/
test_hessian.py
403 lines (317 loc) · 18.3 KB
/
test_hessian.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
import pickle
import unittest
import numpy as np
from pygsti.forwardsims.mapforwardsim import MapForwardSimulator
import pygsti
from pygsti import protocols as proto
from pygsti.modelpacks import smq1Q_XY
from pygsti.modelpacks import smq1Q_XYI
from pygsti.baseobjs import Label as L
from pygsti.report import modelfunction as gsf
from ..testutils import BaseTestCase, compare_files
class TestHessianMethods(BaseTestCase):
def setUp(self):
super(TestHessianMethods, self).setUp()
self.model = smq1Q_XY.target_model()
self.model = self.model.depolarize(spam_noise = .01, op_noise = .001)
self.model = self.model.rotate(max_rotate=.005, seed=1234)
prep_fiducials = smq1Q_XY.prep_fiducials()
meas_fiducials = smq1Q_XY.meas_fiducials()
germs = smq1Q_XY.germs()
op_labels = list(self.model.operations.keys()) # also == std.gates
self.maxLengthList = [1]
#circuits for XY model.
self.gss = pygsti.circuits.make_lsgst_structs(op_labels, prep_fiducials[0:4],
meas_fiducials[0:3], smq1Q_XY.germs(), self.maxLengthList)
self.edesign = proto.CircuitListsDesign([pygsti.circuits.CircuitList(circuit_struct) for circuit_struct in self.gss])
self.ds = pygsti.data.simulate_data(self.model, self.edesign.all_circuits_needing_data, 1000, seed = 1234)
def test_parameter_counting(self):
#XY Model: SPAM=True
n = smq1Q_XY.target_model().num_params
self.assertEqual(n,44) # 2*16 + 3*4 = 44
n = smq1Q_XY.target_model().num_nongauge_params
self.assertEqual(n,28) # full 16 gauge params
#XY Model: SPAM=False
tst = smq1Q_XY.target_model()
del tst.preps['rho0']
del tst.povms['Mdefault']
n = tst.num_params
self.assertEqual(n,32) # 2*16 = 32
n = tst.num_nongauge_params
self.assertEqual(n,18) # gates are all unital & TP => only 14 gauge params (2 casimirs)
#XYI Model: SPAM=True
n = smq1Q_XYI.target_model().num_params
self.assertEqual(n,60) # 3*16 + 3*4 = 60
n = smq1Q_XYI.target_model().num_nongauge_params
self.assertEqual(n,44) # full 16 gauge params: SPAM gate + 3 others
#XYI Model: SPAM=False
tst = smq1Q_XYI.target_model()
del tst.preps['rho0']
del tst.povms['Mdefault']
n = tst.num_params
self.assertEqual(n,48) # 3*16 = 48
n = tst.num_nongauge_params
self.assertEqual(n,34) # gates are all unital & TP => only 14 gauge params (2 casimirs)
#XYI Model: SP0=False
tst = smq1Q_XYI.target_model()
tst.preps['rho0'] = pygsti.modelmembers.states.TPState(tst.preps['rho0'])
n = tst.num_params
self.assertEqual(n,59) # 3*16 + 2*4 + 3 = 59
n = tst.num_nongauge_params
self.assertEqual(n,44) # 15 gauge params (minus one b/c can't change rho?)
#XYI Model: G0=SP0=False
tst.operations[L(())] = pygsti.modelmembers.operations.FullTPOp(tst.operations[L(())])
tst.operations['Gxpi2',0] = pygsti.modelmembers.operations.FullTPOp(tst.operations['Gxpi2',0])
tst.operations['Gypi2',0] = pygsti.modelmembers.operations.FullTPOp(tst.operations['Gypi2',0])
n = tst.num_params
self.assertEqual(n,47) # 3*12 + 2*4 + 3 = 47
n = tst.num_nongauge_params
self.assertEqual(n,35) # full 12 gauge params of single 4x3 gate
def test_hessian_projection(self):
chi2Hessian = pygsti.chi2_hessian(self.model, self.ds)
proj_non_gauge = self.model.compute_nongauge_projector()
projectedHessian = proj_non_gauge@chi2Hessian@proj_non_gauge
self.assertEqual( projectedHessian.shape, (44,44) )
self.assertEqual( np.linalg.matrix_rank(proj_non_gauge), 28)
self.assertEqual( np.linalg.matrix_rank(projectedHessian), 28)
eigvals = np.sort(abs(np.linalg.eigvals(projectedHessian)))
print("eigvals = ",eigvals)
eigvals_chk = np.array([ 5.45537035e-13, 5.45537035e-13, 1.47513013e-12, 1.47513013e-12,
1.57813273e-12, 4.87695508e-12, 1.22061302e-11, 3.75982961e-11,
5.49796401e-11, 5.62019047e-11, 5.62019047e-11, 7.06418308e-11,
1.44881858e-10, 1.48934891e-10, 1.48934891e-10, 2.06194475e-10,
1.91727543e+01, 2.26401298e+02, 5.23331036e+02, 1.16447879e+03,
1.45737904e+03, 1.93375238e+03, 2.02017169e+03, 3.55570313e+03,
3.95986905e+03, 5.52173250e+03, 8.20436174e+03, 9.93573257e+03,
1.36092721e+04, 1.87334336e+04, 2.07723720e+04, 2.17070806e+04,
2.72168569e+04, 3.31886655e+04, 3.72430633e+04, 4.64233389e+04,
6.35672652e+04, 8.61196820e+04, 1.08248150e+05, 1.65647618e+05,
5.72597674e+05, 9.44823397e+05, 1.45785061e+06, 6.85705713e+06])
TOL = 1e-7
for val,chk in zip(eigvals,eigvals_chk):
if abs(val) > TOL or abs(chk) > TOL:
self.assertAlmostEqual(abs(val-chk)/(abs(chk)+TOL), 0.0, places=2)
# (else both chk and val are <= TOL, so both == 0 for our purposes)
#print "eigvals = ",eigvals
def test_confidenceRegion(self):
data = proto.ProtocolData(self.edesign, self.ds)
res = proto.ModelEstimateResults(data, proto.StandardGST(modes="full TP"))
#Add estimate for hessian-based CI --------------------------------------------------
builder = pygsti.objectivefns.PoissonPicDeltaLogLFunction.builder()
res.add_estimate(
proto.estimate.Estimate.create_gst_estimate(
res, smq1Q_XY.target_model(), smq1Q_XY.target_model(),
[self.model] * len(self.maxLengthList), parameters={'final_objfn_builder': builder}),
estimate_key="default"
)
est = res.estimates['default']
est.add_confidence_region_factory('final iteration estimate', 'final')
self.assertWarns(est.add_confidence_region_factory, 'final iteration estimate','final') #overwrites former
self.assertTrue( est.has_confidence_region_factory('final iteration estimate', 'final'))
cfctry = est.create_confidence_region_factory('final iteration estimate', 'final')
self.assertFalse( cfctry.can_construct_views() ) # b/c no hessian or LR enabled yet...
cfctry.compute_hessian(approximate=True)
cfctry.compute_hessian()
self.assertTrue( cfctry.has_hessian )
self.assertFalse( cfctry.can_construct_views() ) # b/c hessian isn't projected yet...
mdl_dummy = cfctry.model # test method
s = pickle.dumps(cfctry) # test pickle
pickle.loads(s)
cfctry.project_hessian('std')
cfctry.project_hessian('none')
cfctry.project_hessian('optimal gate CIs')
cfctry.project_hessian('intrinsic error')
with self.assertRaises(ValueError):
cfctry.project_hessian(95.0, 'normal', 'FooBar') #bad hessianProjection
self.assertTrue( cfctry.can_construct_views() )
ci_std = cfctry.view( 95.0, 'normal', 'std')
ci_noproj = cfctry.view( 95.0, 'normal', 'none')
ci_intrinsic = cfctry.view( 95.0, 'normal', 'intrinsic error')
ci_opt = cfctry.view( 95.0, 'normal', 'optimal gate CIs')
with self.assertRaises(ValueError):
cfctry.view(95.0, 'foobar') #bad region type
self.assertWarns(cfctry.view, 0.95, 'normal', 'none') # percentage < 1.0
#Add estimate for linresponse-based CI --------------------------------------------------
res.add_estimate(
proto.estimate.Estimate.create_gst_estimate(
res, smq1Q_XY.target_model(), smq1Q_XY.target_model(),
[self.model]*len(self.maxLengthList), parameters={'final_objfn_builder': builder}),
estimate_key="linresponse"
)
estLR = res.estimates['linresponse']
#estLR.add_confidence_region_factory('final iteration estimate', 'final') #Could do this, but use alt. method for more coverage
with self.assertRaises(KeyError):
estLR.create_confidence_region_factory('final iteration estimate', 'final') #won't create by default
cfctryLR = estLR.create_confidence_region_factory('final iteration estimate', 'final', create_if_needed=True) #now it will
self.assertTrue( estLR.has_confidence_region_factory('final iteration estimate', 'final'))
#cfctryLR = estLR.create_confidence_region_factory('final iteration estimate', 'final') #done by 'get' call above
self.assertFalse( cfctryLR.can_construct_views() ) # b/c no hessian or LR enabled yet...
cfctryLR.enable_linear_response_errorbars() #parent results object is used to automatically populate params
#self.assertTrue( cfctryLR.can_construct_views() )
ci_linresponse = cfctryLR.view( 95.0, 'normal', None)
mdl_dummy = cfctryLR.model # test method
s = pickle.dumps(cfctryLR) # test pickle
pickle.loads(s)
#Add estimate for with bad objective ---------------------------------------------------------
class FooBar:
def __init__(self):
self.cls_to_build = list # an invalid objective class
self.regularization = {}
self.penalties = {}
res.add_estimate(
proto.estimate.Estimate.create_gst_estimate(
res, smq1Q_XY.target_model(), smq1Q_XY.target_model(),
[self.model]*len(self.maxLengthList), parameters={'final_objfn_builder': FooBar()}),
estimate_key="foo"
)
est = res.estimates['foo']
est.add_confidence_region_factory('final iteration estimate', 'final')
with self.assertRaises(ValueError): # bad objective
est.create_confidence_region_factory('final iteration estimate', 'final').compute_hessian()
# Now test each of the views we created above ------------------------------------------------
for ci_cur in (ci_std, ci_noproj, ci_opt, ci_intrinsic, ci_linresponse):
s = pickle.dumps(ci_cur) # test pickle
pickle.loads(s)
#linear response CI doesn't support profile likelihood intervals
if ci_cur is not ci_linresponse: # (profile likelihoods not implemented in this case)
ar_of_intervals_Gx = ci_cur.retrieve_profile_likelihood_confidence_intervals(L("Gxpi2", 0))
ar_of_intervals_rho0 = ci_cur.retrieve_profile_likelihood_confidence_intervals(L("rho0"))
ar_of_intervals_M0 = ci_cur.retrieve_profile_likelihood_confidence_intervals(L("Mdefault"))
ar_of_intervals = ci_cur.retrieve_profile_likelihood_confidence_intervals()
with self.assertRaises(ValueError):
ci_cur.retrieve_profile_likelihood_confidence_intervals("foobar") #invalid label
def fnOfGate_float(mx,b):
return float(mx[0,0])
def fnOfGate_complex(mx,b):
return complex(mx[0,0] + 1.0j)
def fnOfGate_0D(mx,b):
return np.array( float(mx[0,0]) )
def fnOfGate_1D(mx,b):
return mx[0,:]
def fnOfGate_2D(mx,b):
return mx[:,:]
def fnOfGate_2D_complex(mx,b):
return np.array(mx[:,:] + 1j*mx[:,:],'complex')
def fnOfGate_3D(mx,b):
return np.zeros( (2,2,2), 'd') #just to test for error
fns = (fnOfGate_float, fnOfGate_0D, fnOfGate_1D,
fnOfGate_2D, fnOfGate_3D)
if ci_cur is not ci_linresponse: # complex functions not supported by linresponse CIs
fns += (fnOfGate_complex, fnOfGate_2D_complex)
for fnOfOp in fns:
FnClass = gsf.opfn_factory(fnOfOp)
FnObj = FnClass(self.model, L('Gxpi2',0))
if fnOfOp is fnOfGate_3D:
with self.assertRaises(ValueError):
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
else:
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
df, f0 = self.runSilent(ci_cur.compute_confidence_interval,
FnObj, return_fn_val=True, verbosity=4)
##SHORT-CIRCUIT linear reponse here to reduce run time
if ci_cur is ci_linresponse: continue
def fnOfVec_float(v,b):
return float(v[0])
def fnOfVec_0D(v,b):
return np.array( float(v[0]) )
def fnOfVec_1D(v,b):
return np.array(v[:])
def fnOfVec_2D(v,b):
return np.dot(v.T,v)
def fnOfVec_3D(v,b):
return np.zeros( (2,2,2), 'd') #just to test for error
for fnOfVec in (fnOfVec_float, fnOfVec_0D, fnOfVec_1D, fnOfVec_2D, fnOfVec_3D):
FnClass = gsf.vecfn_factory(fnOfVec)
FnObj = FnClass(self.model, 'rho0', 'prep')
if fnOfVec is fnOfVec_3D:
with self.assertRaises(ValueError):
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
else:
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
df, f0 = self.runSilent(ci_cur.compute_confidence_interval,
FnObj, return_fn_val=True, verbosity=4)
for fnOfVec in (fnOfVec_float, fnOfVec_0D, fnOfVec_1D, fnOfVec_2D, fnOfVec_3D):
FnClass = gsf.vecfn_factory(fnOfVec)
FnObj = FnClass(self.model, 'Mdefault:0', 'effect')
if fnOfVec is fnOfVec_3D:
with self.assertRaises(ValueError):
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
else:
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
df, f0 = self.runSilent(ci_cur.compute_confidence_interval,
FnObj, return_fn_val=True, verbosity=4)
def fnOfSpam_float(rhoVecs, povms):
lbls = list(povms[0].keys())
return float( np.dot( rhoVecs[0].T, povms[0][lbls[0]] ) )
def fnOfSpam_0D(rhoVecs, povms):
lbls = list(povms[0].keys())
return np.array( float( np.dot( rhoVecs[0].T, povms[0][lbls[0]] ) ) )
def fnOfSpam_1D(rhoVecs, povms):
lbls = list(povms[0].keys())
return np.array( [ float(np.dot( rhoVecs[0].T, povms[0][lbls[0]]) ), 0] )
def fnOfSpam_2D(rhoVecs, povms):
lbls = list(povms[0].keys())
return np.array( [[ float(np.dot( rhoVecs[0].T, povms[0][lbls[0]] )), 0],[0,0]] )
def fnOfSpam_3D(rhoVecs, povms):
return np.zeros( (2,2,2), 'd') #just to test for error
for fnOfSpam in (fnOfSpam_float, fnOfSpam_0D, fnOfSpam_1D, fnOfSpam_2D, fnOfSpam_3D):
FnClass = gsf.spamfn_factory(fnOfSpam)
FnObj = FnClass(self.model)
if fnOfSpam is fnOfSpam_3D:
with self.assertRaises(ValueError):
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
else:
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
df, f0 = self.runSilent(ci_cur.compute_confidence_interval,
FnObj, return_fn_val=True, verbosity=4)
def fnOfGateSet_float(mdl):
return float( mdl.operations['Gxpi2',0][0,0] )
def fnOfGateSet_0D(mdl):
return np.array( mdl.operations['Gxpi2',0][0,0] )
def fnOfGateSet_1D(mdl):
return np.array( mdl.operations['Gxpi2',0][0,:] )
def fnOfGateSet_2D(mdl):
return np.array( mdl.operations['Gxpi2',0] )
def fnOfGateSet_3D(mdl):
return np.zeros( (2,2,2), 'd') #just to test for error
for fnOfGateSet in (fnOfGateSet_float, fnOfGateSet_0D, fnOfGateSet_1D, fnOfGateSet_2D, fnOfGateSet_3D):
FnClass = gsf.modelfn_factory(fnOfGateSet)
FnObj = FnClass(self.model)
if fnOfGateSet is fnOfGateSet_3D:
with self.assertRaises(ValueError):
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
else:
df = ci_cur.compute_confidence_interval(FnObj, verbosity=0)
df, f0 = self.runSilent(ci_cur.compute_confidence_interval,
FnObj, return_fn_val=True, verbosity=4)
#TODO: assert values of df & f0 ??
def test_pickle_ConfidenceRegion(self):
data = proto.ProtocolData(self.edesign, self.ds)
res = proto.ModelEstimateResults(data, proto.StandardGST(modes="full TP"))
res.add_estimate(
proto.estimate.Estimate.create_gst_estimate(
res, smq1Q_XY.target_model(), smq1Q_XY.target_model(),
[self.model]*len(self.maxLengthList), parameters={'objective': 'logl'}),
estimate_key="default"
)
est = res.estimates['default']
est.add_confidence_region_factory('final iteration estimate', 'final')
self.assertTrue( est.has_confidence_region_factory('final iteration estimate', 'final'))
cfctry = est.create_confidence_region_factory('final iteration estimate', 'final')
cfctry.compute_hessian()
self.assertTrue( cfctry.has_hessian )
cfctry.project_hessian('std')
ci_std = cfctry.view( 95.0, 'normal', 'std')
s = pickle.dumps(cfctry)
cifctry2 = pickle.loads(s)
s = pickle.dumps(ci_std)
ci_std2 = pickle.loads(s)
#TODO: make sure ci_std and ci_std2 are the same
def test_mapcalc_hessian(self):
chi2Hessian = pygsti.chi2_hessian(self.model, self.ds)
mdl_mapcalc = self.model.copy()
mdl_mapcalc._calcClass = MapForwardSimulator
chi2Hessian_mapcalc = pygsti.chi2_hessian(self.model, self.ds)
self.assertArraysAlmostEqual(chi2Hessian, chi2Hessian_mapcalc)
if __name__ == "__main__":
unittest.main(verbosity=2)