Memoization for Functions: Difference between revisions

No edit summary
No edit summary
Line 23: Line 23:
     fib(30)=1346269;
     fib(30)=1346269;
     fib[28..30] = {514229,832040,1346269};
     fib[28..30] = {514229,832040,1346269};
  30|->1346269 : fib;
  30|->1346268 /: fib;
  {x| 30|->x:fib} = {1346269};
   END
   END


Line 32: Line 35:
* computation of relational image such as fib[28..30], which internally results in three function calls to fib
* computation of relational image such as fib[28..30], which internally results in three function calls to fib
* membership predicates of the form x|->y:fib, where x is a known value
* membership predicates of the form x|->y:fib, where x is a known value
* non-membership predicates of the form x|->y/:fib, where x is a known value


The following points are relevant:
The following points are relevant:

Revision as of 16:22, 9 May 2019


As of version 1.9.0-beta9 ProB allows you to annotate functions in the ABSTRACT_CONSTANTS section for memoization. Memoization is a technique for storing results of function applications and reusing the result if possible to avoid re-computing the function for the same arguments again.

To enable memoization you either need to

  • annotate the function in the ABSTRACT_CONSTANTS section with the pragma /*@desc memo */
  • set the preference MEMOIZE_FUNCTIONS to true. In this case ProB will try to memoize all functions in the ABSTRACT_CONSTANTS section, unless they are obviously small and finite and can thus be precomputed completely.

Take the following example:

 MACHINE MemoizationTests
 ABSTRACT_CONSTANTS
   fib /*@desc memo */,
   fact /*@desc memo */
 PROPERTIES
   fib = %x.(x:NATURAL |
            (IF x=0 or x=1 THEN 1
            ELSE fib(x-1)+fib(x-2)
            END))
   &
   fact = %x.(x:NATURAL|(IF x=0 THEN 1 ELSE x*fact(x-1) END))
 ASSERTIONS
   fib(30)=1346269;
   fib[28..30] = {514229,832040,1346269};
  30|->1346269 : fib;
  30|->1346268 /: fib;
  {x| 30|->x:fib} = {1346269};
 END

Memoization means that the recursive Fibonacci function now runs in linear time rather than in exponential time. Generally, memoization is useful for functions which are complex to compute but which are called repeatedly with the same arguments.

As can be seen above, memoization is active for

  • function calls such as fib(30) (which in turn calls fib(29) and fib(28) which are also memoized)
  • computation of relational image such as fib[28..30], which internally results in three function calls to fib
  • membership predicates of the form x|->y:fib, where x is a known value
  • non-membership predicates of the form x|->y/:fib, where x is a known value

The following points are relevant:

  • Memoization is currently only possible for functions declared in the ABSTRACT_CONSTANTS section
  • Memoization means that all results of function calls are stored. The memoization table is reset when another B machine is loaded or the same B machine is re-loaded.
  • Memoized functions are often treated in a similar way to symbolic functions. If your function is finite and relatively small, it may be better to put the function into the CONCRETE_CONSTANTS section so that it gets computed in its entirety once.
  • Memoization of a function f is currently not active for computations such as dom(f).
  • If a predicate requires the computation of the full function, e.g., ran(f), then the results will be stored and will be available for future function calls or in case the full value is needed again.


With the command-line version probcli you can use the -profile command to obtain some statistics about memoization.