|
|
import stormpy from helpers.helper import get_example_path
import math
class TestMatrix: def test_matrix(self): model = stormpy.build_sparse_model_from_explicit(get_example_path("dtmc", "die.tra"), get_example_path("dtmc", "die.lab")) matrix = model.transition_matrix assert type(matrix) is stormpy.storage.SparseMatrix assert matrix.nr_rows == model.nr_states assert matrix.nr_columns == model.nr_states assert matrix.nr_entries == 20 assert matrix.nr_entries == model.nr_transitions for e in matrix: assert e.value() == 0.5 or e.value() == 0 or (e.value() == 1 and e.column > 6)
def test_backward_matrix(self): model = stormpy.build_sparse_model_from_explicit(get_example_path("dtmc", "die.tra"), get_example_path("dtmc", "die.lab")) matrix = model.backward_transition_matrix assert type(matrix) is stormpy.storage.SparseMatrix assert matrix.nr_rows == model.nr_states assert matrix.nr_columns == model.nr_states assert matrix.nr_entries == 20 assert matrix.nr_entries == model.nr_transitions for e in matrix: assert e.value() == 0.5 or e.value() == 0 or (e.value() == 1 and e.column > 6)
def test_change_matrix(self): model = stormpy.build_sparse_model_from_explicit(get_example_path("dtmc", "die.tra"), get_example_path("dtmc", "die.lab")) matrix = model.transition_matrix for e in matrix: assert e.value() == 0.5 or e.value() == 0 or e.value() == 1 i = 0 for e in matrix: e.set_value(i) i += 0.1 i = 0 for e in matrix: assert e.value() == i i += 0.1
def test_change_matrix_modelchecking(self): import stormpy.logic model = stormpy.build_sparse_model_from_explicit(get_example_path("dtmc", "die.tra"), get_example_path("dtmc", "die.lab")) matrix = model.transition_matrix # Check matrix for e in matrix: assert e.value() == 0.5 or e.value() == 0 or e.value() == 1 # First model checking formulas = stormpy.parse_properties("P=? [ F \"one\" ]") result = stormpy.model_checking(model, formulas[0]) resValue = result.at(model.initial_states[0]) assert math.isclose(resValue, 0.16666666666666663)
# Change probabilities i = 0 for e in matrix: if e.value() == 0.5: if i % 2 == 0: e.set_value(0.3) else: e.set_value(0.7) i += 1 for e in matrix: assert e.value() == 0.3 or e.value() == 0.7 or e.value() == 1 or e.value() == 0 # Second model checking result = stormpy.model_checking(model, formulas[0]) resValue = result.at(model.initial_states[0]) assert math.isclose(resValue, 0.06923076923076932)
# Change probabilities again for state in model.states: for action in state.actions: for transition in action.transitions: if transition.value() == 0.3: transition.set_value(0.8) elif transition.value() == 0.7: transition.set_value(0.2) # Third model checking result = stormpy.model_checking(model, formulas[0]) resValue = result.at(model.initial_states[0]) assert math.isclose(resValue, 0.3555555555555556)
def test_change_parametric_matrix_modelchecking(self): import stormpy.logic
program = stormpy.parse_prism_program(get_example_path("pdtmc", "brp16_2.pm")) formulas = stormpy.parse_properties_for_prism_program("P=? [ F s=5 ]", program) model = stormpy.build_parametric_model(program, formulas) initial_state = model.initial_states[0] assert initial_state == 0 matrix = model.transition_matrix # Check matrix one_pol = stormpy.RationalRF(1) one_pol = stormpy.FactorizedPolynomial(one_pol) one = stormpy.FactorizedRationalFunction(one_pol, one_pol) for e in matrix: assert e.value() == one or len(e.value().gather_variables()) > 0 # First model checking result = stormpy.model_checking(model, formulas[0]) ratFunc = result.at(initial_state) assert len(ratFunc.gather_variables()) > 0
# Change probabilities two_pol = stormpy.RationalRF(2) two_pol = stormpy.FactorizedPolynomial(two_pol) new_val = stormpy.FactorizedRationalFunction(one_pol, two_pol) for e in matrix: if len(e.value().gather_variables()) > 0: e.set_value(new_val) for e in matrix: assert e.value() == new_val or e.value() == one # Second model checking result = stormpy.model_checking(model, formulas[0]) ratFunc = result.at(initial_state) assert len(ratFunc.gather_variables()) == 0
def test_submatrix(self): model = stormpy.build_sparse_model_from_explicit(get_example_path("dtmc", "die.tra"), get_example_path("dtmc", "die.lab")) matrix = model.transition_matrix assert matrix.nr_rows == 13 assert matrix.nr_columns == 13 assert matrix.nr_entries == 20 assert matrix.nr_entries == model.nr_transitions for e in matrix: assert e.value() == 0.5 or e.value() == 0 or (e.value() == 1 and e.column > 6)
row_constraint = stormpy.BitVector(13, [0, 1, 3, 4, 7, 8, 9]) submatrix = matrix.submatrix(row_constraint, row_constraint) assert submatrix.nr_rows == 7 assert submatrix.nr_columns == 7 assert submatrix.nr_entries == 10 for e in submatrix: assert e.value() == 0.5 or e.value() == 0 or (e.value() == 1 and e.column > 3)
|