| 1 | #!/usr/bin/env python |
|---|
| 2 | """ |
|---|
| 3 | Once submit has finished with the jobs, this function is called to have PUQ |
|---|
| 4 | process the results. |
|---|
| 5 | """ |
|---|
| 6 | from __future__ import print_function |
|---|
| 7 | import sys |
|---|
| 8 | import os |
|---|
| 9 | import numpy as np |
|---|
| 10 | import h5py |
|---|
| 11 | import re |
|---|
| 12 | import puq |
|---|
| 13 | from puq.jpickle import unpickle |
|---|
| 14 | import Rappture |
|---|
| 15 | import StringIO |
|---|
| 16 | from scipy.spatial import ConvexHull |
|---|
| 17 | # geometry library |
|---|
| 18 | from shapely.geometry import Polygon |
|---|
| 19 | from shapely.ops import unary_union |
|---|
| 20 | |
|---|
| 21 | # Redirect stdout and stderr to files for debugging. |
|---|
| 22 | # Append to the files created in get_params.py |
|---|
| 23 | sys.stdout = open("uq_debug.out", 'a') |
|---|
| 24 | sys.stderr = open("uq_debug.err", 'a') |
|---|
| 25 | |
|---|
| 26 | |
|---|
| 27 | # Restore the state of a PUQ session from a HDF5 file. |
|---|
| 28 | def load_from_hdf5(name): |
|---|
| 29 | h5 = h5py.File(name, 'r+') |
|---|
| 30 | sw = unpickle(h5['private/sweep'].value) |
|---|
| 31 | sw.fname = os.path.splitext(name)[0] |
|---|
| 32 | h5.close() |
|---|
| 33 | |
|---|
| 34 | sw.psweep._sweep = sw |
|---|
| 35 | |
|---|
| 36 | if hasattr(sw.psweep, 'reinit'): |
|---|
| 37 | sw.psweep.reinit() |
|---|
| 38 | return sw |
|---|
| 39 | |
|---|
| 40 | |
|---|
| 41 | # Plots probability curves |
|---|
| 42 | def plot_pdf_curve(io, h5, xvals, vname, percent): |
|---|
| 43 | print('plot_pdf_curve %s %s' % (vname, percent)) |
|---|
| 44 | # compute upper and lower percentiles |
|---|
| 45 | pm = (100 - percent)/200.0 |
|---|
| 46 | pp = 1 - pm |
|---|
| 47 | |
|---|
| 48 | label = None |
|---|
| 49 | |
|---|
| 50 | # collect data into an array |
|---|
| 51 | xarr = np.empty(len(xvals[vname])) |
|---|
| 52 | yp = np.empty(len(xvals[vname])) |
|---|
| 53 | ym = np.empty(len(xvals[vname])) |
|---|
| 54 | for vindex in sorted(xvals[vname].keys()): |
|---|
| 55 | if label is None: |
|---|
| 56 | label = h5['/output/data/%s[%d]' % (vname, vindex)].attrs['label'] |
|---|
| 57 | xarr[vindex] = xvals[vname][vindex] |
|---|
| 58 | yp[vindex] = pcurves[vname][vindex].ppf(pp) |
|---|
| 59 | ym[vindex] = pcurves[vname][vindex].ppf(pm) |
|---|
| 60 | |
|---|
| 61 | curve = io['output.curve(curve_pdf-%s-%s)' % (vname, percent)] |
|---|
| 62 | if percent == 0: |
|---|
| 63 | curve['about.label'] = "mean" |
|---|
| 64 | else: |
|---|
| 65 | curve['about.label'] = "middle %s%%" % percent |
|---|
| 66 | curve['about.group'] = label |
|---|
| 67 | curve['about.uqtype'] = 'Probability' |
|---|
| 68 | |
|---|
| 69 | pts = "" |
|---|
| 70 | for x, y in zip(xarr, yp): |
|---|
| 71 | pts += "%s %s " % (x, y) |
|---|
| 72 | if percent == 0: |
|---|
| 73 | pts += '\n' |
|---|
| 74 | else: |
|---|
| 75 | for x, y in reversed(zip(xarr, ym)): |
|---|
| 76 | pts += "%s %s " % (x, y) |
|---|
| 77 | pts += "%s %s\n" % (xarr[0], yp[0]) |
|---|
| 78 | |
|---|
| 79 | curve['component.xy'] = pts |
|---|
| 80 | |
|---|
| 81 | |
|---|
| 82 | def add_pts(f1, percent): |
|---|
| 83 | # compute upper and lower percentiles |
|---|
| 84 | pm = (100 - percent) / 200.0 |
|---|
| 85 | pp = 1 - pm |
|---|
| 86 | prob = np.linspace(pm, pp, 31) |
|---|
| 87 | x, y = f1.eval(prob) |
|---|
| 88 | return np.array(zip(x, y)) |
|---|
| 89 | |
|---|
| 90 | |
|---|
| 91 | def plot_pdf_acurve(io, h5, acurves, vname, percent): |
|---|
| 92 | """ |
|---|
| 93 | This function plots the probability curves for parametric |
|---|
| 94 | PDFs. |
|---|
| 95 | """ |
|---|
| 96 | print('plot_pdf_acurve %s %s' % (vname, percent)) |
|---|
| 97 | |
|---|
| 98 | label = None |
|---|
| 99 | prev_pts = None # last set of points |
|---|
| 100 | |
|---|
| 101 | poly = [] |
|---|
| 102 | for vindex in sorted(acurves[vname].keys()): |
|---|
| 103 | if label is None: |
|---|
| 104 | label = h5['/output/data/%s[%d]' % (vname, vindex)].attrs['label'] |
|---|
| 105 | f1 = unpickle(h5['/output/data/%s[%d]' % (vname, vindex)].attrs['curve']) |
|---|
| 106 | bpts = add_pts(f1, percent) |
|---|
| 107 | |
|---|
| 108 | # first data set? Just remember it. |
|---|
| 109 | if prev_pts is None: |
|---|
| 110 | prev_pts = bpts |
|---|
| 111 | continue |
|---|
| 112 | |
|---|
| 113 | pts = np.array((prev_pts, bpts)).ravel().reshape(-1, 2) |
|---|
| 114 | hull = ConvexHull(pts, qhull_options='Pp') |
|---|
| 115 | p1 = Polygon([hull.points[v] for v in hull.vertices]) |
|---|
| 116 | poly.append(p1) |
|---|
| 117 | prev_pts = bpts |
|---|
| 118 | |
|---|
| 119 | u = unary_union(poly) |
|---|
| 120 | |
|---|
| 121 | curve = io['output.curve(curve_pdf-%s-%s)' % (vname, percent)] |
|---|
| 122 | if percent == 0: |
|---|
| 123 | curve['about.label'] = "mean" |
|---|
| 124 | else: |
|---|
| 125 | curve['about.label'] = "middle %s%%" % percent |
|---|
| 126 | curve['about.group'] = label |
|---|
| 127 | curve['about.uqtype'] = 'Probability' |
|---|
| 128 | curve['component.xy'] = np.array(u.exterior.xy) |
|---|
| 129 | |
|---|
| 130 | |
|---|
| 131 | def plot_pdf(io, v, pdf, desc): |
|---|
| 132 | p = io['output.curve(%s)' % desc] |
|---|
| 133 | p['about.label'] = desc |
|---|
| 134 | p['about.uqtype'] = "PDF" |
|---|
| 135 | p['yaxis.label'] = 'Probability' |
|---|
| 136 | |
|---|
| 137 | pts = "%s 0\n" % pdf.x[0] |
|---|
| 138 | for x, y in zip(pdf.x, pdf.y): |
|---|
| 139 | pts += "%s %s\n" % (x, y) |
|---|
| 140 | pts += "%s 0\n" % pdf.x[-1] |
|---|
| 141 | p['component.xy'] = pts |
|---|
| 142 | |
|---|
| 143 | |
|---|
| 144 | def write_responses(io, h5): |
|---|
| 145 | uqtype = h5.attrs['UQtype'] |
|---|
| 146 | for v in h5[uqtype]: |
|---|
| 147 | if '[' in v: |
|---|
| 148 | # It is a curve. Ignore. |
|---|
| 149 | continue |
|---|
| 150 | try: |
|---|
| 151 | desc = h5['%s/%s' % (uqtype, v)].attrs['description'] |
|---|
| 152 | except: |
|---|
| 153 | desc = '' |
|---|
| 154 | try: |
|---|
| 155 | label = h5['%s/%s' % (uqtype, v)].attrs['label'] |
|---|
| 156 | except: |
|---|
| 157 | label = '' |
|---|
| 158 | |
|---|
| 159 | rsp = h5['/%s/%s/response' % (uqtype, v)].value |
|---|
| 160 | |
|---|
| 161 | rout = io['output.response(%s)' % label] |
|---|
| 162 | rout['value'] = rsp |
|---|
| 163 | rout['about.description'] = desc |
|---|
| 164 | rout['about.label'] = label |
|---|
| 165 | rout['about.uqtype'] = 'Response' |
|---|
| 166 | |
|---|
| 167 | rs = unpickle(rsp) |
|---|
| 168 | rout['variables'] = ' '.join([str(p.name) for p in rs.params]) |
|---|
| 169 | labels = ' '.join([repr(str(p.label)) for p in rs.params]) |
|---|
| 170 | rout['labels'] = labels.replace("'", '"') |
|---|
| 171 | |
|---|
| 172 | if type(rs) == puq.response.ResponseFunc: |
|---|
| 173 | rout['equation'] = rs.eqn |
|---|
| 174 | rout['rmse'] = "{:6.3g}".format(rs.rmse()[1]) |
|---|
| 175 | |
|---|
| 176 | rout['data'] = rs.data |
|---|
| 177 | |
|---|
| 178 | |
|---|
| 179 | def write_params(h5, out): |
|---|
| 180 | params = map(str, h5['/input/params'].keys()) |
|---|
| 181 | print('#' * 80, file=out) |
|---|
| 182 | print('INPUT PARAMETERS', file=out) |
|---|
| 183 | |
|---|
| 184 | for pname in params: |
|---|
| 185 | print('-' * 80, file=out) |
|---|
| 186 | p = puq.unpickle(h5['/input/params/' + pname].value) |
|---|
| 187 | cname = p.__class__.__name__[:-9] |
|---|
| 188 | pdf_str = '%s [%s - %s] mean=%s dev=%s mode=%s' % (cname, p.pdf.range[0], p.pdf.range[1], p.pdf.mean, p.pdf.dev, p.pdf.mode) |
|---|
| 189 | |
|---|
| 190 | print("Name:", p.name, file=out) |
|---|
| 191 | try: |
|---|
| 192 | print("Label:", p.label, file=out) |
|---|
| 193 | except: |
|---|
| 194 | pass |
|---|
| 195 | print("Desc:", p.description, file=out) |
|---|
| 196 | print('Value:', pdf_str, file=out) |
|---|
| 197 | print('#' * 80, file=out) |
|---|
| 198 | print(file=out) |
|---|
| 199 | |
|---|
| 200 | |
|---|
| 201 | def write_summary(io, h5): |
|---|
| 202 | outstr = StringIO.StringIO() |
|---|
| 203 | write_params(h5, outstr) |
|---|
| 204 | uqtype = h5.attrs['UQtype'] |
|---|
| 205 | for v in h5[uqtype]: |
|---|
| 206 | if '[' in v: |
|---|
| 207 | # It is a curve. Ignore. |
|---|
| 208 | continue |
|---|
| 209 | desc = h5['%s/%s' % (uqtype, v)].attrs['description'] |
|---|
| 210 | print("QoI: %s (%s)" % (v, desc), file=outstr) |
|---|
| 211 | rs = unpickle(h5['/%s/%s/response' % (uqtype, v)].value) |
|---|
| 212 | if type(rs) == puq.response.ResponseFunc: |
|---|
| 213 | print("\nv=%s\n" % rs.eqn, file=outstr) |
|---|
| 214 | print("SURROGATE MODEL ERROR:{:6.3g}%".format(rs.rmse()[1]), file=outstr) |
|---|
| 215 | sens = puq.unpickle(h5['/%s/%s/sensitivity' % (uqtype, v)].value) |
|---|
| 216 | max_name_len = max(map(len, [p[0] for p in sens])) |
|---|
| 217 | print("\nSENSITIVITY:", file=outstr) |
|---|
| 218 | print("Var%s u* dev" % (' '*(max_name_len)), file=outstr) |
|---|
| 219 | print('-'*(28+max_name_len), file=outstr) |
|---|
| 220 | for item in sens: |
|---|
| 221 | pad = ' '*(max_name_len - len(item[0])) |
|---|
| 222 | print("{}{} {:10.4g} {:10.4g}".format(pad, item[0], |
|---|
| 223 | item[1]['ustar'], item[1]['std']), file=outstr) |
|---|
| 224 | print('-'*(28+max_name_len), file=outstr) |
|---|
| 225 | print(file=outstr) |
|---|
| 226 | iostr = io['output.string(UQ Summary)'] |
|---|
| 227 | iostr['about.label'] = 'UQ Summary' |
|---|
| 228 | iostr['current'] = outstr.getvalue() |
|---|
| 229 | outstr.close() |
|---|
| 230 | |
|---|
| 231 | |
|---|
| 232 | def write_sensitivity(io, h5): |
|---|
| 233 | # If more than one variable, display sensitivity. |
|---|
| 234 | # Curves have indexed variables, so skip them. |
|---|
| 235 | if len(h5['/input/params']) > 1 and ['[' in x for x in h5[uqtype]].count(False): |
|---|
| 236 | for v in h5[uqtype]: |
|---|
| 237 | if '[' in v: |
|---|
| 238 | # curve. skip it. |
|---|
| 239 | continue |
|---|
| 240 | desc = h5['/output/data/%s' % v].attrs['label'] |
|---|
| 241 | sens = unpickle(h5['/%s/%s/sensitivity' % (uqtype, v)].value) |
|---|
| 242 | |
|---|
| 243 | hist = io['output.histogram(sens-%s)' % v] |
|---|
| 244 | hist['about.label'] = desc |
|---|
| 245 | hist['about.uqtype'] = 'Sensitivity' |
|---|
| 246 | hist['about.type'] = 'scatter' |
|---|
| 247 | hist['xaxis.label'] = 'Parameters' |
|---|
| 248 | hist['yaxis.label'] = 'Sensitivity' |
|---|
| 249 | pts = '' |
|---|
| 250 | for name in sens: |
|---|
| 251 | n = name[0] |
|---|
| 252 | try: |
|---|
| 253 | n = h5['/input/params/%s' % n].attrs['label'] |
|---|
| 254 | except: |
|---|
| 255 | pass |
|---|
| 256 | pts += "\"%s\" %s\n" % (n, name[1]['ustar']) |
|---|
| 257 | hist['component.xy'] = pts |
|---|
| 258 | |
|---|
| 259 | sw = load_from_hdf5(sys.argv[1]) |
|---|
| 260 | sw.analyze() |
|---|
| 261 | |
|---|
| 262 | h5 = h5py.File(sys.argv[1], 'r+') |
|---|
| 263 | io = Rappture.PyXml('run_uq.xml') |
|---|
| 264 | |
|---|
| 265 | # curves built from pdfs |
|---|
| 266 | pcurves = {} |
|---|
| 267 | xvals = {} |
|---|
| 268 | acurves = {} |
|---|
| 269 | |
|---|
| 270 | reg1 = re.compile('([ \da-zA-Z_]+)\[([ \d]+)\]') |
|---|
| 271 | |
|---|
| 272 | uqtype = h5.attrs['UQtype'] |
|---|
| 273 | for v in h5[uqtype]: |
|---|
| 274 | rsp = h5['/%s/%s/response' % (uqtype, v)].value |
|---|
| 275 | rs = unpickle(rsp) |
|---|
| 276 | pdf = rs.pdf(fit=False) |
|---|
| 277 | odata = h5['/output/data/%s' % v] |
|---|
| 278 | |
|---|
| 279 | # For curves built from pdfs, just put them in a dict for now |
|---|
| 280 | if 'x' in odata.attrs: |
|---|
| 281 | matches = reg1.findall(v) |
|---|
| 282 | vname, vindex = matches[0] |
|---|
| 283 | vindex = int(vindex) |
|---|
| 284 | if vname not in pcurves: |
|---|
| 285 | pcurves[vname] = {} |
|---|
| 286 | xvals[vname] = {} |
|---|
| 287 | xvals[vname][vindex] = odata.attrs['x'] |
|---|
| 288 | pcurves[vname][vindex] = pdf |
|---|
| 289 | elif 'curve' in odata.attrs: |
|---|
| 290 | matches = reg1.findall(v) |
|---|
| 291 | vname, vindex = matches[0] |
|---|
| 292 | print('ACURVE: %s - %s' % (vname, vindex)) |
|---|
| 293 | if vname not in acurves: |
|---|
| 294 | acurves[vname] = {} |
|---|
| 295 | acurves[vname][int(vindex)] = pdf |
|---|
| 296 | else: |
|---|
| 297 | desc = h5['/output/data/%s' % v].attrs['label'] |
|---|
| 298 | plot_pdf(io, v, pdf, desc) |
|---|
| 299 | |
|---|
| 300 | # now do probability curves |
|---|
| 301 | for vname in xvals: |
|---|
| 302 | plot_pdf_curve(io, h5, xvals, vname, 95) |
|---|
| 303 | plot_pdf_curve(io, h5, xvals, vname, 50) |
|---|
| 304 | |
|---|
| 305 | for vname in acurves: |
|---|
| 306 | try: |
|---|
| 307 | plot_pdf_acurve(io, h5, acurves, vname, 95) |
|---|
| 308 | except: |
|---|
| 309 | pass |
|---|
| 310 | try: |
|---|
| 311 | plot_pdf_acurve(io, h5, acurves, vname, 50) |
|---|
| 312 | except: |
|---|
| 313 | pass |
|---|
| 314 | |
|---|
| 315 | write_sensitivity(io, h5) |
|---|
| 316 | write_responses(io, h5) |
|---|
| 317 | write_summary(io, h5) |
|---|
| 318 | io.close() |
|---|
| 319 | |
|---|
| 320 | h5.close() |
|---|
