plt.subplot(131) |
---|
631 | plot_theory(data, base_value, view=view, use_data=False, limits=limits) |
---|
632 | plt.title("%s t=%.2f ms"%(base.engine, base_time)) |
---|
633 | #cbar_title = "log I" |
---|
634 | if Ncomp > 0: |
---|
635 | if Nbase > 0: plt.subplot(132) |
---|
636 | plot_theory(data, comp_value, view=view, use_data=False, limits=limits) |
---|
637 | plt.title("%s t=%.2f ms"%(comp.engine, comp_time)) |
---|
638 | #cbar_title = "log I" |
---|
639 | if Ncomp > 0 and Nbase > 0: |
---|
640 | plt.subplot(133) |
---|
641 | if not opts['rel_err']: |
---|
642 | err, errstr, errview = resid, "abs err", "linear" |
---|
643 | else: |
---|
644 | err, errstr, errview = abs(relerr), "rel err", "log" |
---|
645 | #err,errstr = base/comp,"ratio" |
---|
646 | plot_theory(data, None, resid=err, view=errview, use_data=False) |
---|
647 | if view == 'linear': |
---|
648 | plt.xscale('linear') |
---|
649 | plt.title("max %s = %.3g"%(errstr, abs(err).max())) |
---|
650 | #cbar_title = errstr if errview=="linear" else "log "+errstr |
---|
651 | #if is2D: |
---|
652 | # h = plt.colorbar() |
---|
653 | # h.ax.set_title(cbar_title) |
---|
654 | |
---|
655 | if Ncomp > 0 and Nbase > 0 and '-hist' in opts: |
---|
656 | plt.figure() |
---|
657 | v = relerr |
---|
658 | v[v == 0] = 0.5*np.min(np.abs(v[v != 0])) |
---|
659 | plt.hist(np.log10(np.abs(v)), normed=1, bins=50) |
---|
660 | plt.xlabel('log10(err), err = |(%s - %s) / %s|' |
---|
661 | % (base.engine, comp.engine, comp.engine)) |
---|
662 | plt.ylabel('P(err)') |
---|
663 | plt.title('Distribution of relative error between calculation engines') |
---|
664 | |
---|
665 | if not opts['explore']: |
---|
666 | plt.show() |
---|
667 | |
---|
668 | return limits |
---|
669 | |
---|
670 | def _print_stats(label, err): |
---|
671 | # type: (str, np.ma.ndarray) -> None |
---|
672 | # work with trimmed data, not the full set |
---|
673 | sorted_err = np.sort(abs(err.compressed())) |
---|
674 | p50 = int((len(sorted_err)-1)*0.50) |
---|
675 | p98 = int((len(sorted_err)-1)*0.98) |
---|
676 | data = [ |
---|
677 | "max:%.3e"%sorted_err[-1], |
---|
678 | "median:%.3e"%sorted_err[p50], |
---|
679 | "98%%:%.3e"%sorted_err[p98], |
---|
680 | "rms:%.3e"%np.sqrt(np.mean(sorted_err**2)), |
---|
681 | "zero-offset:%+.3e"%np.mean(sorted_err), |
---|
682 | ] |
---|
683 | print(label+" "+" ".join(data)) |
---|
684 | |
---|
685 | |
---|
686 | |
---|
687 | # =========================================================================== |
---|
688 | # |
---|
689 | NAME_OPTIONS = set([ |
---|
690 | 'plot', 'noplot', |
---|
691 | 'half', 'fast', 'single', 'double', |
---|
692 | 'single!', 'double!', 'quad!', 'sasview', |
---|
693 | 'lowq', 'midq', 'highq', 'exq', 'zero', |
---|
694 | '2d', '1d', |
---|
695 | 'preset', 'random', |
---|
696 | 'poly', 'mono', |
---|
697 | 'nopars', 'pars', |
---|
698 | 'rel', 'abs', |
---|
699 | 'linear', 'log', 'q4', |
---|
700 | 'hist', 'nohist', |
---|
701 | 'edit', |
---|
702 | 'demo', 'default', |
---|
703 | ]) |
---|
704 | VALUE_OPTIONS = [ |
---|
705 | # Note: random is both a name option and a value option |
---|
706 | 'cutoff', 'random', 'nq', 'res', 'accuracy', |
---|
707 | ] |
---|
708 | |
---|
709 | def columnize(L, indent="", width=79): |
---|
710 | # type: (List[str], str, int) -> str |
---|
711 | """ |
---|
712 | Format a list of strings into columns. |
---|
713 | |
---|
714 | Returns a string with carriage returns ready for printing. |
---|
715 | """ |
---|
716 | column_width = max(len(w) for w in L) + 1 |
---|
717 | num_columns = (width - len(indent)) // column_width |
---|
718 | num_rows = len(L) // num_columns |
---|
719 | L = L + [""] * (num_rows*num_columns - len(L)) |
---|
720 | columns = [L[k*num_rows:(k+1)*num_rows] for k in range(num_columns)] |
---|
721 | lines = [" ".join("%-*s"%(column_width, entry) for entry in row) |
---|
722 | for row in zip(*columns)] |
---|
723 | output = indent + ("\n"+indent).join(lines) |
---|
724 | return output |
---|
725 | |
---|
726 | |
---|
727 | def get_pars(model_info, use_demo=False): |
---|
728 | # type: (ModelInfo, bool) -> ParameterSet |
---|
729 | """ |
---|
730 | Extract demo parameters from the model definition. |
---|
731 | """ |
---|
732 | # Get the default values for the parameters |
---|
733 | pars = {} |
---|
734 | for p in model_info.parameters.call_parameters: |
---|
735 | parts = [('', p.default)] |
---|
736 | if p.polydisperse: |
---|
737 | parts.append(('_pd', 0.0)) |
---|
738 | parts.append(('_pd_n', 0)) |
---|
739 | parts.append(('_pd_nsigma', 3.0)) |
---|
740 | parts.append(('_pd_type', "gaussian")) |
---|
741 | for ext, val in parts: |
---|
742 | if p.length > 1: |
---|
743 | dict(("%s%d%s"%(p.id,k,ext), val) for k in range(p.length)) |
---|
744 | else: |
---|
745 | pars[p.id+ext] = val |
---|
746 | |
---|
747 | # Plug in values given in demo |
---|
748 | if use_demo: |
---|
749 | pars.update(model_info.demo) |
---|
750 | return pars |
---|
751 | |
---|
752 | |
---|
753 | def parse_opts(): |
---|
754 | # type: () -> Dict[str, Any] |
---|
755 | """ |
---|
756 | Parse command line options. |
---|
757 | """ |
---|
758 | MODELS = core.list_models() |
---|
759 | flags = [arg for arg in sys.argv[1:] |
---|
760 | if arg.startswith('-')] |
---|
761 | values = [arg for arg in sys.argv[1:] |
---|
762 | if not arg.startswith('-') and '=' in arg] |
---|
763 | args = [arg for arg in sys.argv[1:] |
---|
764 | if not arg.startswith('-') and '=' not in arg] |
---|
765 | models = "\n ".join("%-15s"%v for v in MODELS) |
---|
766 | if len(args) == 0: |
---|
767 | print(USAGE) |
---|
768 | print("\nAvailable models:") |
---|
769 | print(columnize(MODELS, indent=" ")) |
---|
770 | sys.exit(1) |
---|
771 | if len(args) > 3: |
---|
772 | print("expected parameters: model N1 N2") |
---|
773 | |
---|
774 | name = args[0] |
---|
775 | try: |
---|
776 | model_info = core.load_model_info(name) |
---|
777 | except ImportError as exc: |
---|
778 | print(str(exc)) |
---|
779 | print("Could not find model; use one of:\n " + models) |
---|
780 | sys.exit(1) |
---|
781 | |
---|
782 | invalid = [o[1:] for o in flags |
---|
783 | if o[1:] not in NAME_OPTIONS |
---|
784 | and not any(o.startswith('-%s='%t) for t in VALUE_OPTIONS)] |
---|
785 | if invalid: |
---|
786 | print("Invalid options: %s"%(", ".join(invalid))) |
---|
787 | sys.exit(1) |
---|
788 | |
---|
789 | |
---|
790 | # pylint: disable=bad-whitespace |
---|
791 | # Interpret the flags |
---|
792 | opts = { |
---|
793 | 'plot' : True, |
---|
794 | 'view' : 'log', |
---|
795 | 'is2d' : False, |
---|
796 | 'qmax' : 0.05, |
---|
797 | 'nq' : 128, |
---|
798 | 'res' : 0.0, |
---|
799 | 'accuracy' : 'Low', |
---|
800 | 'cutoff' : 0.0, |
---|
801 | 'seed' : -1, # default to preset |
---|
802 | 'mono' : False, |
---|
803 | 'show_pars' : False, |
---|
804 | 'show_hist' : False, |
---|
805 | 'rel_err' : True, |
---|
806 | 'explore' : False, |
---|
807 | 'use_demo' : True, |
---|
808 | 'zero' : False, |
---|
809 | } |
---|
810 | engines = [] |
---|
811 | for arg in flags: |
---|
812 | if arg == '-noplot': opts['plot'] = False |
---|
813 | elif arg == '-plot': opts['plot'] = True |
---|
814 | elif arg == '-linear': opts['view'] = 'linear' |
---|
815 | elif arg == '-log': opts['view'] = 'log' |
---|
816 | elif arg == '-q4': opts['view'] = 'q4' |
---|
817 | elif arg == '-1d': opts['is2d'] = False |
---|
818 | elif arg == '-2d': opts['is2d'] = True |
---|
819 | elif arg == '-exq': opts['qmax'] = 10.0 |
---|
820 | elif arg == '-highq': opts['qmax'] = 1.0 |
---|
821 | elif arg == '-midq': opts['qmax'] = 0.2 |
---|
822 | elif arg == '-lowq': opts['qmax'] = 0.05 |
---|
823 | elif arg == '-zero': opts['zero'] = True |
---|
824 | elif arg.startswith('-nq='): opts['nq'] = int(arg[4:]) |
---|
825 | elif arg.startswith('-res='): opts['res'] = float(arg[5:]) |
---|
826 | elif arg.startswith('-accuracy='): opts['accuracy'] = arg[10:] |
---|
827 | elif arg.startswith('-cutoff='): opts['cutoff'] = float(arg[8:]) |
---|
828 | elif arg.startswith('-random='): opts['seed'] = int(arg[8:]) |
---|
829 | elif arg == '-random': opts['seed'] = np.random.randint(1000000) |
---|
830 | elif arg == '-preset': opts['seed'] = -1 |
---|
831 | elif arg == '-mono': opts['mono'] = True |
---|
832 | elif arg == '-poly': opts['mono'] = False |
---|
833 | elif arg == '-pars': opts['show_pars'] = True |
---|
834 | elif arg == '-nopars': opts['show_pars'] = False |
---|
835 | elif arg == '-hist': opts['show_hist'] = True |
---|
836 | elif arg == '-nohist': opts['show_hist'] = False |
---|
837 | elif arg == '-rel': opts['rel_err'] = True |
---|
838 | elif arg == '-abs': opts['rel_err'] = False |
---|
839 | elif arg == '-half': engines.append(arg[1:]) |
---|
840 | elif arg == '-fast': engines.append(arg[1:]) |
---|
841 | elif arg == '-single': engines.append(arg[1:]) |
---|
842 | elif arg == '-double': engines.append(arg[1:]) |
---|
843 | elif arg == '-single!': engines.append(arg[1:]) |
---|
844 | elif arg == '-double!': engines.append(arg[1:]) |
---|
845 | elif arg == '-quad!': engines.append(arg[1:]) |
---|
846 | elif arg == '-sasview': engines.append(arg[1:]) |
---|
847 | elif arg == '-edit': opts['explore'] = True |
---|
848 | elif arg == '-demo': opts['use_demo'] = True |
---|
849 | elif arg == '-default': opts['use_demo'] = False |
---|
850 | # pylint: enable=bad-whitespace |
---|
851 | |
---|
852 | if len(engines) == 0: |
---|
853 | engines.extend(['single', 'double']) |
---|
854 | elif len(engines) == 1: |
---|
855 | if engines[0][0] != 'double': |
---|
856 | engines.append('double') |
---|
857 | else: |
---|
858 | engines.append('single') |
---|
859 | elif len(engines) > 2: |
---|
860 | del engines[2:] |
---|
861 | |
---|
862 | n1 = int(args[1]) if len(args) > 1 else 1 |
---|
863 | n2 = int(args[2]) if len(args) > 2 else 1 |
---|
864 | use_sasview = any(engine=='sasview' and count>0 |
---|
865 | for engine, count in zip(engines, [n1, n2])) |
---|
866 | |
---|
867 | # Get demo parameters from model definition, or use default parameters |
---|
868 | # if model does not define demo parameters |
---|
869 | pars = get_pars(model_info, opts['use_demo']) |
---|
870 | |
---|
871 | |
---|
872 | # Fill in parameters given on the command line |
---|
873 | presets = {} |
---|
874 | for arg in values: |
---|
875 | k, v = arg.split('=', 1) |
---|
876 | if k not in pars: |
---|
877 | # extract base name without polydispersity info |
---|
878 | s = set(p.split('_pd')[0] for p in pars) |
---|
879 | print("%r invalid; parameters are: %s"%(k, ", ".join(sorted(s)))) |
---|
880 | sys.exit(1) |
---|
881 | presets[k] = float(v) if not k.endswith('type') else v |
---|
882 | |
---|
883 | # randomize parameters |
---|
884 | #pars.update(set_pars) # set value before random to control range |
---|
885 | if opts['seed'] > -1: |
---|
886 | pars = randomize_pars(model_info, pars, seed=opts['seed']) |
---|
887 | print("Randomize using -random=%i"%opts['seed']) |
---|
888 | if opts['mono']: |
---|
889 | pars = suppress_pd(pars) |
---|
890 | pars.update(presets) # set value after random to control value |
---|
891 | #import pprint; pprint.pprint(model_info) |
---|
892 | constrain_pars(model_info, pars) |
---|
893 | if use_sasview: |
---|
894 | constrain_new_to_old(model_info, pars) |
---|
895 | if opts['show_pars']: |
---|
896 | print(str(parlist(model_info, pars, opts['is2d']))) |
---|
897 | |
---|
898 | # Create the computational engines |
---|
899 | data, _ = make_data(opts) |
---|
900 | if n1: |
---|
901 | base = make_engine(model_info, data, engines[0], opts['cutoff']) |
---|
902 | else: |
---|
903 | base = None |
---|
904 | if n2: |
---|
905 | comp = make_engine(model_info, data, engines[1], opts['cutoff']) |
---|
906 | else: |
---|
907 | comp = None |
---|
908 | |
---|
909 | # pylint: disable=bad-whitespace |
---|
910 | # Remember it all |
---|
911 | opts.update({ |
---|
912 | 'name' : name, |
---|
913 | 'def' : model_info, |
---|
914 | 'n1' : n1, |
---|
915 | 'n2' : n2, |
---|
916 | 'presets' : presets, |
---|
917 | 'pars' : pars, |
---|
918 | 'data' : data, |
---|
919 | 'engines' : [base, comp], |
---|
920 | }) |
---|
921 | # pylint: enable=bad-whitespace |
---|
922 | |
---|
923 | return opts |
---|
924 | |
---|
925 | def explore(opts): |
---|
926 | # type: (Dict[str, Any]) -> None |
---|
927 | """ |
---|
928 | Explore the model using the Bumps GUI. |
---|
929 | """ |
---|
930 | import wx # type: ignore |
---|
931 | from bumps.names import FitProblem # type: ignore |
---|
932 | from bumps.gui.app_frame import AppFrame # type: ignore |
---|
933 | |
---|
934 | problem = FitProblem(Explore(opts)) |
---|
935 | is_mac = "cocoa" in wx.version() |
---|
936 | app = wx.App() |
---|
937 | frame = AppFrame(parent=None, title="explore") |
---|
938 | if not is_mac: frame.Show() |
---|
939 | frame.panel.set_model(model=problem) |
---|
940 | frame.panel.Layout() |
---|
941 | frame.panel.aui.Split(0, wx.TOP) |
---|
942 | if is_mac: frame.Show() |
---|
943 | app.MainLoop() |
---|
944 | |
---|
945 | class Explore(object): |
---|
946 | """ |
---|
947 | Bumps wrapper for a SAS model comparison. |
---|
948 | |
---|
949 | The resulting object can be used as a Bumps fit problem so that |
---|
950 | parameters can be adjusted in the GUI, with plots updated on the fly. |
---|
951 | """ |
---|
952 | def __init__(self, opts): |
---|
953 | # type: (Dict[str, Any]) -> None |
---|
954 | from bumps.cli import config_matplotlib # type: ignore |
---|
955 | from . import bumps_model |
---|
956 | config_matplotlib() |
---|
957 | self.opts = opts |
---|
958 | model_info = opts['def'] |
---|
959 | pars, pd_types = bumps_model.create_parameters(model_info, **opts['pars']) |
---|
960 | # Initialize parameter ranges, fixing the 2D parameters for 1D data. |
---|
961 | if not opts['is2d']: |
---|
962 | for p in model_info.parameters.user_parameters(is2d=False): |
---|
963 | for ext in ['', '_pd', '_pd_n', '_pd_nsigma']: |
---|
964 | k = p.name+ext |
---|
965 | v = pars.get(k, None) |
---|
966 | if v is not None: |
---|
967 | v.range(*parameter_range(k, v.value)) |
---|
968 | else: |
---|
969 | for k, v in pars.items(): |
---|
970 | v.range(*parameter_range(k, v.value)) |
---|
971 | |
---|
972 | self.pars = pars |
---|
973 | self.pd_types = pd_types |
---|
974 | self.limits = None |
---|
975 | |
---|
976 | def numpoints(self): |
---|
977 | # type: () -> int |
---|
978 | """ |
---|
979 | Return the number of points. |
---|
980 | """ |
---|
981 | return len(self.pars) + 1 # so dof is 1 |
---|
982 | |
---|
983 | def parameters(self): |
---|
984 | # type: () -> Any # Dict/List hierarchy of parameters |
---|
985 | """ |
---|
986 | Return a dictionary of parameters. |
---|
987 | """ |
---|
988 | return self.pars |
---|
989 | |
---|
990 | def nllf(self): |
---|
991 | # type: () -> float |
---|
992 | """ |
---|
993 | Return cost. |
---|
994 | """ |
---|
995 | # pylint: disable=no-self-use |
---|
996 | return 0. # No nllf |
---|
997 | |
---|
998 | def plot(self, view='log'): |
---|
999 | # type: (str) -> None |
---|
1000 | """ |
---|
1001 | Plot the data and residuals. |
---|
1002 | """ |
---|
1003 | pars = dict((k, v.value) for k, v in self.pars.items()) |
---|
1004 | pars.update(self.pd_types) |
---|
1005 | self.opts['pars'] = pars |
---|
1006 | limits = compare(self.opts, limits=self.limits) |
---|
1007 | if self.limits is None: |
---|
1008 | vmin, vmax = limits |
---|
1009 | self.limits = vmax*1e-7, 1.3*vmax |
---|
1010 | |
---|
1011 | |
---|
1012 | def main(): |
---|
1013 | # type: () -> None |
---|
1014 | """ |
---|
1015 | Main program. |
---|
1016 | """ |
---|
1017 | opts = parse_opts() |
---|
1018 | if opts['explore']: |
---|
1019 | explore(opts) |
---|
1020 | else: |
---|
1021 | compare(opts) |
---|
1022 | |
---|
1023 | if __name__ == "__main__": |
---|
1024 | main() |
---|