JijModeling 1.12.1 Release Notes

Feature Enhancements

Human Readable Errors in Interpreters

Now Interpreter fails with more human-readable error message. For example:

import jijmodeling as jm

D = jm.Placeholder("D", dtype=jm.DataType.INTEGER)
T = jm.Placeholder("T", dtype=jm.DataType.INTEGER)
N = jm.Placeholder("N", dtype=jm.DataType.INTEGER)
H = jm.Placeholder("H", shape=(D, T, N))
R = jm.Placeholder("R", shape=(D, N))

x = jm.BinaryVar("x", shape=(D, T, N))
d = jm.Element("d", belong_to=(0, D))
t = jm.Element("t", belong_to=(0, T))
i = jm.Element("i", belong_to=(0, N))

problem = jm.Problem(name="shift_optimization")
problem += jm.Constraint("night_shift", R[d+1, i] <= x[d, T-1, i], forall=[d, i])
problem += jm.sum([d, t, i], x[d, t, i])

problem

\[\begin{split}\begin{array}{cccc}\text{Problem:} & \text{shift\_optimization} & & \\& & \min \quad \displaystyle \sum_{d = 0}^{D - 1} \sum_{t = 0}^{T - 1} \sum_{i = 0}^{N - 1} x_{d, t, i} & \\\text{{s.t.}} & & & \\ & \text{night\_shift} & \displaystyle R_{d + 1, i} \leq x_{d, T - 1, i} & \forall d \in \left\{0,\ldots,D - 1\right\} \forall i \in \left\{0,\ldots,N - 1\right\} \\\text{{where}} & & & \\& x & 3\text{-dim binary variable}\\\end{array}\end{split}\]

num_days = 7
num_times = 3
num_people = 5

data = problem.generate_random_dataset(
    options={
        "D": {"value": num_days},
        "T": {"value": num_times},
        
        "N": {"value": num_people},
    }
)
interp = jm.Interpreter(data)

try:
    interp.eval_problem(problem)
except Exception as e:
    print(e)

Traceback (most recent last):
    while evaluating Problem `shift_optimization',
    while evaluating constraint: `night_shift',
    while evaluating expression `R[d + 1, i] - x[d, T - 1, i]',
    while evaluating expression `R[d + 1, i]',

IndexError: Index [7, 0] is out of range for shape [7, 5]

Bugfixes

Fixes the Performance Bug in Interpreter.eval_*.

Previously, evaluating expressions with Interpreter takes much more times than intended. This was due to the direct call to the arithmetic operations on OMMX Functions, which goes through conversion between intermediate expressions and re-allocates memory from the ground up. This results in quadratic performance degradation in some summation operations.

Now JijModeling uses dedicated intermediate form of functions, so the overhead of conversion should be greatly tamed now.