next up previous
Next: In The Works Up: EMME/2 News 9 April 1990 Previous: EMME/2 Benchmarks (Part II)

Generating Random Numbers in Network and Matrix Calculator

Some users have contacted us with the suggestion that the network and matrix calculator should be enhanced in order to provide for a random number generator. Indeed, random numbers can be very useful in many cases, such as when introducing random variations on the normally assumed average time or demand values, in order to assess the robustness of the obtained solutions.

While most users might not be aware of this, random number generators can be implemented easily with the existing versions of the network and matrix calculator modules. The trick is to properly use the two intrinsics get() and put(), which are provided mainly to optimize common subexpressions within the same expression (see section III-3.3 of the User's Manual for a description).

While normally a get() can only be performed when a corresponding put() has already been executed within the same expression, a special provision is made to allow a call get(1) even if no corresponding put has been made in the same expression yet. In this case, get(1) returns the value stored in the first put() of the last expression used within the same module, or zero on the first call.

This feature can thus be used to automatically generate consecutive numbers, e.g. with the simple expression ``put(get(1)1)+''.

This feature can be used in a similar manner to implement a random number generator according to the linear congruential method. The following expressions are different implementations of the same basic method, all yielding real numbers between 0 and 1, with cycle length of 11979, 6655 and 6057 respectively:




These expressions can be used alone to initialize vectors or matrices with random values, or as a subexpression in longer calculations. In the latter case, if the get()/put() mechanism is used not just for the random numbers, you have to make sure that the put() in the random number generator precedes the other calls to put() in the expression.

Note that the random sequence is restarted whenever the module is entered. If the random numbers are to be generated in a sequence which involves more than a single call to a module (e.g. when using it in a complex macro procedure), the internal state of random number generator can be saved in a scalar (say msX) before leaving the module by using a dummy calculation with the expression get(1). Later on, when the same random sequence is to be continued, the generator can be "seeded" by an initial dummy calculation of the expression put(ms(X)).

If you are interested in more details on the theory and practice of random number generators, such as why the specific choice of the numbers used in the above expressions, refer to the following references:

next up previous
Next: In The Works Up: EMME/2 News 9 April 1990 Previous: EMME/2 Benchmarks (Part II)

Heinz Spiess, EMME/2 Support Center, Thu Jun 6 14:21:14 MET DST 1996