Changes in / [599993b9:2a12d8d8] in sasmodels


Ignore:
Files:
1 deleted
6 edited

Legend:

Unmodified
Added
Removed

  • doc/guide/plugin.rst

    r57c609b r2015f02  
    428428        def random(): 
    429429        ... 
    430  
    431 This function provides a model-specific random parameter set which shows model 
    432 features in the USANS to SANS range.  For example, core-shell sphere sets the 
    433 outer radius of the sphere logarithmically in `[20, 20,000]`, which sets the Q 
    434 value for the transition from flat to falling.  It then uses a beta distribution 
    435 to set the percentage of the shape which is shell, giving a preference for very 
    436 thin or very thick shells (but never 0% or 100%).  Using `-sets=10` in sascomp 
    437 should show a reasonable variety of curves over the default sascomp q range. 
    438 The parameter set is returned as a dictionary of `{parameter: value, ...}`. 
    439 Any model parameters not included in the dictionary will default according to 
     430         
     431This function provides a model-specific random parameter set which shows model  
     432features in the USANS to SANS range.  For example, core-shell sphere sets the  
     433outer radius of the sphere logarithmically in `[20, 20,000]`, which sets the Q  
     434value for the transition from flat to falling.  It then uses a beta distribution  
     435to set the percentage of the shape which is shell, giving a preference for very  
     436thin or very thick shells (but never 0% or 100%).  Using `-sets=10` in sascomp  
     437should show a reasonable variety of curves over the default sascomp q range.   
     438The parameter set is returned as a dictionary of `{parameter: value, ...}`.   
     439Any model parameters not included in the dictionary will default according to  
    440440the code in the `_randomize_one()` function from sasmodels/compare.py. 
    441441 
    … …  
    701701    erf, erfc, tgamma, lgamma:  **do not use** 
    702702        Special functions that should be part of the standard, but are missing 
    703         or inaccurate on some platforms. Use sas_erf, sas_erfc, sas_gamma 
    704         and sas_lgamma instead (see below). 
     703        or inaccurate on some platforms. Use sas_erf, sas_erfc and sas_gamma 
     704        instead (see below). Note: lgamma(x) has not yet been tested. 
    705705 
    706706Some non-standard constants and functions are also provided: 
    … …  
    769769        Gamma function sas_gamma\ $(x) = \Gamma(x)$. 
    770770 
    771         The standard math function, tgamma(x), is unstable for $x < 1$ 
     771        The standard math function, tgamma(x) is unstable for $x < 1$ 
    772772        on some platforms. 
    773773 
    774774        :code:`source = ["lib/sas_gamma.c", ...]` 
    775775        (`sas_gamma.c <https://github.com/SasView/sasmodels/tree/master/sasmodels/models/lib/sas_gamma.c>`_) 
    776  
    777     sas_gammaln(x): 
    778         log gamma function sas_gammaln\ $(x) = \log \Gamma(|x|)$. 
    779  
    780         The standard math function, lgamma(x), is incorrect for single 
    781         precision on some platforms. 
    782  
    783         :code:`source = ["lib/sas_gammainc.c", ...]` 
    784         (`sas_gammainc.c <https://github.com/SasView/sasmodels/tree/master/sasmodels/models/lib/sas_gammainc.c>`_) 
    785  
    786     sas_gammainc(a, x), sas_gammaincc(a, x): 
    787         Incomplete gamma function 
    788         sas_gammainc\ $(a, x) = \int_0^x t^{a-1}e^{-t}\,dt / \Gamma(a)$ 
    789         and complementary incomplete gamma function 
    790         sas_gammaincc\ $(a, x) = \int_x^\infty t^{a-1}e^{-t}\,dt / \Gamma(a)$ 
    791  
    792         :code:`source = ["lib/sas_gammainc.c", ...]` 
    793         (`sas_gammainc.c <https://github.com/SasView/sasmodels/tree/master/sasmodels/models/lib/sas_gammainc.c>`_) 
    794776 
    795777    sas_erf(x), sas_erfc(x): 
    … …  
    829811        If $n$ = 0 or 1, it uses sas_J0($x$) or sas_J1($x$), respectively. 
    830812 
    831         Warning: JN(n,x) can be very inaccurate (0.1%) for x not in [0.1, 100]. 
    832  
    833813        The standard math function jn(n, x) is not available on all platforms. 
    834814 
    … …  
    839819        Sine integral Si\ $(x) = \int_0^x \tfrac{\sin t}{t}\,dt$. 
    840820 
    841         Warning: Si(x) can be very inaccurate (0.1%) for x in [0.1, 100]. 
    842  
    843821        This function uses Taylor series for small and large arguments: 
    844822 
    845         For large arguments use the following Taylor series, 
     823        For large arguments, 
    846824 
    847825        .. math:: 
    … …  
    851829             - \frac{\sin(x)}{x}\left(\frac{1}{x} - \frac{3!}{x^3} + \frac{5!}{x^5} - \frac{7!}{x^7}\right) 
    852830 
    853         For small arguments , 
     831        For small arguments, 
    854832 
    855833        .. math:: 

  • explore/precision.py

    rfba9ca0 r2a7e20e  
    9595            neg:    [-100,100] 
    9696 
    97         For arbitrary range use "start:stop:steps:scale" where scale is 
    98         one of log, lin, or linear. 
    99  
    10097        *diff* is "relative", "absolute" or "none" 
    10198 
    … …  
    105102        linear = not xrange.startswith("log") 
    106103        if xrange == "zoom": 
    107             start, stop, steps = 1000, 1010, 2000 
     104            lin_min, lin_max, lin_steps = 1000, 1010, 2000 
    108105        elif xrange == "neg": 
    109             start, stop, steps = -100.1, 100.1, 2000 
     106            lin_min, lin_max, lin_steps = -100.1, 100.1, 2000 
    110107        elif xrange == "linear": 
    111             start, stop, steps = 1, 1000, 2000 
    112             start, stop, steps = 0.001, 2, 2000 
     108            lin_min, lin_max, lin_steps = 1, 1000, 2000 
     109            lin_min, lin_max, lin_steps = 0.001, 2, 2000 
    113110        elif xrange == "log": 
    114             start, stop, steps = -3, 5, 400 
     111            log_min, log_max, log_steps = -3, 5, 400 
    115112        elif xrange == "logq": 
    116             start, stop, steps = -4, 1, 400 
    117         elif ':' in xrange: 
    118             parts = xrange.split(':') 
    119             linear = parts[3] != "log" if len(parts) == 4 else True 
    120             steps = int(parts[2]) if len(parts) > 2 else 400 
    121             start = float(parts[0]) 
    122             stop = float(parts[1]) 
    123  
     113            log_min, log_max, log_steps = -4, 1, 400 
    124114        else: 
    125115            raise ValueError("unknown range "+xrange) 
    … …  
    131121            # value to x in the given precision. 
    132122            if linear: 
    133                 start = max(start, self.limits[0]) 
    134                 stop = min(stop, self.limits[1]) 
    135                 qrf = np.linspace(start, stop, steps, dtype='single') 
    136                 #qrf = np.linspace(start, stop, steps, dtype='double') 
     123                lin_min = max(lin_min, self.limits[0]) 
     124                lin_max = min(lin_max, self.limits[1]) 
     125                qrf = np.linspace(lin_min, lin_max, lin_steps, dtype='single') 
     126                #qrf = np.linspace(lin_min, lin_max, lin_steps, dtype='double') 
    137127                qr = [mp.mpf(float(v)) for v in qrf] 
    138                 #qr = mp.linspace(start, stop, steps) 
     128                #qr = mp.linspace(lin_min, lin_max, lin_steps) 
    139129            else: 
    140                 start = np.log10(max(10**start, self.limits[0])) 
    141                 stop = np.log10(min(10**stop, self.limits[1])) 
    142                 qrf = np.logspace(start, stop, steps, dtype='single') 
    143                 #qrf = np.logspace(start, stop, steps, dtype='double') 
     130                log_min = np.log10(max(10**log_min, self.limits[0])) 
     131                log_max = np.log10(min(10**log_max, self.limits[1])) 
     132                qrf = np.logspace(log_min, log_max, log_steps, dtype='single') 
     133                #qrf = np.logspace(log_min, log_max, log_steps, dtype='double') 
    144134                qr = [mp.mpf(float(v)) for v in qrf] 
    145                 #qr = [10**v for v in mp.linspace(start, stop, steps)] 
     135                #qr = [10**v for v in mp.linspace(log_min, log_max, log_steps)] 
    146136 
    147137        target = self.call_mpmath(qr, bits=500) 
    … …  
    186176    """ 
    187177    if diff == "relative": 
    188         err = np.array([(abs((t-a)/t) if t != 0 else a) for t, a in zip(target, actual)], 'd') 
     178        err = np.array([abs((t-a)/t) for t, a in zip(target, actual)], 'd') 
    189179        #err = np.clip(err, 0, 1) 
    190180        pylab.loglog(x, err, '-', label=label) 
    … …  
    207197    return model_info 
    208198 
    209 # Hack to allow second parameter A in two parameter functions 
    210 A = 1 
    211 def parse_extra_pars(): 
    212     global A 
    213  
    214     A_str = str(A) 
    215     pop = [] 
    216     for k, v in enumerate(sys.argv[1:]): 
    217         if v.startswith("A="): 
    218             A_str = v[2:] 
    219             pop.append(k+1) 
    220     if pop: 
    221         sys.argv = [v for k, v in enumerate(sys.argv) if k not in pop] 
    222         A = float(A_str) 
    223  
    224 parse_extra_pars() 
    225  
    226199 
    227200# =============== FUNCTION DEFINITIONS ================ 
    … …  
    324297    ocl_function=make_ocl("return sas_gamma(q);", "sas_gamma", ["lib/sas_gamma.c"]), 
    325298    limits=(-3.1, 10), 
    326 ) 
    327 add_function( 
    328     name="gammaln(x)", 
    329     mp_function=mp.loggamma, 
    330     np_function=scipy.special.gammaln, 
    331     ocl_function=make_ocl("return sas_gammaln(q);", "sas_gammaln", ["lib/sas_gammainc.c"]), 
    332     #ocl_function=make_ocl("return lgamma(q);", "sas_gammaln"), 
    333 ) 
    334 add_function( 
    335     name="gammainc(x)", 
    336     mp_function=lambda x, a=A: mp.gammainc(a, a=0, b=x)/mp.gamma(a), 
    337     np_function=lambda x, a=A: scipy.special.gammainc(a, x), 
    338     ocl_function=make_ocl("return sas_gammainc(%.15g,q);"%A, "sas_gammainc", ["lib/sas_gammainc.c"]), 
    339 ) 
    340 add_function( 
    341     name="gammaincc(x)", 
    342     mp_function=lambda x, a=A: mp.gammainc(a, a=x, b=mp.inf)/mp.gamma(a), 
    343     np_function=lambda x, a=A: scipy.special.gammaincc(a, x), 
    344     ocl_function=make_ocl("return sas_gammaincc(%.15g,q);"%A, "sas_gammaincc", ["lib/sas_gammainc.c"]), 
    345299) 
    346300add_function( 
    … …  
    509463lanczos_gamma = """\ 
    510464    const double coeff[] = { 
    511             76.18009172947146, -86.50532032941677, 
    512             24.01409824083091, -1.231739572450155, 
     465            76.18009172947146,     -86.50532032941677, 
     466            24.01409824083091,     -1.231739572450155, 
    513467            0.1208650973866179e-2,-0.5395239384953e-5 
    514468            }; 
    … …  
    521475""" 
    522476add_function( 
    523     name="loggamma(x)", 
     477    name="log gamma(x)", 
    524478    mp_function=mp.loggamma, 
    525479    np_function=scipy.special.gammaln, 
    … …  
    645599 
    646600ALL_FUNCTIONS = set(FUNCTIONS.keys()) 
    647 ALL_FUNCTIONS.discard("loggamma")  # use cephes-based gammaln instead 
     601ALL_FUNCTIONS.discard("loggamma")  # OCL version not ready yet 
    648602ALL_FUNCTIONS.discard("3j1/x:taylor") 
    649603ALL_FUNCTIONS.discard("3j1/x:trig") 
    … …  
    661615    -r indicates that the relative error should be plotted (default), 
    662616    -x<range> indicates the steps in x, where <range> is one of the following 
    663         log indicates log stepping in [10^-3, 10^5] (default) 
    664         logq indicates log stepping in [10^-4, 10^1] 
    665         linear indicates linear stepping in [1, 1000] 
    666         zoom indicates linear stepping in [1000, 1010] 
    667         neg indicates linear stepping in [-100.1, 100.1] 
    668         start:stop:n[:stepping] indicates an n-step plot in [start, stop] 
    669             or [10^start, 10^stop] if stepping is "log" (default n=400) 
    670 Some functions (notably gammainc/gammaincc) have an additional parameter A 
    671 which can be set from the command line as A=value.  Default is A=1. 
    672  
    673 Name is one of: 
     617      log indicates log stepping in [10^-3, 10^5] (default) 
     618      logq indicates log stepping in [10^-4, 10^1] 
     619      linear indicates linear stepping in [1, 1000] 
     620      zoom indicates linear stepping in [1000, 1010] 
     621      neg indicates linear stepping in [-100.1, 100.1] 
     622and name is "all" or one of: 
    674623    """+names) 
    675624    sys.exit(1) 

  • sasmodels/kernelpy.py

    r91bd550 r91bd550  
    3737        self.info = model_info 
    3838        self.dtype = np.dtype('d') 
    39         logger.info("make python model " + self.info.name) 
    4039 
    4140    def make_kernel(self, q_vectors): 

  • sasmodels/model_test.py

    r12eec1e r012cd34  
    4747import sys 
    4848import unittest 
    49 import traceback 
    5049 
    5150try: 
    … …  
    7574# pylint: enable=unused-import 
    7675 
     76 
    7777def make_suite(loaders, models): 
    7878    # type: (List[str], List[str]) -> unittest.TestSuite 
    … …  
    8686    *models* is the list of models to test, or *["all"]* to test all models. 
    8787    """ 
     88    ModelTestCase = _hide_model_case_from_nose() 
    8889    suite = unittest.TestSuite() 
    8990 
    … …  
    9495        skip = [] 
    9596    for model_name in models: 
    96         if model_name not in skip: 
    97             model_info = load_model_info(model_name) 
    98             _add_model_to_suite(loaders, suite, model_info) 
     97        if model_name in skip: 
     98            continue 
     99        model_info = load_model_info(model_name) 
     100 
     101        #print('------') 
     102        #print('found tests in', model_name) 
     103        #print('------') 
     104 
     105        # if ispy then use the dll loader to call pykernel 
     106        # don't try to call cl kernel since it will not be 
     107        # available in some environmentes. 
     108        is_py = callable(model_info.Iq) 
     109 
     110        # Some OpenCL drivers seem to be flaky, and are not producing the 
     111        # expected result.  Since we don't have known test values yet for 
     112        # all of our models, we are instead going to compare the results 
     113        # for the 'smoke test' (that is, evaluation at q=0.1 for the default 
     114        # parameters just to see that the model runs to completion) between 
     115        # the OpenCL and the DLL.  To do this, we define a 'stash' which is 
     116        # shared between OpenCL and DLL tests.  This is just a list.  If the 
     117        # list is empty (which it will be when DLL runs, if the DLL runs 
     118        # first), then the results are appended to the list.  If the list 
     119        # is not empty (which it will be when OpenCL runs second), the results 
     120        # are compared to the results stored in the first element of the list. 
     121        # This is a horrible stateful hack which only makes sense because the 
     122        # test suite is thrown away after being run once. 
     123        stash = [] 
     124 
     125        if is_py:  # kernel implemented in python 
     126            test_name = "%s-python"%model_name 
     127            test_method_name = "test_%s_python" % model_info.id 
     128            test = ModelTestCase(test_name, model_info, 
     129                                 test_method_name, 
     130                                 platform="dll",  # so that 
     131                                 dtype="double", 
     132                                 stash=stash) 
     133            suite.addTest(test) 
     134        else:   # kernel implemented in C 
     135 
     136            # test using dll if desired 
     137            if 'dll' in loaders or not use_opencl(): 
     138                test_name = "%s-dll"%model_name 
     139                test_method_name = "test_%s_dll" % model_info.id 
     140                test = ModelTestCase(test_name, model_info, 
     141                                     test_method_name, 
     142                                     platform="dll", 
     143                                     dtype="double", 
     144                                     stash=stash) 
     145                suite.addTest(test) 
     146 
     147            # test using opencl if desired and available 
     148            if 'opencl' in loaders and use_opencl(): 
     149                test_name = "%s-opencl"%model_name 
     150                test_method_name = "test_%s_opencl" % model_info.id 
     151                # Using dtype=None so that the models that are only 
     152                # correct for double precision are not tested using 
     153                # single precision.  The choice is determined by the 
     154                # presence of *single=False* in the model file. 
     155                test = ModelTestCase(test_name, model_info, 
     156                                     test_method_name, 
     157                                     platform="ocl", dtype=None, 
     158                                     stash=stash) 
     159                #print("defining", test_name) 
     160                suite.addTest(test) 
    99161 
    100162    return suite 
    101  
    102 def _add_model_to_suite(loaders, suite, model_info): 
    103     ModelTestCase = _hide_model_case_from_nose() 
    104  
    105     #print('------') 
    106     #print('found tests in', model_name) 
    107     #print('------') 
    108  
    109     # if ispy then use the dll loader to call pykernel 
    110     # don't try to call cl kernel since it will not be 
    111     # available in some environmentes. 
    112     is_py = callable(model_info.Iq) 
    113  
    114     # Some OpenCL drivers seem to be flaky, and are not producing the 
    115     # expected result.  Since we don't have known test values yet for 
    116     # all of our models, we are instead going to compare the results 
    117     # for the 'smoke test' (that is, evaluation at q=0.1 for the default 
    118     # parameters just to see that the model runs to completion) between 
    119     # the OpenCL and the DLL.  To do this, we define a 'stash' which is 
    120     # shared between OpenCL and DLL tests.  This is just a list.  If the 
    121     # list is empty (which it will be when DLL runs, if the DLL runs 
    122     # first), then the results are appended to the list.  If the list 
    123     # is not empty (which it will be when OpenCL runs second), the results 
    124     # are compared to the results stored in the first element of the list. 
    125     # This is a horrible stateful hack which only makes sense because the 
    126     # test suite is thrown away after being run once. 
    127     stash = [] 
    128  
    129     if is_py:  # kernel implemented in python 
    130         test_name = "%s-python"%model_info.name 
    131         test_method_name = "test_%s_python" % model_info.id 
    132         test = ModelTestCase(test_name, model_info, 
    133                                 test_method_name, 
    134                                 platform="dll",  # so that 
    135                                 dtype="double", 
    136                                 stash=stash) 
    137         suite.addTest(test) 
    138     else:   # kernel implemented in C 
    139  
    140         # test using dll if desired 
    141         if 'dll' in loaders or not use_opencl(): 
    142             test_name = "%s-dll"%model_info.name 
    143             test_method_name = "test_%s_dll" % model_info.id 
    144             test = ModelTestCase(test_name, model_info, 
    145                                     test_method_name, 
    146                                     platform="dll", 
    147                                     dtype="double", 
    148                                     stash=stash) 
    149             suite.addTest(test) 
    150  
    151         # test using opencl if desired and available 
    152         if 'opencl' in loaders and use_opencl(): 
    153             test_name = "%s-opencl"%model_info.name 
    154             test_method_name = "test_%s_opencl" % model_info.id 
    155             # Using dtype=None so that the models that are only 
    156             # correct for double precision are not tested using 
    157             # single precision.  The choice is determined by the 
    158             # presence of *single=False* in the model file. 
    159             test = ModelTestCase(test_name, model_info, 
    160                                     test_method_name, 
    161                                     platform="ocl", dtype=None, 
    162                                     stash=stash) 
    163             #print("defining", test_name) 
    164             suite.addTest(test) 
    165  
    166163 
    167164def _hide_model_case_from_nose(): 
    … …  
    351348    return abs(target-actual)/shift < 1.5*10**-digits 
    352349 
    353 # CRUFT: old interface; should be deprecated and removed 
    354 def run_one(model_name): 
    355     # msg = "use check_model(model_info) rather than run_one(model_name)" 
    356     # warnings.warn(msg, category=DeprecationWarning, stacklevel=2) 
    357     try: 
    358         model_info = load_model_info(model_name) 
    359     except Exception: 
    360         output = traceback.format_exc() 
    361         return output 
    362  
    363     success, output = check_model(model_info) 
    364     return output 
    365  
    366 def check_model(model_info): 
    367     # type: (ModelInfo) -> str 
    368     """ 
    369     Run the tests for a single model, capturing the output. 
    370  
    371     Returns success status and the output string. 
     350def run_one(model): 
     351    # type: (str) -> str 
     352    """ 
     353    Run the tests for a single model, printing the results to stdout. 
     354 
     355    *model* can by a python file, which is handy for checking user defined 
     356    plugin models. 
    372357    """ 
    373358    # Note that running main() directly did not work from within the 
    … …  
    384369    # Build a test suite containing just the model 
    385370    loaders = ['opencl'] if use_opencl() else ['dll'] 
    386     suite = unittest.TestSuite() 
    387     _add_model_to_suite(loaders, suite, model_info) 
     371    models = [model] 
     372    try: 
     373        suite = make_suite(loaders, models) 
     374    except Exception: 
     375        import traceback 
     376        stream.writeln(traceback.format_exc()) 
     377        return 
    388378 
    389379    # Warn if there are no user defined tests. 
    … …  
    400390    for test in suite: 
    401391        if not test.info.tests: 
    402             stream.writeln("Note: %s has no user defined tests."%model_info.name) 
     392            stream.writeln("Note: %s has no user defined tests."%model) 
    403393        break 
    404394    else: 
    … …  
    416406    output = stream.getvalue() 
    417407    stream.close() 
    418     return result.wasSuccessful(), output 
     408    return output 
    419409 
    420410 

  • sasmodels/sasview_model.py

    rbd547d0 rbd547d0  
    803803            return value, [value], [1.0] 
    804804 
    805     @classmethod 
    806     def runTests(cls): 
    807         """ 
    808         Run any tests built into the model and captures the test output. 
    809  
    810         Returns success flag and output 
    811         """ 
    812         from .model_test import check_model 
    813         return check_model(cls._model_info) 
    814  
    815805def test_cylinder(): 
    816806    # type: () -> float 

  • sasmodels/special.py

    rfba9ca0 rdf69efa  
    113113        The standard math function, tgamma(x) is unstable for $x < 1$ 
    114114        on some platforms. 
    115  
    116     sas_gammaln(x): 
    117         log gamma function sas_gammaln\ $(x) = \log \Gamma(|x|)$. 
    118  
    119         The standard math function, lgamma(x), is incorrect for single 
    120         precision on some platforms. 
    121  
    122     sas_gammainc(a, x), sas_gammaincc(a, x): 
    123         Incomplete gamma function 
    124         sas_gammainc\ $(a, x) = \int_0^x t^{a-1}e^{-t}\,dt / \Gamma(a)$ 
    125         and complementary incomplete gamma function 
    126         sas_gammaincc\ $(a, x) = \int_x^\infty t^{a-1}e^{-t}\,dt / \Gamma(a)$ 
    127115 
    128116    sas_erf(x), sas_erfc(x): 
    … …  
    219207from numpy import pi, nan, inf 
    220208from scipy.special import gamma as sas_gamma 
    221 from scipy.special import gammaln as sas_gammaln 
    222 from scipy.special import gammainc as sas_gammainc 
    223 from scipy.special import gammaincc as sas_gammaincc 
    224209from scipy.special import erf as sas_erf 
    225210from scipy.special import erfc as sas_erfc 
Note: See TracChangeset for help on using the changeset viewer.