|
How to compile a simple test program?
$ mkdir test
$ cd test
$ cp /usr/share/gapc/examples/elm.gap .
$ gapc -i buyerpp elm.gap -t --kbacktrace
Meaning that the instance buyerpp (which is the product buyer*pretty)
is used, a good table configuration is automatically derived and backtracing
is used for pretty printing (faster).
$ make -f out.mf
The default filename of the generated makefile is out.mf (using -o foo.cc above changes it to foo.mf). This command compiles the generated C++.
$ ./out '1+2*3*4+5'
Answer:
( 30 , (1+((2*(3*4))+5)) )
( 30 , (1+(((2*3)*4)+5)) )
( 30 , ((1+(2*(3*4)))+5) )
( 30 , ((1+((2*3)*4))+5) )
Stochastic backtracing does not compile?
When using stochastic backtracing, the generated program needs to have libgsl installed.
Under a Debian-like distribution you can just install the libgsl package that is suggested
by the gapc package:
$ apt-cache show bellmansgapc | grep Suggest
Suggests: libgsl0-dev, libgmp3-dev, libatlas-base-dev
# apt-get install libgsl0-dev
The generated code contains a simple check for libgsl, such that a more informative error
message is displayed instead of the default link error, in case that the library is not
available.
How to use stochastic backtracing?
The adpf.gap contains an example instance:
instance pfsampleshape = fold ( ( (p_func | p_func_id ) * shape5 )
suchthat sample_filter ) ;
The p_func_id algebra is used during backtracing and the sample_filter does the actual
sampling - the p_func_id algebra extends the p_func algebra and overwrites the objective
function - instead of summation it is now the identity. The shape5 algebra is a pretty-printing
like algebra that is used during backtracing to print the candidates.
The instance is compiled via:
$ cp /usr/share/gapc/examples/adpf.gap .
$ cp /usr/share/gapc/examples/adpf_filter.hh .
$ gapc -t -i pfsampleshape --sample adpf.gap
$ make -f out.mf
The generated program respects the -r, which takes the number
of samples (of stochastic backtraces) as argument, e.g.:
$ ./out -r 1000 cgaugucaguac | head
Answer:
( 0.0280974 , )
( 0.0280974 , )
( 0.0280974 , )
( 0.0280974 , [] )
( 0.0280974 , )
( 0.0280974 , )
( 0.0280974 , )
( 0.0280974 , )
( 0.0280974 , )
The first component is not of interest - the second component
is the result of the pretty printing algebra (shape5). Counting the results we get the (sampled) shape probabilities:
$ ./out -r 1000 cgaugucaguac | \
awk -F'[(),]' '/^\(/ { ++s; ++a[$3]; } END { for (i in a) print i, a[i]/s; } '
[] 0.017
0.983
You can specify the seed of the random number generator via
exporting GSL_RNG_SEED environment variable.
|
