JijModeling 1.8.0 Release Note

JijModeling 1.8.0 Release Note#

import jijmodeling as jm
import ommx.v1
import pytest

Compiling into OMMX Message#

image.png

Interpreter.eval_scalar#

  • eval_scalar evaluates JijModeling expression which does not contain decision variables into a float

  • This method is useful for testing and debugging

# Initialize with instance data
interpreter = jm.Interpreter({"a": [1, 2, 3]})

# Placeholder is evaluated with instance data
a = jm.Placeholder("a", ndim=1)
assert interpreter.eval_scalar(a[1]) == 2.0

# Summation is supported
n = a.len_at(0, latex="n")
i = jm.Element("i", belong_to=n)
sum_a = jm.sum(i, a[i])
sum_a
\[\displaystyle \sum_{i = 0}^{n - 1} a_{i}\]
assert interpreter.eval_scalar(sum_a) == 1 + 2 + 3

# If it finds decision variable, raises error
v = jm.BinaryVar("v")
with pytest.raises(jm.InterpreterError):
    interpreter.eval_scalar(v)

Interpreter.eval_expr#

JijModeling expression containing decision variables are evaluated by eval_expr into ommx.v1.Function

# Decision variables
x = jm.BinaryVar("x", shape=(n,))

# Linear function
f = jm.sum(i, a[i] * x[i])
f
\[\displaystyle \sum_{i = 0}^{n - 1} a_{i} \cdot x_{i}\]
f_ = interpreter.eval_expr(f)
assert type(f_) is ommx.v1.Function
f_

Function(x0 + 2*x1 + 3*x2)

  • Decision variables are registered into the interpreter when it is evaluated.

  • The decision variables get their ID when they are registered. Interpreter issues the ID sequentially from 0.

    • Here x[0] has ID 0, x[1] has ID 1, and so on.

  • Note that __repr__ of ommx.v1.Function show the variables as x0, x1, …, by their IDs since it does not contain the information of the original variable names.

y = jm.BinaryVar("y")
interpreter.eval_expr(y)  # Shown by their ID `x3` instead of their name `y`

Function(x3)

# Get corresponding decision variables in OMMX
x0 = interpreter.get_decision_variable_by_name("x", [0])
x1 = interpreter.get_decision_variable_by_name("x", [1])
x2 = interpreter.get_decision_variable_by_name("x", [2])
assert type(x0) is ommx.v1.DecisionVariable

# Compare obtained ommx.v1.Function
assert f_.almost_equal(ommx.v1.Function(x0 + 2*x1 + 3 * x2))

Interpreter.eval_problem#

problem = jm.Problem("release-1.8.0")
problem += f
problem += jm.Constraint("quad", jm.sum(i, x[i] * x[i]) == 1)
problem
\[\begin{split}\begin{array}{cccc}\text{Problem:} & \text{release-1.8.0} & & \\& & \min \quad \displaystyle \sum_{i = 0}^{n - 1} a_{i} \cdot x_{i} & \\\text{{s.t.}} & & & \\ & \text{quad} & \displaystyle \sum_{i = 0}^{n - 1} x_{i} \cdot x_{i} = 1 & \\\text{{where}} & & & \\& x & 1\text{-dim binary variable}\\\end{array}\end{split}\]
instance = interpreter.eval_problem(problem)
assert type(instance) is ommx.v1.Instance

Limitations#

  • jijmodeling.CustomPenaltyTerm is not supported yet.

    • This will be supported in next version with OMMX update.