Bilalh / Gen

Set of tools produced during the creation of the PhD thesis "Model selection and testing for an automated constraint modelling toolchain" which focused on how to find good constraint programming models automatically and how to test them.

Building

Requirements

ghc 7.10+ to compile
Python 3.4+ to run (`./scripts/misc/install_python3.4.sh` can install python 3.4 locally as ~/bin/python3 )
Compatible Conjure, Savilerow & Minion to run

Firstly

git clone [e-mail]:stacs_cp/gen.git gen

To build (places gen in ~/.cabal/bin)

cd hs
export CONJURE_LIB=<where the conjure-private repo is>
make

Usage

gen

gen <subcommand>

Command completion for gen (add to .bash_profile, .bashrc or .zshrc as appropriate):

  source <repo_path>/hs/scripts/_gen.sh

gen reduce

Using https://bitbucket.org/stacs_cp/savilerow/issues/136/sr-generates-invalid-minion as an example

Put a file named spec.essence in a directory. Add the param as given.param if required.

Basic usage (. being the directory with spec.essence), using 30 seconds to each reduction on 1 core.

gen reduce  -p30 --from-essence . --always-compact        # Don't have choices
gen reduce  -p30 --from-essence . --choices [eprime|json] # Have choices

Add -N to hide output of running each specification since it generally not useful,

The kind of error can specified e.g.

gen reduce -p30 --from-essence --always-compact --kind=Savilerow_

Full list of kinds can be found from:

gen reduce --list-kinds
gen reduce --list-statuses

This reduces this specification

language Essence 1.3

given module_EnumSize: int
given fin4: int
given fin5: int
given fin6: int
given var2: matrix indexed by [int(1..module_EnumSize), int(1..fin4)] of int(fin5..fin6)
find b : bool
such that b = and([true | m1 : int(1..module_EnumSize)
                        , q5 : int(1..fin4)
                        , t1 : int(0..var2[m1, q5] - 1)
                        ])

to this much smaller one.

language Essence 1.3

find unused: bool
such that and([true | q5 : int(1..2), t1 : int(0..[2; int(1..1)][q5])])

gen essence

Run for 15 minutes with 60 seconds for the toolchain on 4 cores:

gen essence -p60 -t900 -c4

Add -N to hide the output of the toolchain since it generally not useful, It is still written to file.

Add -d to specific the domain depth & -e to specific the expression depth.

Add -w <JSON-FILE> to specific the weights in below format, anything not specified defaults to 100.

{"TypeSet":30
,"TypeInt":53
}

Use gen weights to output example weightings files.

More examples

Run for 15 minutes with 30 seconds for the refinement, 4 cores:

gen essence  -p30 -t900 -c4  --mode=refine

Run with 30 seconds for the toolchain on 4 cores using a directory (including sub-directories) of essence files:

gen json -d <essence-dir>
gen essence --given <essence-dir> -p30 -c4

gen instance Help

Only the required options are shown see --help for more info

gen instance-undirected [OPTIONS] essence
  Generate discriminating instance for the given essence specification using a
  baseline method

Required:
  -p    --per-model-time=INT         Time per model
  -i    --iterations=INT             Number of races
  -m    --mode=ITEM                  The suffix of the models directory
  -c    --cores=INT                  Number of cores to use, required unless
                                     CORES is set

gen instance-nsample [OPTIONS] essence
  Generate discriminating instance for the given essence specification using
  the nsample method

Required:
  -p    --per-model-time=INT         Time per model
  -i    --iterations=INT             Number of races
  -m    --mode=ITEM                  The suffix of the models directory
  -c    --cores=INT                  Number of cores to use, required unless
                                     CORES is set
  -f    --influence_radius=INT       Number of races


gen instance-noRacing [OPTIONS] essence
  Only Generate instances i.e. no racing. Results will be in the _params
  sub-directory of -o

Required:
  -i    --iterations=INT             Number of instances to generate


gen instance-summary [OPTIONS]
  Generate a summary of the results, placed in the summary sub-directory.

Required:
  -o    --output-directory=DIR  Output directory

Instance Experiments

See https://bitbucket.org/stacs_cp/essence-refinements to see how to run the experiments

Instance Experiments Results

The results of the instance experiments is hosted on the host server since bitbucket limit is 2GB

ssh://[e-mail]/cs/home/bh246/git/instance_experiments.git

Note this does not contain the .minion files or the .aux files

scripts should symlinked to gen/scripts/instance_results r should be symlinked to gen/scripts/instance_r

run ./scripts/sync_from_server_no_large.sh <server> to get results from <server> assumes that the results on ~/sampling on the <server>

run ./scripts/process_new_results.sh when new results are synced to generated the resulting data.