Comparison tolerance: Difference between revisions

From APL Wiki
Jump to navigation Jump to search
No edit summary
(Formula and description)
Line 3: Line 3:
{{quote|In an early talk Ken was explaining the advantages of tolerant comparison. A member of the audience asked incredulously, "Surely you don't mean that when <nowiki>A=B and B=C</nowiki>, A may not equal C?" Without skipping a beat, Ken replied, "Any carpenter knows that!" and went on to the next question.|—Paul Berry<ref>[[Roger Hui]]. [https://keiapl.org/anec/ Ken Iverson Quotations and Anecdotes]. 2005-09-30.</ref>}}
{{quote|In an early talk Ken was explaining the advantages of tolerant comparison. A member of the audience asked incredulously, "Surely you don't mean that when <nowiki>A=B and B=C</nowiki>, A may not equal C?" Without skipping a beat, Ken replied, "Any carpenter knows that!" and went on to the next question.|—Paul Berry<ref>[[Roger Hui]]. [https://keiapl.org/anec/ Ken Iverson Quotations and Anecdotes]. 2005-09-30.</ref>}}


The formula now used for comparison tolerance was introduced in [[SHARP APL]] by [[Robert Bernecky]] and others,<ref>[[Robert Bernecky]]. [https://www.jsoftware.com/papers/satn23.htm "Comparison Tolerance"]. SATN-23. 1977-06-10.</ref> following a proposal by [[Dick Lathwell]].<ref>[[Dick Lathwell]]. [https://doi.org/10.1145/800114.803685 APL comparison tolerance] at [[APL76]] (also reproduced in [https://www.jsoftware.com/papers/satn23.htm SATN-23]).</ref>
The rule for comparison tolerance is that numbers <syntaxhighlight lang=apl inline>a</syntaxhighlight> and <syntaxhighlight lang=apl inline>b</syntaxhighlight> are equal when
<syntaxhighlight lang=apl>(|a-b) ≤ ⎕CT×(|a)⌈(|b)</syntaxhighlight>
where <syntaxhighlight lang=apl inline>≤</syntaxhighlight> is evaluated intolerantly. This means that the allowed difference between the numbers increases as they become larger in magnitude (a relative tolerance). Comparison with zero is intolerant: only zero can equal zero. A typical value for <syntaxhighlight lang=apl inline>⎕CT</syntaxhighlight> is <syntaxhighlight lang=apl inline>1e¯14</syntaxhighlight>, which is usually large enough to accomodate multiple iterations of double-precision rounding (which introduces error on the order of <syntaxhighlight lang=apl inline>1e¯16</syntaxhighlight>) while being far smaller than typical precision of real-world measurements.
 
Comparison tolerance was available in some form since [[APL\360]], where it was described as a "fuzz" applied to some functions. The value <syntaxhighlight lang=apl inline>⎕CT</syntaxhighlight> to control it was defined with the introduction of [[system variable]]s in [[APL.SV]]. The formula now used for comparison tolerance was proposed by [[Dick Lathwell]] in 1976<ref>[[Dick Lathwell]]. [https://doi.org/10.1145/800114.803685 APL comparison tolerance] at [[APL76]] (also reproduced in [https://www.jsoftware.com/papers/satn23.htm SATN-23]).</ref> and later introduced in [[SHARP APL]] by [[Robert Bernecky]] and others<ref>[[Robert Bernecky]]. [https://www.jsoftware.com/papers/satn23.htm "Comparison Tolerance"]. SATN-23. 1977-06-10.</ref> and included in the extended APL standard [[ISO/IEC 13751:2001]].<ref>[[Adin Falkoff]] and D. L. Orth. [https://doi.org/10.1145/800137.804495 "Development of an APL standard"] at [[APL79]].</ref> However, comparison tolerance is not supported in many newer APLs such as [[ngn/apl]], [[dzaima/APL]], and [[Kap]].


The application of comparison tolerance to [[search function]]s presents problems for standard hash-based search methods.<ref>[[Roger Hui]]. [https://www.jsoftware.com/papers/Hashing.htm "Hashing for Tolerant Index-Of"] at [[Dyalog '10]].</ref><ref>[[Roger Hui]]. "Tolerant Unique" ([https://www.dyalog.com/uploads/conference/dyalog17/presentations/D10_Tolerant_Unique.zip materials (1.5 MB)], [https://dyalog.tv/Dyalog17/?v=fPWky9IOG40 video (27 mins)]) at [[Dyalog '17]].</ref>
The application of comparison tolerance to [[search function]]s presents problems for standard hash-based search methods.<ref>[[Roger Hui]]. [https://www.jsoftware.com/papers/Hashing.htm "Hashing for Tolerant Index-Of"] at [[Dyalog '10]].</ref><ref>[[Roger Hui]]. "Tolerant Unique" ([https://www.dyalog.com/uploads/conference/dyalog17/presentations/D10_Tolerant_Unique.zip materials (1.5 MB)], [https://dyalog.tv/Dyalog17/?v=fPWky9IOG40 video (27 mins)]) at [[Dyalog '17]].</ref>

Revision as of 22:55, 25 February 2024

Tolerant comparison is an inexact form of comparison used to mitigate the impact of floating-point rounding error on programs. It considers two numbers equal when their relative difference is smaller than a parameter called the comparison tolerance, and accessed with the system variable ⎕CT. In addition to the comparison functions, tolerance applies to Match and Not Match, Floor, Ceiling, and Modulus, and search functions defined in terms of Match (not Interval Index).

In an early talk Ken was explaining the advantages of tolerant comparison. A member of the audience asked incredulously, "Surely you don't mean that when A=B and B=C, A may not equal C?" Without skipping a beat, Ken replied, "Any carpenter knows that!" and went on to the next question.

—Paul Berry[1]

The rule for comparison tolerance is that numbers a and b are equal when 

(|a-b) ≤ ⎕CT×(|a)⌈(|b)

where ≤ is evaluated intolerantly. This means that the allowed difference between the numbers increases as they become larger in magnitude (a relative tolerance). Comparison with zero is intolerant: only zero can equal zero. A typical value for ⎕CT is 1e¯14, which is usually large enough to accomodate multiple iterations of double-precision rounding (which introduces error on the order of 1e¯16) while being far smaller than typical precision of real-world measurements.

Comparison tolerance was available in some form since APL\360, where it was described as a "fuzz" applied to some functions. The value ⎕CT to control it was defined with the introduction of system variables in APL.SV. The formula now used for comparison tolerance was proposed by Dick Lathwell in 1976[2] and later introduced in SHARP APL by Robert Bernecky and others[3] and included in the extended APL standard ISO/IEC 13751:2001.[4] However, comparison tolerance is not supported in many newer APLs such as ngn/apl, dzaima/APL, and Kap.

The application of comparison tolerance to search functions presents problems for standard hash-based search methods.[5][6]

External links

References

  1. ↑ Roger Hui. Ken Iverson Quotations and Anecdotes. 2005-09-30.
  2. ↑ Dick Lathwell. APL comparison tolerance at APL76 (also reproduced in SATN-23).
  3. ↑ Robert Bernecky. "Comparison Tolerance". SATN-23. 1977-06-10.
  4. ↑ Adin Falkoff and D. L. Orth. "Development of an APL standard" at APL79.
  5. ↑ Roger Hui. "Hashing for Tolerant Index-Of" at Dyalog '10.
  6. ↑ Roger Hui. "Tolerant Unique" (materials (1.5 MB), video (27 mins)) at Dyalog '17.
APL built-ins [edit]
Primitives (Timeline) Functions
Scalar
Monadic Conjugate ∙ Negate ∙ Signum ∙ Reciprocal ∙ Magnitude ∙ Exponential ∙ Natural Logarithm ∙ Floor ∙ Ceiling ∙ Factorial ∙ Not ∙ Pi Times ∙ Roll ∙ Type ∙ Imaginary ∙ Square Root ∙ Round
Dyadic Add ∙ Subtract ∙ Times ∙ Divide ∙ Residue ∙ Power ∙ Logarithm ∙ Minimum ∙ Maximum ∙ Binomial ∙ Comparison functions ∙ Boolean functions (And, Or, Nand, Nor) ∙ GCD ∙ LCM ∙ Circular ∙ Complex ∙ Root
Non-Scalar
Structural Shape ∙ Reshape ∙ Tally ∙ Depth ∙ Ravel ∙ Enlist ∙ Table ∙ Catenate ∙ Reverse ∙ Rotate ∙ Transpose ∙ Raze ∙ Mix ∙ Split ∙ Enclose ∙ Nest ∙ Cut (K) ∙ Pair ∙ Link ∙ Partitioned Enclose ∙ Partition
Selection First ∙ Pick ∙ Take ∙ Drop ∙ Unique ∙ Identity ∙ Stop ∙ Select ∙ Replicate ∙ Expand ∙ Set functions (Intersection ∙ Union ∙ Without) ∙ Bracket indexing ∙ Index ∙ Cartesian Product ∙ Sort
Selector Index generator ∙ Grade ∙ Index Of ∙ Interval Index ∙ Indices ∙ Deal ∙ Prefix and suffix vectors
Computational Match ∙ Not Match ∙ Membership ∙ Find ∙ Nub Sieve ∙ Encode ∙ Decode ∙ Matrix Inverse ∙ Matrix Divide ∙ Format ∙ Execute ∙ Materialise ∙ Range
Operators Monadic Each ∙ Commute ∙ Constant ∙ Replicate ∙ Expand ∙ Reduce ∙ Windowed Reduce ∙ Scan ∙ Outer Product ∙ Key ∙ I-Beam ∙ Spawn ∙ Function axis ∙ Identity (Null, Ident)
Dyadic Bind ∙ Compositions (Compose, Reverse Compose, Beside, Withe, Atop, Over) ∙ Inner Product ∙ Determinant ∙ Power ∙ At ∙ Under ∙ Rank ∙ Depth ∙ Variant ∙ Stencil ∙ Cut ∙ Direct definition (operator) ∙ Identity (Lev, Dex)
Quad names Index origin ∙ Comparison tolerance ∙ Migration level ∙ Atomic vector
Retrieved from ‘https://aplwiki.com/index.php?title=Comparison_tolerance&oldid=11069’