Repository navigation
Expand file tree
/
Copy pathglossary.yaml
More file actions
890 lines (815 loc) · 45.6 KB
/
Copy pathglossary.yaml
File metadata and controls
890 lines (815 loc) · 45.6 KB
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
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
# The glossary: the single source of the register's phrases and the symbols a user meets.
#
# Compiled by scripts/build_generated.py into the docsite Glossary page (term and short only,
# grouped alphabetically). The other fields are the source of the paper's register table
# (docs/superpowers/notes/notation.md holds the long form).
#
# Fields
# id unique lower-case slug
# term the phrase the prose uses
# symbol optional, LaTeX without delimiters
# short one line for a user (no line break)
# register the register's phrase for the object
# formal the formal object
# not what it is not
# defined_in optional, where it is defined or proved (lemma and ledger ids)
# layer optional: decision, design or method, the layer of the nested optimizations the
# word belongs to (see `preamble`)
# used_in optional: the register or registers in which this is the word's ordinary meaning
# elsewhere optional: how another register uses the same word (the caveat; it lives here and
# in the detailed sections, never on an intuitive surface)
# aliases optional list of strings: variants the prose may use for the term (plural,
# abbreviation); each resolves to this entry
# avoid optional list of {phrase, reason}: variants that must not appear in prose, each
# with the reason (notes/vocabulary-review.md)
#
# File-level keys besides `entries`
# preamble the three layers in full, shown at the top of the docs Glossary page
# assumed optional list of {term, reason}: standard vocabulary assumed of the reader (calculus,
# linear algebra, convex optimization at the level of Boyd and Vandenberghe,
# scikit-learn's documented terms), each with the reason it needs no entry here
# stems optional list of strings: the stems that mark a term of art in this subject; a
# phrase built on one must resolve to an entry, an alias or an assumed term
preamble: |
Three optimizations are nested in this project, and one word can mean something different in
each. The decision layer is the decision-maker's choice of a design: its words are objective
(a metric the decision-maker wants more of), constraint (a requirement's limit) and the Pareto
notions. The design layer is the designer's optimization of a configuration, inside which the
fitted model appears as a function of the metrics: generalized geometric program and log-convex
belong here. The method layer is the fit, an optimization over the exponents: its word is loss,
the function the solver minimizes, which is scikit-learn's word. The words loss, metric,
attribute and objective mean different things in decision analysis, in optimization and in
machine learning. This glossary gives each one meaning, names the layer and the register of every
entry, and the surfaces hold to it: plain words on the intuitive surfaces, the caveats in the
entries and the detailed sections.
entries:
- id: 'configuration'
term: 'configuration'
symbol: 'c'
short: 'The design variables the engineer chooses (wheel class, gains); carried as covariates, not fitted.'
register: 'configuration'
formal: 'the design variables of one evaluated design'
not: 'an input of the fit; the fit sees attributes and metrics only'
defined_in: 'Prediction from configuration'
- id: 'attributes'
term: 'attributes'
symbol: 'a'
short: 'What the evaluation returns (seconds, kilograms); raw, in mixed units and directions.'
register: 'attributes'
formal: 'the evaluation outcomes of a configuration; the evaluation is a black box (simulator, physical test or human assessment)'
not: 'metrics'
defined_in: 'A4'
- id: 'metrics'
term: 'metrics'
symbol: 'm \in \mathbb{R}^d_{++}'
short: 'The conditioned attributes: strictly positive, larger is better; the input to the fit.'
register: 'metrics'
formal: 'the conditioned attributes, a point of the positive orthant'
not: 'the raw attributes; zero is tolerated in data and masked in the fit'
defined_in: 'A1, A4, L9'
- id: 'log-metrics'
term: 'log-metrics'
symbol: 'u = \log m'
short: 'The metrics in log space, where every fit happens and a monomial becomes a hyperplane.'
register: 'log-metrics'
formal: 'the componentwise logarithm of the metrics'
not: 'the chart coordinates Z'
defined_in: 'L1, A1'
- id: 'conditioning'
term: 'conditioning'
short: 'The analyst''s step that turns attributes into metrics; LogAffinePreconditioner does it.'
register: 'conditioning'
formal: 'the log-affine, invertible map from attributes to metrics, log m_j = alpha_j log(shift_j + attribute_j) + b_j; the lemma scripts write the shift a_j and the raw attribute r_j (s_j in the statements of A4 and L9), the main text the attribute a_j and the shift s_j'
not: 'part of the fit'
defined_in: 'A4, L9'
- id: 'number-of-designs'
term: 'number of designs'
symbol: 'N'
short: 'The rows of the table; each row is one evaluated configuration.'
register: 'number of designs'
formal: 'the number of evaluated configurations'
not: 'a sample drawn from the attainable set'
defined_in: 'The model'
- id: 'number-of-metrics'
term: 'number of metrics'
symbol: 'd'
short: 'The length of the metric vector; the dimension of the trade space.'
register: 'number of metrics'
formal: 'the dimension of the trade space'
not: 'the number of configuration variables'
defined_in: 'The model, A0'
- id: 'number-of-terms'
term: 'number of terms'
symbol: 'K'
short: 'How many monomials the frontier has; the analyst''s choice, never selected by the fit.'
register: 'number of terms'
formal: 'the number of monomials of the posynomial frontier'
not: 'a selected or estimated quantity'
defined_in: 'L10, A6'
- id: 'invariant-value'
term: 'invariant value'
symbol: 'V(m) = e^{-\beta}\prod_j m_j^{w_j}'
short: 'A weighted geometric mean of the metrics; doubling every metric doubles it, so it reads as a factor of improvement.'
register: 'invariant value'
formal: 'the weighted geometric mean of the metrics, scaled by e^(-beta) so that V = 1 on the fitted reference surface (the lemma script L1 states the unscaled mean, beta = 0)'
not: 'a probability or a bounded score; V above 1 is possible at finite temperature'
defined_in: 'L1, A1, A2'
- id: 'hyperbolic'
term: 'hyperbolic'
short: 'The level curves of V, products of powers of the metrics held constant, the rectangular hyperbola in two metrics. Not negative-curvature geometry, and not the graph measure of Färe, Grosskopf and Lovell.'
register: 'hyperbolic'
formal: 'the level curves of V, a product of powers of the metrics set equal to a constant. In two metrics with equal exponents the rectangular hyperbola m1 m2 = constant'
not: 'negative-curvature geometry (the setting of hyperbolic principal component analysis), or the hyperbolic graph measure of efficiency of Färe, Grosskopf and Lovell'
defined_in: 'Definitions'
- id: 'exponent-vector'
term: 'exponent vector'
symbol: 'w_k'
short: 'The weights of term k on the simplex over the metrics; its positive entries name the metrics that contend.'
register: 'exponent vector'
formal: 'a point on the simplex over the d metrics'
not: 'a loading vector with sign or unit norm'
defined_in: 'A2, L3, L4'
- id: 'balanced-point'
term: 'balanced point'
symbol: 'U^\top p'
short: 'At the exact-maximum fit, the mean of the log-metric rows under the optimal dual weights p; its components are equal on the support of the exponents.'
register: 'balanced point'
formal: 'the point U^T p, p an optimal dual weight vector (at finite tau, the softmax weights of the designs)'
not: 'a design of the table, or the mean of the data'
defined_in: 'L3, R1'
- id: 'cell'
term: 'cell'
short: 'The designs assigned to one term, those whose largest term it is; the cells partition the rows, one per term.'
register: 'cell'
formal: 'the set of rows whose maximizing term is k'
not: 'a term, which is a piece of the model, not a set of designs'
defined_in: 'A8, L15'
aliases:
- 'cells'
- id: 'binding-term'
term: 'binding term'
short: 'The term whose piece the surface follows at a point (the supervisor: "binds" in the infinite-temperature sense); at a finite temperature the term carries the largest weight.'
register: 'binding term'
formal: 'at a point, the term k with the largest value a_k + s * sum_j w_kj log c_j (plotting._ternary_region_of); exact at the exact maximum (infinite temperature), where the surface follows that term''s piece, an active constraint; at a finite temperature the term carries the largest weight, and the surface lies within 1/tau of the binding piece. The designs so assigned are the term''s cell'
not: 'a cluster of designs'
defined_in: 'plotting._ternary_region_of (docstring)'
- id: 'characteristic-point'
term: 'characteristic point'
short: 'A term''s characteristic point is the point of the surface whose shares of the metrics are the term''s exponents, where the term is largest for a given total.'
register: 'characteristic point'
formal: 'by the weighted inequality of the means, among points with the same total of the metrics (each in units of its typical size) a term''s monomial is largest where the shares equal its exponents; equivalently, on the term''s piece of the surface it is the point with the smallest total. The ternary''s numbered markers sit at these shares (star, diamond, square for terms 1, 2, 3)'
not: 'a design of the table; a balanced point (the dual-weighted mean)'
defined_in: 'plot_ternary (docstring, Notes)'
- id: 'term'
term: 'monomial term'
short: 'One monomial of the posynomial: the model object, one trade-off; in log space one flat piece of the fitted reference surface.'
register: 'monomial term'
formal: 'a product of powers of the metrics; in log space a hyperplane, the flat piece of the fitted reference surface where that term is the largest'
not: 'a cluster of designs; a face of the Pareto boundary or of the data''s convex hull'
defined_in: 'L10'
aliases:
- 'term'
- 'terms'
- 'monomial terms'
avoid:
- phrase: 'facet'
reason: 'a second name for the term that brought the geometry (a face of a solid) to mind before the algebra; the model object has one name, monomial term, shortened to term; the word stays only in its polyhedral sense (R1, L11), each such sentence marked vocab-exempt (WP24, A1)'
- id: 'monomial'
term: 'monomial'
short: 'A positive coefficient times the metrics raised to real powers; one term of a posynomial, a hyperplane in log-metrics.'
register: 'monomial'
formal: 'a positive constant times each metric raised to a real power; its logarithm is affine in the log-metrics'
not: 'a posynomial, or a cluster of designs'
defined_in: 'Definition Posynomial, L4.17'
aliases:
- 'monomials'
- id: 'log-convex'
term: 'log-convex'
short: 'Convex as a function of the log-metrics u = log m; the fitted reference surface is log-convex, so no number of terms represents a dent or a hole.'
register: 'log-convex'
formal: 'a function convex in u = log m'
not: 'convex in the metrics m'
defined_in: 'L4, L10'
aliases:
- 'log-convex surface'
- id: 'posynomial'
term: 'posynomial'
short: 'A sum of monomials; the fit''s K-term sum is one, and the fitted reference surface is where that sum equals one.'
register: 'posynomial'
formal: 'a sum of monomials with positive coefficients; here the sum of the fit''s terms, each a monomial in the metrics'
not: 'the invariant, which is a root of the sum, or a polynomial with coefficients of either sign'
defined_in: 'Definition Posynomial, L10'
aliases:
- 'posynomials'
- id: 'generalized-posynomial'
term: 'generalized posynomial'
short: 'A function built from posynomials by sums, products, positive powers and maxima (Boyd et al. 2007); the invariant V_K belongs to this class.'
register: 'generalized posynomial'
formal: 'a function of the class closed under sums, products, positive powers and maxima that contains the posynomials; a class of functions'
not: 'a name for the fitted reference surface, which is a level set of a posynomial, or for a class of problems'
defined_in: 'Boyd et al. 2007, section 5'
aliases:
- 'generalized posynomials'
- id: 'geometric-program'
term: 'geometric program'
short: 'The minimization of a posynomial under posynomial inequality and monomial equality constraints; convex after x = exp(y).'
register: 'geometric program'
formal: 'minimize a posynomial subject to posynomial constraints at most one and monomial constraints equal to one, over positive variables; the logarithmic change of variables makes it a convex program'
not: 'the several-term fit as one program, which is not convex in the terms (L16). With the weights given, the fit of all terms is a generalized geometric program'
defined_in: 'L4.14 to L4.24'
aliases:
- 'geometric programs'
- 'GP'
- 'geometric programming'
- id: 'generalized-geometric-program'
term: 'generalized geometric program'
short: 'A program that minimizes a generalized posynomial under generalized-posynomial inequality constraints and monomial equality constraints (Boyd et al. 2007, section 5.4); convex after x = exp(y).'
register: 'generalized geometric program'
formal: 'minimize a generalized posynomial subject to generalized-posynomial constraints at most one and monomial constraints equal to one, over positive variables. The logarithmic change of variables makes it a convex program, and added variables turn it into a geometric program'
not: 'a function class, or the unsupervised several-term fit as one program (not convex in the terms); the older use of the phrase for signomial programs; a design problem that uses the fitted model is one'
defined_in: 'Boyd et al. 2007, section 5.4; L16 (iii)'
aliases:
- 'generalized geometric programs'
- 'generalized geometric programming'
- 'GGP'
- id: 'fitted-reference-surface'
term: 'fitted reference surface'
symbol: '\Phi(u) = 0'
short: 'The level V = 1 of the fitted invariant at finite temperature, a model of the data underlying the table; rows above it exist by construction.'
register: 'fitted reference surface'
formal: 'the level set V = 1 of the fitted invariant, where the posynomial sum equals one; it models the data (the evaluated designs)'
not: 'the Pareto boundary, or a bound that no design exceeds'
defined_in: 'L3, L5, G-1'
avoid:
- phrase: 'generalized posynomial reference surface'
reason: 'the surface is the level set of a posynomial; generalized names the class the invariant belongs to'
- phrase: 'generalized posynomial surface'
reason: 'the surface is the level set of a posynomial; generalized names the class the invariant belongs to'
- phrase: 'model of the table'
reason: 'the surface models the data underlying the table; the table and its reading are views'
- phrase: 'models the table'
reason: 'the surface models the data underlying the table; the table and its reading are views'
- phrase: 'fitted surface'
reason: 'the fitted reference surface is the one named object; a bare fitted surface leaves open which of the three frontier objects is meant'
- phrase: 'learned surface'
reason: 'the fitted reference surface is the one named object; learned surface is a second name for it'
- phrase: 'posynomial surface'
reason: 'the fitted reference surface is the one named object; the posynomial is the fit, the surface is its level V = 1'
- phrase: 'fitted posynomial reference surface'
reason: 'the surface is named fitted reference surface; the posynomial qualifier adds a second name'
- id: 'supporting-envelope'
term: 'supporting envelope'
short: 'The fit at infinite temperature, the exact maximum; a hyperplane that supports the observed rows from above.'
register: 'supporting envelope'
formal: 'the hyperplane of the infinite-temperature fit, found by a linear program'
not: 'an envelope of the attainable set'
defined_in: 'L5, I1, G-1'
avoid:
- phrase: 'sample envelope'
reason: 'the envelope of the sample is the supporting envelope; use the glossary name so the three frontier objects stay apart'
- id: 'pareto-boundary'
term: 'Pareto boundary'
short: 'The nondominated points of the attainable set, a property of the system the data sample.'
register: 'Pareto boundary'
formal: 'the set of nondominated points of the attainable set'
not: 'the fitted reference surface or the supporting envelope'
defined_in: 'L11, G-1, G-13'
- id: 'frontier'
term: 'frontier'
short: 'Everyday word for one of three objects; the pages name them as the fitted reference surface, the supporting envelope or the Pareto boundary.'
register: 'frontier'
formal: 'not a defined object; allowed where the context fixes which object is meant'
not: 'a synonym for the Pareto boundary'
defined_in: 'Words'
- id: 'energy-differential'
term: 'energy differential'
symbol: '\log V'
short: 'The log of the invariant value: a design''s level relative to the fitted reference surface, by analogy with a potential; negative inside, positive beyond; returned by score_samples.'
register: 'energy differential'
formal: 'the potential evaluated at the design, a path-independent difference'
not: 'a physical energy or a Euclidean distance'
defined_in: 'L1, L2, L13'
- id: 'level-offset'
term: 'level offset'
symbol: '\log V'
short: 'Another name for the energy differential, read as the offset from the level V = 1.'
register: 'level offset'
formal: 'the energy differential read against the level V = 1'
not: 'a distance'
defined_in: 'L1, L2'
- id: 'improvement-factor'
term: 'improvement factor'
symbol: 'V^{-1/w_j}'
short: 'The factor on one metric alone that brings a design to V = 1.'
register: 'improvement factor'
formal: 'the single-metric scaling that carries the design to V = 1'
not: 'the factor that brings the design to the Pareto boundary'
defined_in: 'L7'
- id: 'chart'
term: 'chart'
symbol: '(Z, \log V)'
short: 'The reversible transformation between metric space and the fitting space; nothing is lost either way.'
register: 'chart'
formal: 'the reversible map from metrics to (Z, log V)'
not: 'defined for a row with a zero metric'
defined_in: 'L8, L9'
- id: 'position-along-the-frontier'
term: 'position along the frontier'
symbol: 'Z \in \mathbb{R}^{d-1}'
short: 'The other d - 1 chart coordinates, free of overall scale; where in the trade space a design sits.'
register: 'position along the frontier'
formal: 'the Helmert log-ratio coordinates of the chart'
not: 'a position on the Pareto boundary'
defined_in: 'L8, A7, I6'
- id: 'white-space'
term: 'white space'
short: 'A region of the chart where the fit speaks and no evaluated design is; the one word for it.'
register: 'white space'
formal: 'a hole in the distribution of Z'
not: 'evidence that a design there is attainable; the term gap or an entry of the gaps ledger'
layer: 'design'
used_in: 'this method'
defined_in: 'Novelty and white space'
avoid:
- phrase: 'hole in the chart'
reason: 'one word for an unoccupied region of the chart: white space'
- phrase: 'holes in the chart'
reason: 'one word for an unoccupied region of the chart: white space'
- phrase: 'gaps in the table'
reason: 'gap names the term gap and the gaps ledger; an unoccupied region is white space'
- phrase: 'table has gaps'
reason: 'gap names the term gap and the gaps ledger; an unoccupied region is white space'
- id: 'temperature'
term: 'fit temperature'
symbol: '\tau'
short: 'How hard the maximum is in the one-sided fit; infinite is the exact maximum, finite gives a band of near-frontier designs.'
register: 'fit temperature'
formal: 'the entropic softness of the log-mean-exp'
not: 'a noise level or a model parameter fitted from the data'
defined_in: 'L5, A3, G-9'
avoid:
- phrase: 'working temperature'
reason: 'the fit temperature is the one name for tau; working temperature is a second name for it'
- phrase: 'fitted temperature'
reason: 'the fit temperature is a setting the analyst chooses or selects, not a fitted quantity'
- phrase: 'selected temperature'
reason: 'the fit temperature is the one name for tau, however it was chosen'
- phrase: 'soft maximum temperature'
reason: 'the fit temperature is the one name for tau, the entropic softness of the soft maximum'
- phrase: 'matching temperature'
reason: 'the fit temperature is the one name for tau; the band is 1/tau'
- id: 'term-temperature'
term: 'term temperature'
symbol: '\kappa'
short: 'The temperature of the soft weights over the terms in the several-term fit; infinity is the sharp limit of one label per design.'
register: 'term temperature'
formal: 'the softness of the weights softmax_k(kappa score) over the K terms'
not: 'the fit temperature'
defined_in: 'L10, A6'
avoid:
- phrase: 'extension temperature'
reason: 'the term temperature (kappa) is the one name for the softness of the weights over the terms'
- id: 'support-weights'
term: 'support weights'
symbol: 'p_i'
short: 'The softmax weights of the designs in the fit; which designs hold up the fitted reference surface.'
register: 'support weights'
formal: 'the softmax weights of the rows in the log-mean-exp'
not: 'probabilities of a sampling model'
defined_in: 'L3, A3'
- id: 'reach'
term: 'reach'
symbol: '\beta_k'
short: 'How far the data extend in direction w_k: the soft maximum of their scores along it, the largest score at infinite temperature.'
register: 'reach'
formal: 'the log-mean-exp of the scores w_k . u_i over the rows, J_tau(U w_k)'
not: 'a fitted constant unrelated to the data'
defined_in: 'L5, Reach and share'
- id: 'term-share'
term: 'term share'
symbol: '\mathrm{share}_k'
short: 'The fraction of the near-frontier designs in a term''s cell (term_share_), showing which terms carry most of the frontier. It depends on the band, and it gives the order of the terms, largest first.'
register: 'term share'
formal: 'the fraction of the near-frontier rows whose cell is term k; returned as term_share_'
not: 'the row share, which counts all rows, or a share of a total of metrics'
defined_in: 'L5, Reach and share'
aliases:
- 'term shares'
- id: 'near-frontier-band'
term: 'near-frontier band'
symbol: '\lvert\log V\rvert < 1/\tau'
short: 'The designs with log V strictly between -1/tau and 1/tau. The fit weighs every design in its soft maximum, and a design inside the band weighs within a factor e of one on the surface.'
register: 'near-frontier band'
formal: 'the set of rows whose log V lies strictly between -1/tau and 1/tau for the fitted invariant'
not: 'a confidence band or a noise level'
defined_in: 'L5.25, Temperatures and band'
- id: 'effective-sample-size'
term: 'effective sample size'
symbol: 'n_{\mathrm{eff}}'
short: 'The inverse sum of squares of the support weights; returned as n_eff_.'
register: 'effective sample size'
formal: '1 / sum_i p_i^2 for the support weights p_i'
not: 'the number of rows N'
defined_in: 'L3, A3'
- id: 'contending-set'
term: 'contending set'
short: 'The metrics with positive exponent in the exact-maximum fit; the members of the n-lemma.'
register: 'contending set'
formal: 'the support of the exponent vector at infinite temperature'
not: 'the participants at the fit temperature; a statement about the whole system'
defined_in: 'L3, I1, G-2, G-12'
- id: 'n-lemma'
term: 'n-lemma'
symbol: '\nu_k'
short: 'How many of the d metrics contend in a term; a dilemma is 2, a trilemma 3; found from the data, not assumed.'
register: 'n-lemma'
formal: 'the size of the contending set of one term'
not: 'a count of metrics in tension in the data'
defined_in: 'L3, G-2'
- id: 'elasticity'
term: 'elasticity'
symbol: '-w_k / w_j'
short: 'Within a term, the percentage of one metric given up per percentage gained in another.'
register: 'elasticity'
formal: 'the derivative of one log-metric with respect to another along the term''s hyperplane; a ratio of exponents; the log-log form of the tension, constant on the term'
not: 'a causal exchange or an intervention effect'
defined_in: 'L7, A1, A2, G-3'
aliases:
- 'elasticities'
avoid:
- phrase: 'exchange rate'
reason: 'it reads as money; the quantity is a ratio of exponents, the percentage of one metric given up per percentage gained in another, and the theorem that derives it (L7) is named for elasticities'
- phrase: 'exchange rates'
reason: 'it reads as money; the quantity is a ratio of exponents, the percentage of one metric given up per percentage gained in another, and the theorem that derives it (L7) is named for elasticities'
- phrase: 'constant rate'
reason: 'the rate at which one metric trades for another within a term is the elasticity, constant within a term'
- phrase: 'varying rate'
reason: 'the rate at which one metric trades for another is the elasticity, which changes from term to term'
- id: 'objective'
term: 'objective'
short: 'A metric the decision-maker wants more of; the objectives are a subset of the metrics, and a metric need not be one.'
register: 'objective'
formal: 'a metric in the subset of the metrics on which the decision-maker chooses among designs'
not: 'the function the solver minimizes (that is the loss), a requirement''s limit (that is a constraint), or any metric of the fit'
layer: 'decision'
used_in: 'decision analysis, multi-objective optimization'
elsewhere: 'optimization and machine learning call the function a solver minimizes the objective; here that function is the loss'
defined_in: 'Preliminaries and notation'
aliases:
- 'objectives'
avoid:
- phrase: 'fit objective'
reason: 'the function the fit minimizes is the loss; objective is the decision-maker''s subset of the metrics'
- phrase: 'soft objective'
reason: 'the function the soft fit minimizes is the loss; objective is the decision-maker''s subset of the metrics'
- phrase: 'noisy objective'
reason: 'the function the fit minimizes under noise is the loss; objective is the decision-maker''s subset of the metrics'
- phrase: 'convex objective'
reason: 'the function the solver minimizes is the loss; objective is the decision-maker''s subset of the metrics'
- phrase: 'dual objective'
reason: 'the dual of the fit''s problem has a value, not an objective; objective is the decision-maker''s subset of the metrics'
- phrase: 'primal objective'
reason: 'the primal value of the fit''s problem is the loss at its minimizer; objective is the decision-maker''s subset of the metrics'
- id: 'constraint'
term: 'constraint'
short: 'A requirement''s limit on a measure: a design meets it or does not. The fit never sees constraints; they are read after it.'
register: 'constraint'
formal: 'a bound on an attribute or a metric, applied to designs after the fit'
not: 'a condition of the fit itself (the exponents lie on the simplex); an objective'
layer: 'decision'
used_in: 'decision analysis, optimization'
defined_in: 'Requirements after the fit; Appendix B.12'
aliases:
- 'constraints'
- id: 'trade-off'
term: 'trade-off'
short: 'Metrics that cannot all rise together: raising one costs another. The plain word; its local measure is the tension, its constant-per-term form the elasticity.'
register: 'trade-off'
formal: 'the negative tension between metrics on a term of the fitted reference surface'
not: 'a causal exchange, or a count of the metrics in tension in the data'
layer: 'decision'
used_in: 'decision analysis, engineering'
aliases:
- 'trade-offs'
avoid:
- phrase: 'tradeoff'
reason: 'one spelling; the glossary''s is trade-off'
- phrase: 'tradeoffs'
reason: 'one spelling; the glossary''s is trade-off'
- id: 'tension'
term: 'tension'
symbol: 'T_{ij}'
short: 'How much one metric changes when another rises by a unit along the surface, with the others held, at a point; negative where both are carried by the term.'
register: 'tension'
formal: 'the partial derivative of m_j with respect to m_i along a term of the fitted reference surface, in the conditioned metrics: -(w_ki / w_kj)(m_j / m_i); zero where i carries no exponent, undefined where j carries none, reciprocal across the diagonal, piecewise at K > 1; the elasticity is its log-log form, constant on the term'
not: 'a correlation of the data (association), or the effect of changing a design variable; it is a rate along the fitted reference surface'
layer: 'design'
used_in: 'this method'
elsewhere: 'in mechanics a tension is a force; here it is a derivative between two metrics'
defined_in: 'L21, Definition (tension)'
aliases:
- 'tensions'
- id: 'term-tension-matrix'
term: 'tension matrix'
short: 'The tensions of every ordered pair of metrics along one term, at the term''s balanced point; from the fit.'
register: 'tension matrix'
formal: 'T_k with T_k[i, j] = -(w_ki / w_kj)(m*_j / m*_i), m* the balanced point of term k; blank where j carries no exponent'
not: 'a design''s tensions (from the data), or the association among a term''s near-frontier designs'
layer: 'design'
used_in: 'this method'
defined_in: 'L21'
aliases:
- 'tension matrices'
- 'term''s tension matrix'
- id: 'design-tensions'
term: 'design tensions'
short: 'The tension formula at one design''s own metric ratios, on the term it belongs to; from the data. In attribute units they carry the conditioning.'
register: 'design tensions'
formal: 'T_k[i, j] at the design''s metrics, k its term; in attributes, ds_j/ds_i = -(w_ki alpha_i)/(w_kj alpha_j) (a_j + s_j)/(a_i + s_i) (L9 pull-back)'
not: 'the term''s tension matrix (from the fit); a one-sided value near a seam is reported with the other term''s'
layer: 'design'
used_in: 'this method'
defined_in: 'L21, L9'
aliases:
- 'design''s tensions'
- id: 'participation-threshold'
term: 'participation threshold'
short: 'A working-temperature exponent above 1e-3 counts as participating; the threshold is a parameter of trade_off_report.'
register: 'participation threshold'
formal: 'the exponent cutoff for membership in the participating set'
not: 'a statistical test'
defined_in: 'G-2, G-5'
- id: 'numerically-zero'
term: 'numerically zero'
short: 'An exponent at or below 1e-12; the code treats it as zero, which is not the same as mathematically zero.'
register: 'numerically zero'
formal: 'an exponent below 1e-12'
not: 'mathematically zero'
defined_in: 'G-2, G-5'
- id: 'assignment'
term: 'assignment'
short: 'The term that scores a design highest; reading it off the fitted model is analysis. Only the soft fit (finite term temperature) weighs every term against every design. At the default (infinite), the assignments drive the refit.'
register: 'assignment'
formal: 'the maximizing term of the posynomial at the design'
not: 'an identity that survives ties between scores'
defined_in: 'L13, R2, G-10'
- id: 'term-gap'
term: 'term gap'
symbol: 'G_k'
short: 'The first-order gap of a term''s refit; zero exactly when the optimality conditions hold.'
register: 'term gap'
formal: 'the first-order gap of the term''s refit on the simplex'
not: 'the distance to the optimum'
defined_in: 'L12, L3, L4, I8, F12'
aliases:
- 'Frank-Wolfe gap'
avoid:
- phrase: 'weighted gap'
reason: 'the weighted refit of a term in the soft fit has a term gap G_k; weighted gap is a second name for it'
- phrase: 'maximal gap'
reason: 'the largest term gap over the terms; name it so, and the Termination record field max_gap carries it'
- id: 'row-share'
term: 'row share'
symbol: '\lvert C_k\rvert/N'
short: 'The fraction of all designs in a term''s cell; it weights the term gaps in the termination record''s bound.'
register: 'row share'
formal: 'the number of rows in cell k over N; returned as Termination.shares'
not: 'the term share, which counts the near-frontier rows only'
defined_in: 'L14, A8'
aliases:
- 'row shares'
- id: 'assignment-margin'
term: 'assignment margin'
symbol: '\gamma'
short: 'The smallest gap by which a design''s term beats the runner-up; positive means each design has one maximizing term.'
register: 'assignment margin'
formal: 'the minimum over rows of the best score minus the second-best'
not: 'the loss''s margin'
defined_in: 'L13, R2'
- id: 'stable-radius'
term: 'stable radius'
short: 'A perturbation of the terms smaller than this leaves every design''s assignment unchanged.'
register: 'stable radius'
formal: 'gamma divided by twice the largest row norm'
not: 'a bound on estimation error'
defined_in: 'L13, I9, R2'
avoid:
- phrase: 'explicit radius'
reason: 'the radius within which score changes keep the assignment is the stable radius'
- id: 'mean-slack'
term: 'mean slack'
symbol: 'F(W)'
short: 'The mean shortfall of the designs below the fitted reference surface, F(W) = -(1/N) sum_i Phi_K(u_i), returned as mean_slack_. The several-term fit minimizes it; it falls with K by construction and cannot choose K.'
register: 'mean slack'
formal: 'minus the mean over the rows of Phi_K; returned as mean_slack_'
not: 'evidence about the boundary between observations (G-18)'
defined_in: 'L14, L15, G-18'
- id: 'loss'
term: 'loss'
short: 'The function the solver minimizes: the soft maximum of the scores at one term, the mean slack at several; scikit-learn''s word. We choose it for the method layer.'
register: 'loss'
formal: 'J_tau at one term; the mean slack at several terms'
not: 'an objective (the decision-maker''s subset of the metrics); a measure of how good a design is'
layer: 'method'
used_in: 'machine learning (scikit-learn)'
elsewhere: 'optimization texts call it the objective; here that word is reserved for the decision layer'
defined_in: 'Preliminaries and notation'
avoid:
- phrase: 'solver objective'
reason: 'the function the solver minimizes is the loss'
- phrase: 'refit objective'
reason: 'the function a term refit minimizes is the loss'
- phrase: 'weight objective'
reason: 'the function minimized over the weights is the loss'
- id: 'certificate'
term: 'certificate'
short: 'That the optimizer reached a local minimizer of the loss; for a soft fit, that the second-order test holds.'
register: 'certificate'
formal: 'for a soft fit, the second-order sufficient conditions of L17. For a sharp fit, levels 1 to 3 (L15, L12, L13 and L14)'
not: 'evidence that the surface describes the designs or holds outside the table'
defined_in: 'L17, I16, L14'
- id: 'termination-status'
term: 'termination status'
short: 'Which level of guarantee the returned run meets: certified, stationary without margin, uncertified, guard stop, iteration cap, not applicable, or stationary (soft fit only).'
register: 'termination status'
formal: 'the label of the termination record'
not: 'a statement about the choice among restarts'
defined_in: 'L12 to L14, G-15, G-16'
- id: 'conditioning-of-the-exponents'
term: 'conditioning of the exponents'
symbol: 'conditioning_'
short: 'The smallest eigenvalue of the restricted Hessian at each term; small means a nearly flat direction.'
register: 'conditioning of the exponents'
formal: 'the smallest eigenvalue of the restricted Hessian of the loss on the simplex tangent space'
not: 'a confidence interval'
defined_in: 'L3, L4, R1, G-2'
- id: 'ridge'
term: 'ridge'
short: 'A penalty on the squared size of the exponents added to the loss. Noise on the metrics adds one, and the penalty pulls the fitted exponents toward equal weights.'
register: 'ridge'
formal: 'the term (tau/2) w'' Sigma w added to the loss of the fit by Gaussian noise of covariance Sigma on the metrics (L20)'
not: 'a ridge regression fit or a ridge of the data. The model has no such penalty; the noise induces it'
defined_in: 'L20, L6, L18'
- id: 'active-face'
term: 'active face'
short: 'The part of the simplex where the exponents that are zero at the fit stay zero; the fit moves only along it.'
register: 'active face'
formal: 'the face of the simplex spanned by the coordinates with positive exponent at the fit'
not: 'a face of the Pareto boundary or of the data''s convex hull'
defined_in: 'L20, R3'
- id: 'novelty'
term: 'novelty'
symbol: 'V - 1'
short: 'How far a design lies beyond the fitted reference surface; a score for white space and new capabilities.'
register: 'novelty'
formal: 'V minus 1'
not: 'a measure of attainability'
defined_in: 'Novelty and white space'
- id: 'proposal'
term: 'proposal'
short: 'A configuration chosen for a chart point that has no datum, with the chart point predicted for it and a tolerance. A conjecture until it is checked.'
register: 'proposal'
formal: 'a triple of a configuration, the chart point the regressor predicts for it, and a tolerance (the band 1/tau by default)'
not: 'a design known to land at the target, or a result'
defined_in: 'Prediction from configuration; WP41 note'
aliases:
- 'proposals'
- id: 'selection-rule'
term: 'selection rule'
short: 'How one configuration is picked for a chart point when many are predicted near it, because the way back from the chart to the design space is not unique.'
register: 'selection rule'
formal: 'a map from a target chart point to one configuration among those predicted near it'
not: 'a property of the model or of the fit'
defined_in: 'WP41 note'
aliases:
- 'selection rules'
- id: 'check'
term: 'check'
short: 'Evaluating a proposal once and comparing the measured chart point with the predicted one. Three outcomes: confirmed, refuted (a row the table lacked, so the fit is rerun) or infeasible.'
register: 'check'
formal: 'the comparison of the measured chart point of a proposal with its predicted chart point within the tolerance, together with the feasibility conditions'
not: 'an estimate of the rate at which proposals are confirmed'
defined_in: 'Appendix B.6; WP41 note'
aliases:
- 'checks'
- 'checked'
- id: 'ternary-plot'
term: 'ternary plot'
short: 'A triangle whose corners are three metrics: a design sits by its shares of their total, tinted by region, with a numbered marker for each term''s exponent vector. For reading, not for measuring distance.'
register: 'ternary plot'
formal: 'the barycentric image of the shares of three metrics. A term''s exponent vector sums to one, so it is a point of the triangle, where that term''s V is largest at a fixed total'
not: 'a plot of V, a distance between designs, or a place where V = 1; V also depends on the total'
defined_in: 'plot_ternary (docstring, Notes)'
aliases:
- 'triangle plot'
- id: 'share'
term: 'share'
short: 'A position in the triangle: a metric''s fraction of the total of the three.'
register: 'share'
formal: 'c_j = m_j / S, where S is the sum of the three metrics, each in units of its typical size; the three shares sum to one'
not: 'a row share or a term share (fractions of designs), or a weight of a term (that is a coefficient of the fit)'
defined_in: 'plot_ternary (docstring, Notes)'
aliases:
- 'shares'
- id: 'region'
term: 'region'
short: 'In the ternary plot, the set of positions where one term is the largest; an analysis of the fit, not a label on a design.'
register: 'region'
formal: 'the set of share positions where one term attains the maximum in log V; the regions partition the triangle'
not: 'a part of the fit, a cluster of designs, or a property a design carries'
defined_in: 'plot_ternary (docstring, Notes)'
aliases:
- 'regions'
- id: 'moe'
term: 'MoE'
short: 'Measure of effectiveness: a measure of how well a system serves its purpose that a stakeholder states without knowing the design; here the pointing error and the launch cost.'
register: 'MoE'
formal: 'a metric a stakeholder states on the system in use, independent of the design'
not: 'an MoP, which measures the system''s performance under the engineer''s scenario; not by itself an objective (the decision-maker chooses which metrics are objectives)'
layer: 'decision'
used_in: 'systems engineering'
defined_in: 'cubesat_r2.sysml (the mark in the doc of the metric''s element); WP23-review.md'
aliases:
- 'MoEs'
- 'measure of effectiveness'
- id: 'mop'
term: 'MoP'
short: 'Measure of performance: a measure of a system''s performance under the scenario the engineer states; here the settling time, the total mass, the control effort, the power margin, the energy balance, the peak slew power, the depth of discharge.'
register: 'MoP'
formal: 'a metric of the system''s performance under a stated scenario'
not: 'an MoE, which a stakeholder states without knowing the design; not by itself an objective'
layer: 'decision'
used_in: 'systems engineering'
defined_in: 'cubesat_r2.sysml (the mark in the doc of the metric''s element); WP23-review.md'
aliases:
- 'MoPs'
- 'measure of performance'
- id: 'launch-cost'
term: 'launch cost'
short: 'The launch rate, an illustrative USD per kilogram, times the total mass; the launch-budget requirement reads it. It is not a traded metric and not a dataset column.'
register: 'launch cost'
formal: 'rate times total mass, increasing in mass'
not: 'a traded metric or a dataset column; a cited launch price'
defined_in: 'model_v2/cubesat.sysml (calc def LaunchCost, LaunchBudgetRequirement); case-model.md, The requirements and what each reads'
- id: 'baseline-regime'
term: 'baseline regime'
short: 'The second regime of the case: the designs drawn with the proportional gain Kp from 0.03 to 1 (the stock gains), dataset cubesat-stock.'
register: 'baseline regime'
formal: 'the data of record whose gains are sampled from the stock gain box, Kp from 0.03 to 1; dataset identifier cubesat-stock (the CaseSpec key stock)'
not: 'the primary regime (aggressive gains, Kp from 0.2 to 2), the baseline configuration (one design of the SysML v2 model), or a term of the fit'
defined_in: 'Section 7, Case construction; examples/adcs/case.py (CaseSpec.regimes, regime_word)'
aliases:
- 'baseline regimes'
avoid:
- phrase: 'stock regime'
reason: 'the supervisor named the second regime the baseline regime (WP56-decisions.md); one object, one name; the identifier cubesat-stock and the key stock stay in the data'
assumed:
- term: 'convex program'
reason: 'convex optimization at the level of Boyd and Vandenberghe 2004'
- term: 'linear program'
reason: 'convex optimization at the level of Boyd and Vandenberghe 2004'
- term: 'simplex'
reason: 'the probability simplex of Boyd and Vandenberghe 2004; named in Preliminaries'
- term: 'KKT conditions'
reason: 'Boyd and Vandenberghe 2004, section 5.5.3'
- term: 'tangent space'
reason: 'multivariate calculus and convex optimization at the level of Boyd and Vandenberghe 2004: the directions along which a point can move on a constraint set; L20 and R3 use it for the active face'
- term: 'Hessian'
reason: 'multivariate calculus'
- term: 'eigenvalue'
reason: 'linear algebra'
- term: 'log-sum-exp'
reason: 'Boyd and Vandenberghe 2004, section 3.1.5; the paper uses the log-mean-exp, which it defines'
- term: 'principal component analysis'
reason: 'standard multivariate statistics; the scikit-learn documented vocabulary (PCA)'
- term: 'unsupervised learning'
reason: 'the scikit-learn documented vocabulary (Pedregosa et al. 2011)'
- term: 'feature engineering'
reason: 'the scikit-learn documented vocabulary (Pedregosa et al. 2011; Buitinck et al. 2013)'
- term: 'kernel'
reason: 'the scikit-learn documented vocabulary of kernel methods (Pedregosa et al. 2011)'
- term: 'estimator'
reason: 'the scikit-learn documented vocabulary (Pedregosa et al. 2011; Buitinck et al. 2013)'
- term: 'transformer'
reason: 'the scikit-learn documented vocabulary (Pedregosa et al. 2011; Buitinck et al. 2013)'
- term: 'facet'
reason: 'polyhedral geometry: a maximal proper face of a polyhedron; used only where a polyhedron is meant (R1''s relative interior of a facet of the supporting envelope, L11''s blocking facet), each sentence marked vocab-exempt (WP24, A1); everywhere else the model object is the monomial term'
stems:
- 'objective'
- 'loss'
- 'tension'
- 'posynomial'
- 'monomial'
- 'program'
- 'frontier'
- 'envelope'
- 'surface'
- 'chart'
- 'lemma'
- 'rate'
- 'margin'
- 'radius'
- 'temperature'
- 'gap'
- 'proposal'
- 'check'