LASCO - Learning from Answer Sets COmpiler

LASCO is a tool that allows to compute inductive solutions of Learning from Answer Sets (LAS) tasks by genereating an ASP encoding that can be processed by state-of-the-art ASP solvers such as Clingo or DLV.

Full details are provided in my Master's Thesis.

---

🚀 Download binaries

Precompiled binaries are available in the GitHub Releases:

• lasco-linux
• lasco-macos
• lasco-windows.exe

---

📦 Dependencies

The LASCO binary depends on external ASP grounders and solvers for full functionality:

• gringo is required to solve tasks involving variables (i.e., non-ground programs).
• clingo or dlv must be installed if you want to solve the compiled encoding directly using the --solve option.

Note: Recommended versions of these tools are as follows. gringo 5.7.1; clingo 5.7.1; dlv 2.1.2

Note: If no solver is specified, LASCO will only produce the ASP encoding without solving it.

---

🔨 Build from source

To build LASCO from source, you need:

• GHC
• Cabal
• alex
• happy

1. Clone the repository:

git clone https://github.com/robyBorelli/lasco.git
cd lasco

2. Build using Cabal:

cd code
cabal build

3. The binary will be located inside dist-newstyle/build/.../lasco.

4. Test the compiled binary:

cabal run lasco -- --help

---

⚙️ Command-line options

Usage: lasco [options]

Available options:
  -i,--input IFILE              Specifies the input file containing the learning task
                                to solve.
  -o,--output OFILE             Specifies the output file which will contain the
                                output encoding.
  -e,--encoder ENCODER          Specifies the encoding type:
                                  - 'exponential'
                                  - 'disjunctive' (default)
  --solve SOLVER                Solves the encoding using:
                                  - 'clingo'
                                  - 'dlv'
  --solve-mode MODE             Solving mode:
                                  - 'first', 'all', or 'optimum'
                                  - an integer (e.g. 3) for number of solutions
  -t, --parallel-mode THREADS   Specifies the number of threads to use with the SOLVER. 
                                Works only with 'clingo' SOLVER.  (default: 1)
  -v,--verbose                  Prints intermediate steps and debug information.
  --enable-comments             Enables comments in the generated encoding.
  --show-hypos [i1,i2,...]      Prints hypotheses of specified indexes and exits.
  --version                     Prints program version.
  -h,--help                     Show this help text.

Examples

# Basic usage
lasco --input examples/ex1_normal_rules.las

# Solves at optimum using clingo
lasco --input examples/ex1_normal_rules.las --solve clingo --solve-mode optimum

# Shows the first two rules in the hypothesis space
lasco --input examples/ex1_normal_rules.las --show-hypos [1,2]

# Basic usage with exponential encoder
lasco --input examples/ex1_normal_rules.las --encoder exponential

---

📥 Input Syntax

Input files must use Answer Set Programming (ASP) syntax with some extensions to define the hypothesis and examples.

1. Background Knowledge

The background is a set of standard ASP rules, including:

• Facts (e.g., edge(a,b).)
• Normal rules (e.g., reachable(X,Y) :- edge(X,Y).)
• Constraints (e.g., :- not reachable(a,c).)
• Arithmetic (e.g., X = Y + 1.)
• Comparison predicates (e.g., X < Y.)

Syntax rules:

• Variables start with an uppercase letter: X, Node, Time.
• Constants start with a lowercase letter: a, node1, zero.
• Every word containing LASCO or lasco as a substring, is a reserved word.

2. Hypotheses

A rule in the hypothesis space is defined as:

weight ~ rule.

where:

• weight is a positive integer.
• rule is an ASP rule.

---

3. Examples

A standard example is defined as:

#type(inclusion_set, exclusion_set).

where:

• type is either pos (positive example) or neg (negative example).
• inclusion_set is a set of ground atoms. (e.g., {p, q, reachable(a), edge(a,b)})
• exclusion_set is a set of ground atoms.

A context example is defined as:

#type(inclusion_set, exclusion_set, {context}).

where:

• context is an ASP program.

4. Full example

p :- not q.

1 ~ q.
2 ~ q :- not p.

#pos({p},{q}).
#neg({q},{p}).

---