Using MOSES to evolve programs

This Jupyter notebook shows how MOSES can be used to evolve programs and mevis can be used to visualize the results. The .ipynb file can be found here.

Sources:

• OpenCog Wiki

  • Python console: https://wiki.opencog.org/w/Python

  • MOSES: Meta-Optimizing Semantic Evolutionary Search

• GitHub

  • asmoses with examples/asmoses.py

from opencog.asmoses.pyasmoses import moses

import mevis as mv

moses = moses()

Define training data

A list of input-output pairs

• Input: first two columns

• Output: last column

input_data = [
    [0, 0,  0],
    [1, 1,  0],
    [1, 0,  1],
    [0, 1,  1],
]

print('Training data: {}'.format(input_data))

Training data: [[0, 0, 0], [1, 1, 0], [1, 0, 1], [0, 1, 1]]

Evolve Python programs

output = moses.run(input=input_data, python=True)

best_result = output[0]
program_text = best_result.program.decode()

print('Score:', best_result.score)
print('\nPython program:\n', program_text)

Score: 0.0

Python program:
 #!/usr/bin/env python

#score: 0
import operator as op
from functools import reduce
from math import log, exp, sin
def l0(i): return 0 < i
def adds(*args): return sum(args)
def muls(*args): return reduce(op.mul, args)
def pdiv(a, b): return a / (b + 0.000001)
def impulse(a): return 1.0 if a else 0.0
def moses_eval(i):
    return ((not(i[0]) or not(i[1])) and (i[0] or i[1]))

model = output[0].eval

print('Testing model on data:')
print('[0, 0]: {}'.format(model([0, 0])))
print('[0, 1]: {}'.format(model([0, 1])))
print('[1, 0]: {}'.format(model([1, 0])))
print('[1, 1]: {}'.format(model([1, 1])))

Testing model on data:
[0, 0]: 0
[0, 1]: 1
[1, 0]: 1
[1, 1]: False

Evolve Scheme programs

output = moses.run(input=input_data, scheme=True)

best_result = output[0]
program_text = best_result.program.decode()

print('Score:', best_result.score)
print('\nScheme program:\n', program_text)

Score: 0

Scheme program:
 (AndLink (OrLink (NotLink (PredicateNode "$1")) (NotLink (PredicateNode "$2"))) (OrLink (PredicateNode "$1") (PredicateNode "$2")))

Evolve a complex Scheme program and put it into an AtomSpace

from opencog.atomspace import AtomSpace
from opencog.scheme import scheme_eval
from opencog.utilities import set_default_atomspace

input_data = [
    [0, 0, 0, 0, 0,  1],
    [0, 0, 0, 0, 1,  0],
    [0, 0, 0, 1, 0,  0],
    [0, 0, 0, 1, 1,  0],
    [0, 0, 1, 0, 0,  1],
    [0, 0, 1, 0, 1,  0],
    [0, 0, 1, 1, 0,  1],
    [0, 0, 1, 1, 1,  1],
    [0, 1, 0, 0, 0,  0],
    [0, 1, 0, 0, 1,  1],
    [0, 1, 0, 1, 0,  0],
    [0, 1, 0, 1, 1,  1],
    [0, 1, 1, 0, 0,  1],
    [0, 1, 1, 0, 1,  1],
    [0, 1, 1, 1, 0,  0],
    [0, 1, 1, 1, 1,  0],
    [1, 0, 0, 0, 0,  1],
    [1, 0, 0, 0, 1,  1],
    [1, 0, 0, 1, 0,  0],
    [1, 0, 0, 1, 1,  1],
    [1, 0, 1, 0, 0,  0],
    [1, 0, 1, 0, 1,  1],
    [1, 0, 1, 1, 0,  0],
    [1, 0, 1, 1, 1,  0],
    [1, 1, 0, 0, 0,  1],
    [1, 1, 0, 0, 1,  1],
    [1, 1, 0, 1, 0,  0],
    [1, 1, 0, 1, 1,  1],
    [1, 1, 1, 0, 0,  0],
    [1, 1, 1, 0, 1,  1],
    [1, 1, 1, 1, 0,  1],
    [1, 1, 1, 1, 1,  0],
]

output = moses.run(input=input_data, scheme=True)
best_result = output[0]

score = best_result.score
program_text = best_result.program.decode()

print('Score:', best_result.score)
print('\nPython program:\n', program_text)

Score: -8

Python program:
 (OrLink (AndLink (OrLink (AndLink (NotLink (PredicateNode "$3")) (PredicateNode "$4")) (NotLink (PredicateNode "$2"))) (PredicateNode "$5")) (AndLink (OrLink (PredicateNode "$1") (NotLink (PredicateNode "$5"))) (PredicateNode "$2") (PredicateNode "$3")) (AndLink (PredicateNode "$2") (NotLink (PredicateNode "$4")) (NotLink (PredicateNode "$5"))))

atomspace = AtomSpace()
set_default_atomspace(atomspace)

string = scheme_eval(atomspace, program_text)
print(string.decode())

(OrLink
  (AndLink
    (PredicateNode "$3")
    (PredicateNode "$2")
    (OrLink
      (PredicateNode "$1")
      (NotLink
        (PredicateNode "$5"))))
  (AndLink
    (PredicateNode "$2")
    (NotLink
      (PredicateNode "$4"))
    (NotLink
      (PredicateNode "$5")))
  (AndLink
    (PredicateNode "$5")
    (OrLink
      (AndLink
        (PredicateNode "$4")
        (NotLink
          (PredicateNode "$3")))
      (NotLink
        (PredicateNode "$2")))))

mv.plot(atomspace, 'vis', 'dot')

Details for selected element

General

App state

Reset

Display mode

Enter full screen

Export

SVG PNG JPG

Data selection

Graph

Node label text

Edge label text

Node size

Normalize

Minimum

Maximum

Edge size

Normalize

Minimum

Maximum

Nodes

Visibility

Show nodes

Size

Scaling factor

Position

Release fixed nodes

Drag behavior

Fix node position

Hover behavior

Show neighborhood

Show tooltips (if provided)

Node images

Visibility

Show node images

Size

Scaling factor

Node labels

Visibility

Show node labels

Show borders

Size

Scaling factor

Rotation

Angle

Edges

Visibility

Show edges

Size

Scaling factor

Form

Curvature

Hover behavior

Show tooltips (if provided)

Edge labels

Visibility

Show edge labels

Show borders

Size

Scaling factor

Rotation

Angle

Layout algorithm

Simulation

Active

Algorithm

Parameters

Gravitational constant

Spring length

Spring constant

Avoid overlap

Central gravity

mv.store(atomspace, 'moses_result.scm', overwrite=True)

graph = mv.convert(atomspace)
mv.export(graph, 'moses_result.gml', overwrite=True)