7.8 asmat

This program performs the same function as the program represented in Figure 7.1, but uses the Multinode Model API instead of direct creating a pair of actors to solve the task. A workspace for the state transition matrix and the action emission matrix is defined using an internally generated assembler program which is dumped to file prg_src in the current directory.

The asmat program prints the number of node states defined by the assembler program and prints the amount of spur accumulated during interaction of a single-node model with a deterministic environment defined by three matrices hard-coded in the program source text. State transition and action emission probabilities learned in the process of interaction are dumped to file var_out in the current directory.

A random seed can be specified by a program argument. If the random seed is non-negative, then the single-node model will operate normally. If the random seed is negative, then the single-node model will generate output signals completely randomly.

Like the apsamp program, the asmat program can be built for various schemes of using a pair of actors upon which the multinode model is based. To build the program for a selected scheme, delete executable and object files of the program if they exist, and run command

make name_of_executable_of_program_asmat AP_MODE=name_of_scheme

in directory samples. The supported names of schemes are: ‘S1_S1’, ‘S1_S3’, ‘S1_L3’, ‘S3_S1’, ‘S3_S3’, ‘S3_L3’, ‘L3_L3’. If a scheme is not specified when building the program, then default scheme ‘S1_S1’ will be used. See apsamp, for a description of how to decode the names of schemes.

Here is given the output of the example program built for scheme ‘S1_S1’.

$ ./asmat -3
nstate = 30
spur = 399

$ ./asmat 3
nstate = 30
spur = 2108