|
| 1 | +""" |
| 2 | +Driver for running model on design of experiments cases using OpenTURNS sampling methods |
| 3 | +""" |
| 4 | +from __future__ import print_function |
| 5 | +import numpy as np |
| 6 | +from six import iteritems |
| 7 | + |
| 8 | +from openmdao.api import DOEDriver, OptionsDictionary |
| 9 | +from openmdao.drivers.doe_generators import DOEGenerator |
| 10 | + |
| 11 | +OPENTURNS_NOT_INSTALLED = False |
| 12 | +try: |
| 13 | + import openturns as ot |
| 14 | +except ImportError: |
| 15 | + OPENTURNS_NOT_INSTALLED = True |
| 16 | + |
| 17 | + |
| 18 | +class OpenturnsMonteCarloDOEGenerator(DOEGenerator): |
| 19 | + LIMIT = 1e12 |
| 20 | + |
| 21 | + def __init__(self, n_samples=10, dist=None): |
| 22 | + super(OpenturnsMonteCarloDOEGenerator, self).__init__() |
| 23 | + |
| 24 | + self.n_samples = n_samples |
| 25 | + self.distribution = dist |
| 26 | + self.called = False |
| 27 | + |
| 28 | + def __call__(self, uncertain_vars, model=None): |
| 29 | + if self.distribution is None: |
| 30 | + dists = [] |
| 31 | + for name, meta in iteritems(uncertain_vars): |
| 32 | + size = meta["size"] |
| 33 | + meta_low = meta["lower"] |
| 34 | + meta_high = meta["upper"] |
| 35 | + for j in range(size): |
| 36 | + if isinstance(meta_low, np.ndarray): |
| 37 | + p_low = meta_low[j] |
| 38 | + else: |
| 39 | + p_low = meta_low |
| 40 | + p_low = max(p_low, -self.LIMIT) |
| 41 | + |
| 42 | + if isinstance(meta_high, np.ndarray): |
| 43 | + p_high = meta_high[j] |
| 44 | + else: |
| 45 | + p_high = meta_high |
| 46 | + p_high = min(p_high, self.LIMIT) |
| 47 | + |
| 48 | + dists.append(ot.Uniform(p_low, p_high)) |
| 49 | + self.distribution = ot.ComposedDistribution(dists) |
| 50 | + else: |
| 51 | + size = 0 |
| 52 | + for name, meta in iteritems(uncertain_vars): |
| 53 | + size += meta["size"] |
| 54 | + if (size) != (self.distribution.getDimension()): |
| 55 | + raise RuntimeError( |
| 56 | + "Bad distribution dimension: should be equal to uncertain variables size {} " |
| 57 | + ", got {}".format(size, self.distribution.getDimension()) |
| 58 | + ) |
| 59 | + samples = self.distribution.getSample(self.n_samples) |
| 60 | + self._cases = np.array(samples) |
| 61 | + self.called = True |
| 62 | + sample = [] |
| 63 | + for i in range(self._cases.shape[0]): |
| 64 | + j = 0 |
| 65 | + for name, meta in iteritems(uncertain_vars): |
| 66 | + size = meta["size"] |
| 67 | + sample.append((name, self._cases[i, j : j + size])) |
| 68 | + j += size |
| 69 | + yield sample |
| 70 | + |
| 71 | + def get_cases(self): |
| 72 | + if not self.called: |
| 73 | + raise RuntimeError("Have to run the driver before getting cases") |
| 74 | + return self._cases |
| 75 | + |
| 76 | + |
| 77 | +class OpenturnsDOEDriver(DOEDriver): |
| 78 | + """ |
| 79 | + Baseclass for OpenTURNS design-of-experiments Drivers |
| 80 | + """ |
| 81 | + |
| 82 | + def __init__(self, **kwargs): |
| 83 | + super(OpenturnsDOEDriver, self).__init__() |
| 84 | + |
| 85 | + self.options.declare( |
| 86 | + "distribution", |
| 87 | + types=ot.ComposedDistribution, |
| 88 | + default=None, |
| 89 | + allow_none=True, |
| 90 | + desc="Joint distribution of uncertain variables", |
| 91 | + ) |
| 92 | + self.options.declare( |
| 93 | + "n_samples", types=int, default=2, desc="number of sample to generate" |
| 94 | + ) |
| 95 | + self.options.update(kwargs) |
| 96 | + |
| 97 | + self.options["generator"] = OpenturnsMonteCarloDOEGenerator( |
| 98 | + self.options["n_samples"], self.options["distribution"] |
| 99 | + ) |
| 100 | + |
| 101 | + def _set_name(self): |
| 102 | + self._name = "OpenTURNS_DOE_MonteCarlo" |
| 103 | + |
| 104 | + def get_cases(self): |
| 105 | + return self.options["generator"].get_cases() |
0 commit comments