def test_solver_minmax_respects_two_minimum():
'''
Tests 3 papers, 4 reviewers. Reviewers review min: 2, max: 3 papers. Each paper needs 3 reviews.
Reviewer 4 has very high cost. Other reviewers have 0 cost.
Purpose: Make sure all reviewers (including reviewer 4) get at least their minimum
'''
num_papers = 3
num_reviewers = 4
min_papers_per_reviewer = 2
max_papers_per_reviewer = 3
paper_revs_reqd = 3
cost_matrix = np.transpose(
np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [2000, 2000, 2000]]))
constraint_matrix = np.transpose(
np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]]))
rev_mins = [min_papers_per_reviewer] * num_reviewers
rev_maxs = [max_papers_per_reviewer] * num_reviewers
papers_reqd = [paper_revs_reqd] * num_papers
solver = MinMaxSolver(rev_mins,
rev_maxs,
papers_reqd,
encoder(cost_matrix, constraint_matrix),
allow_zero_score_assignments=True)
res = solver.solve()
assert res.shape == (3, 4)
# make sure every reviewer has at least 1 paper
nrows, ncols = res.shape
for rix in range(nrows):
reviewer_count_reviews = 0
for pix in range(ncols):
if res[rix, pix] != 0:
reviewer_count_reviews += 1
assert reviewer_count_reviews >= 1
def test_solver_finds_lowest_cost_soln():
'''
4 reviewers 3 papers. Papers 0,1 need 1 review; Paper 2 needs 2 reviews. Reviewers can do max of 2 reviews
Setup so that lowest cost solution should be
Reviewer 0 reviews paper 0
1 reviews paper 1
2 reviews paper 2
3 reviews paper 2
Purpose: Finds the lowest cost solution
'''
cost_matrix = np.transpose(np.array([
[0, 1, 1],
[1, 0, 1],
[1, 1, 0],
[2, 2, 0]
]))
constraint_matrix = np.zeros(np.shape(cost_matrix))
solver = MinMaxSolver(
[1,1,1,1],
[2,2,2,2],
[1,1,2],
encoder(cost_matrix, constraint_matrix)
)
res = solver.solve()
assert res.shape == (3,4)
expected_cost = 0
check_solution(solver, expected_cost)
def test_solver_minmax_avoid_zero_scores_get_no_solution():
'''
Tests 3 papers, 4 reviewers.
Reviewers review min: 2, max: 3 papers.
Each paper needs 3 reviews.
Reviewer 4 has very high cost.
Other reviewers have 0 cost.
Purpose: Make sure all reviewers (including reviewer 4) get at least their minimum
'''
num_papers = 3
num_reviewers = 4
min_papers_per_reviewer = 2
max_papers_per_reviewer = 3
paper_revs_reqd = 3
aggregate_score_matrix = np.transpose(
np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 1, 0]]))
constraint_matrix = np.transpose(
np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]]))
rev_mins = [min_papers_per_reviewer] * num_reviewers
rev_maxs = [max_papers_per_reviewer] * num_reviewers
papers_reqd = [paper_revs_reqd] * num_papers
solver = MinMaxSolver(rev_mins,
rev_maxs,
papers_reqd,
encoder(aggregate_score_matrix, constraint_matrix),
allow_zero_score_assignments=False)
res = solver.solve()
assert solver.solved == False
def test_solver4_minmax():
'''
Tests 6 papers, 6 reviewers. Reviewers review min: 2, max: 3 papers. Each paper needs 2 reviews.
All scores set to 1 so that any match that does not violate constraints is optimal
Purpose: Honors minimums == 2 for all reviewers
'''
cost_matrix = np.array([[1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1]])
constraint_matrix = np.array([[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0]])
solver = MinMaxSolver([2, 2, 2, 2, 2, 2], [3, 3, 3, 3, 3, 3],
[2, 2, 2, 2, 2, 2],
encoder(cost_matrix, constraint_matrix))
res = solver.solve()
assert res.shape == (6, 6)
# make sure every reviewer is reviewing 2 papers
nrows, ncols = res.shape
for rix in range(nrows):
reviewer_count_reviews = 0
for pix in range(ncols):
if res[rix, pix] != 0:
reviewer_count_reviews += 1
assert reviewer_count_reviews == 2
check_solution(solver, solver.optimal_cost)
def test_solver_minmax_find_lowest_cost_and_respect_constraints():
'''
Tests 5 papers, 4 reviewers. Reviewers review min: 1, max: 3 papers. Each paper needs 2 reviews.
Constrained such that:
Reviewer 0: available for all papers
1: cannot review papers 0,3
2: cannot review papers 3,4
3: cannot review papers 1,2
Scores set such that a lowest-cost solution can be found along all reviewer-paper arcs with cost = -10 and no others.
Purpose: Finds the lowest cost solution in combination with honoring constraints (i.e. ignores lower-cost paths that are constrained to be ommitted)
'''
cost_matrix = np.transpose(
np.array([[-10, 1, 1, -10, -10], [-100, -10, -10, -100, 1],
[1, -10, -10, -100, -100], [-10, -100, -100, -10, -10]]))
constraint_matrix = np.transpose(
np.array([[0, 0, 0, 0, 0], [-1, 0, 0, -1, 0], [0, 0, 0, -1, -1],
[0, -1, -1, 0, 0]]))
solver = MinMaxSolver([1, 1, 1, 1], [3, 3, 3, 3], [2, 2, 2, 2, 2],
encoder(cost_matrix, constraint_matrix))
res = solver.solve()
assert res.shape == (5, 4)
# make sure result does not violate constraints (i.e. no flow at i,j if there is a -1 constraint at i,j
# make sure the score at i,j = -10 if there is flow there.
nrows, ncols = res.shape
for i in range(nrows):
for j in range(ncols):
assert not (constraint_matrix[i, j] == -1 and res[i, j] > 0
), "Solution violates constraint at [{},{}]".format(
i, j)
assert not (
res[i, j] > 0 and cost_matrix[i, j] > -10
), "Solution contains an arc that is not part of an lowest-cost solution"
check_solution(solver, solver.optimal_cost)
def test_solver_minmax_respects_constraints():
'''
Tests 5 papers, 4 reviewers. Reviewers review min: 1, max: 3 papers. Each paper needs 2 reviews.
Constrained such that:
Reviewer 0: available for all papers
1: cannot review papers 2,3
2: cannot review papers 2,3
3: cannot review papers 0, 1
All scores set to 1 so that any match that does not violate constraints is optimal
Purpose: Honors constraints in its solution
'''
cost_matrix = np.transpose(
np.array([[1, 1, 1, 1, 1], [1, 1, 1, 1, 1], [1, 1, 1, 1, 1],
[1, 1, 1, 1, 1]]))
constraint_matrix = np.transpose(
np.array([[0, 0, 0, 0, 0], [0, 0, -1, -1, 0], [0, 0, -1, -1, 0],
[-1, -1, 0, 0, 0]]))
solver = MinMaxSolver([1, 1, 1, 1], [3, 3, 3, 3], [2, 2, 2, 2, 2],
encoder(cost_matrix, constraint_matrix))
res = solver.solve()
assert res.shape == (5, 4)
# make sure result does not violate constraints (i.e. no flow at i,j if there is a -1 constraint at i,j
nrows, ncols = res.shape
for i in range(nrows):
for j in range(ncols):
assert not (constraint_matrix[i, j] == -1 and res[i, j] > 0
), "Solution violates constraint at [{},{}]".format(
i, j)
check_solution(solver, solver.optimal_cost)
def test_solver_minmax_custom_supply():
aggregate_score_matrix_A = np.transpose(
np.array([[0.2, 0.1, 0.4], [0.5, 0.2, 0.3], [0.2, 0.0, 0.6],
[0.7, 0.9, 0.3]]))
constraint_matrix = np.zeros(np.shape(aggregate_score_matrix_A))
solver_A = MinMaxSolver([1, 1, 1, 1], [2, 1, 3, 1], [2, 2, 2],
encoder(aggregate_score_matrix_A,
constraint_matrix))
res_A = solver_A.solve()
assert res_A.shape == (3, 4)
def test_solver_minmax_custom_demands_paper_with_0_demand():
aggregate_score_matrix_A = np.transpose(
np.array([[0.2, 0.1, 0.4], [0.5, 0.2, 0.3], [0.2, 0.0, 0.6],
[0.7, 0.9, 0.3]]))
constraint_matrix = np.zeros(np.shape(aggregate_score_matrix_A))
solver_A = MinMaxSolver([0, 0, 0, 0], [2, 2, 2, 2], [2, 1, 0],
encoder(aggregate_score_matrix_A,
constraint_matrix))
res_A = solver_A.solve()
assert res_A.shape == (3, 4)
assert np.sum(res_A, axis=1)[2] == 0
def test_solver5():
'''
Tests 3 papers, 4 reviewers. Reviewers review min: 1, max: 3 papers. Each paper needs 3 reviews.
Reviewer 4 has very high cost. Other reviewers have 0 cost.
Purpose: Make sure all reviewers get at least their minimum
'''
num_papers = 3
num_reviewers = 4
min_papers_per_reviewer = 1
max_papers_per_reviewer = 3
paper_revs_reqd = 3
cost_matrix = np.transpose(np.array([
[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
[2000, 2000, 2000]]))
constraint_matrix = np.transpose(np.array([
[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
[0, 0, 0]]))
rev_mins = [min_papers_per_reviewer] * num_reviewers
rev_maxs = [max_papers_per_reviewer] * num_reviewers
papers_reqd = [paper_revs_reqd] * num_papers
solver = MinMaxSolver(
rev_mins,
rev_maxs,
papers_reqd,
encoder(cost_matrix, constraint_matrix)
)
res = solver.solve()
assert res.shape == (3, 4)
# make sure every reviewer has at least 1 paper
nrows, ncols = res.shape
for rix in range(nrows):
reviewer_count_reviews = 0
for pix in range(ncols):
if res[rix,pix] != 0:
reviewer_count_reviews += 1
assert reviewer_count_reviews >= 1
# TestSolver.silent = False
check_solution(solver,solver.optimal_cost)
def test_solver_minmax_impossible_constraints():
'''
Test to ensure that the MinMaxSolver's 'solved' attribute is correctly set
when no solution is possible due to constraints.
'''
# 20 papers, 5 reviewers
num_papers = 20
num_reviewers = 5
cost_matrix = np.zeros((num_papers, num_reviewers))
constraint_matrix = -1 * np.ones(
(num_papers,
num_reviewers)) # all pairs are constrained! should be impossible
minimums = [5] * 5
maximums = [20] * 5
demands = [3] * 20
solver = MinMaxSolver(minimums, maximums, demands,
encoder(cost_matrix, constraint_matrix))
solver.solve()
assert not solver.solved
def test_solver_minmax_random():
'''When costs are all zero, compute random assignments'''
cost_matrix_A = np.transpose(
np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]]))
constraint_matrix = np.zeros(np.shape(cost_matrix_A))
solver_A = MinMaxSolver([1, 1, 1, 1], [2, 2, 2, 2], [1, 1, 2],
encoder(cost_matrix_A, constraint_matrix))
res_A = solver_A.solve()
assert res_A.shape == (3, 4)
cost_matrix_B = np.transpose(
np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]]))
constraint_matrix = np.zeros(np.shape(cost_matrix_B))
solver_B = MinMaxSolver([1, 1, 1, 1], [2, 2, 2, 2], [1, 1, 2],
encoder(cost_matrix_B, constraint_matrix))
res_B = solver_B.solve()
assert res_B.shape == (3, 4)
# ensure that the cost matrices are random
# (i.e. overwhelmingly likely to be different)
assert not np.array_equal(solver_A.cost_matrix, solver_B.cost_matrix)
