% causal_model_mechanisms.pl % % This file contains mechanisms to deal with the details % of causal models, i.e., whether the different causal % relations actually have an effect or not. % % Written by Anders Bouwer, 8 Jan 2002 % % Last updated 30 June 2003 % effect_status(N, CausalParRel, Direction, Status) % % CausalParRel = causal_relation(X, Rel, Y) % this predicate finds out whether the CausalParRel % in state N actually has an effect or not (Status), % and in which direction the affected parameter would % change if this was the only effect. % % X has no effect on Y because X is zero or unknown (in the case of % X influencing Y), or X is steady or dX is unknown (in the case of % a proportional relationship). % % this clause is put first, because it takes the least amount % of time to check % effect_status(N, causal_relation(X, Rel, Y), none, none):- show_causal_model(N, Rel, X, Y, second), expected_effect(N, causal_relation(X, Rel, Y), none). % X has an effect on Y (as expected) % effect_status(N, causal_relation(X, Rel, Y), Direction, effect):- show_causal_model(N, Rel, X, Y, second), expected_effect(N, causal_relation(X, Rel, Y), Direction), Direction \== none, actual_effect(N, Y, Direction). % X has a submissive effect on Y (Y changes in the other direction) % effect_status(N, causal_relation(X, Rel, Y), Direction, submissive):- show_causal_model(N, Rel, X, Y, second), expected_effect(N, causal_relation(X, Rel, Y), Direction), actual_effect(N, Y, Direction2), reverse_direction(Direction, Direction2). % X has an effect on Y but this is counterbalanced % and consequently Y does not change at all % effect_status(N, causal_relation(X, Rel, Y), Direction, balanced):- show_causal_model(N, Rel, X, Y, second), expected_effect(N, causal_relation(X, Rel, Y), Direction), Direction \== none, actual_effect(N, Y, none). % correspondence relationships % N.B. Arguments of correspondences are switched % with respect to the GARP representation! % % effect status of correspondence relationships % A correspondence relation (of any kind) is active when % it can be used in a calculation % % This version uses the same visualization format as % gen_causal_rel etc. % corr_effect_status(N, corr_relation(N, VFrom, VRel, VTo, Dir), active):- index_pred_state(N, par_relations, CorrRel), ( % q_corr CorrRel = q_correspondence(Q1, Q2), contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1)), visualize(Rel, Q1, Q2, VRel, VFrom, VTo, Dir) ; % dir_q_corr CorrRel = dir_q_correspondence(Q1, Q2), contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1)), visualize(Rel, Q1, Q2, VRel, VFrom, VTo, Dir) ; % v_corr CorrRel = v_correspondence(Q1, V1, Q2, V2), contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1)), % strip off quantity name, if necessary strip_atomize(V1, VA1), strip_atomize(V2, VA2), From =.. [VA1, Q1], To =.. [VA2, Q2], visualize(Rel, From, To, VRel, VFrom, VTo, Dir) ; % dir_v_corr CorrRel = dir_v_correspondence(Q1, V1, Q2, V2), contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1)), % strip off quantity name, if necessary strip_atomize(V1, VA1), strip_atomize(V2, VA2), From =.. [VA1, Q1], To =.. [VA2, Q2], visualize(Rel, From, To, VRel, VFrom, VTo, Dir) ). % A correspondence relation (of any kind) is inactive when % it is not active % corr_effect_status(N, corr_relation(N, VFrom, VRel, VTo, Dir), inactive):- index_pred_state(N, par_relations, CorrRel), ( % q_corr CorrRel = q_correspondence(Q1, Q2), not(contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1))), visualize(q_correspondence, Q1, Q2, VRel, VFrom, VTo, Dir) ; % dir_q_corr CorrRel = dir_q_correspondence(Q1, Q2), not(contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1))), visualize(dir_q_correspondence, Q1, Q2, VRel, VFrom, VTo, Dir) ; % v_corr CorrRel = v_correspondence(Q1, V1, Q2, V2), not(contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1))), % strip off quantity name, if necessary strip_atomize(V1, VA1), strip_atomize(V2, VA2), From =.. [VA1, Q1], To =.. [VA2, Q2], visualize(v_correspondence, From, To, VRel, VFrom, VTo, Dir) ; % dir_v_corr CorrRel = dir_v_correspondence(Q1, V1, Q2, V2), not(contains_corr_calculation(N, corr_calculation, corr_calc(N, Q2, V2, Rel, Q1, V1))), % strip off quantity name, if necessary strip_atomize(V1, VA1), strip_atomize(V2, VA2), From =.. [VA1, Q1], To =.. [VA2, Q2], visualize(dir_v_correspondence, From, To, VRel, VFrom, VTo, Dir) ). % A correspondence relation (of any kind) is active when % it can be used in a calculation % % This version has separate parameters for the values % corr_effect_status(N, corr_relation(N, Q1, V1, Rel, Q2, V2), active):- contains_corr_calculation(N, corr_calculation, corr_calc(N, Q1, V1, Rel, Q2, V2)). % A correspondence relation (of any kind) is inactive when % it is not active % corr_effect_status(N, corr_relation(N, Q1, V1, Rel, Q2, V2), inactive):- index_pred_state(N, par_relations, CorrRel), ( CorrRel = q_correspondence(Q2, Q1), Rel = q_correspondence ; CorrRel = q_correspondence(Q1, Q2), Rel = q_corr_rev ; CorrRel = dir_q_correspondence(Q2, Q1), Rel = dir_q_correspondence ; CorrRel = v_correspondence(Q2, V2, Q1, V1), Rel = v_correspondence ; CorrRel = v_correspondence(Q1, V1, Q2, V2), Rel = v_corr_rev ; CorrRel = dir_v_correspondence(Q2, V2, Q1, V1), Rel = dir_v_correspondence ), not(contains_corr_calculation(N, corr_calculation, corr_calc(N, Q1, V1, Rel, Q2, V2))). reverse_direction(pos, neg). reverse_direction(neg, pos). % Note that the different clauses of this predicate % partly give the same result, so duplicates may occur! % positive_value(N, X):- index_pred_state(N, par_relations, greater(X, zero)),!. % if the above par_relation is not present, we have to check % it by investigating the value in relation to its quantity space % positive_value(N, X):- index_pred_state(N, par_values, value(X, _, Value, _Der)),!, nonvar(Value), engine:qspace(X, _Predicate, QSpaceList, _Fail), qspace_member(Value, QSpaceList, Index), qspace_member(zero, QSpaceList, ZeroIndex), Index > ZeroIndex,!. % if the value is plus, we can also say it's positive, % also if zero is not included in the qspace % positive_value(N, X):- index_pred_state(N, par_values, value(X, _, QV, _Der)),!, QV == plus. % Note that the different clauses of this predicate % partly give the same result, so duplicates may occur! % negative_value(N, X):- index_pred_state(N, par_relations, smaller(X, zero)),!. % if the above par_relation is not present, we have to check % it by investigating the value in relation to its quantity space % negative_value(N, X):- index_pred_state(N, par_values, value(X, _, Value, _Der)),!, nonvar(Value), engine:qspace(X, _Predicate, QSpaceList, _Fail), qspace_member(Value, QSpaceList, Index), qspace_member(zero, QSpaceList, ZeroIndex), Index < ZeroIndex,!. % This succeeds when X is unknown in state N % unknown_value(N, X):- index_pred_state(N, par_values, value(X, _, Value, _Der)),!, var(Value). % This succeeds when the derivative of X is unknown in state N % unknown_derivative(N, X):- index_pred_state(N, par_values, value(X, _, _Value, Der)),!, var(Der). increasing(N, X):- index_pred_state(N, par_values, value(X, _, _Value, Der)),!, Der == plus ; index_pred_state(N, par_relations, d_greater(X, zero)). decreasing(N, X):- index_pred_state(N, par_values, value(X, _, _Value, Der)),!, Der == min ; % or index_pred_state(N, par_relations, d_smaller(X, zero)). zero_value(N, X):- index_pred_state(N, par_values, value(X, _, Val, _Der)),!, Val == zero ; % or index_pred_state(N, par_relations, equal(X, zero)). steady(N, X):- index_pred_state(N, par_values, value(X, _, _, Der)),!, Der == zero ; % or index_pred_state(N, par_relations, d_equal(X, zero)). % Value or derivative of X is expected to have a % positive effect on Y (causing Y to increase) % expected_effect(N, causal_relation(X, 'I+', _Y), pos):- positive_value(N, X),!. expected_effect(N, causal_relation(X, 'I-', _Y), pos):- negative_value(N, X),!. expected_effect(N, causal_relation(X, 'P+', _Y), pos):- increasing(N, X),!. expected_effect(N, causal_relation(X, 'P-', _Y), pos):- decreasing(N, X),!. % Value or derivative of X is expected to have a % negative effect on Y (causing Y to decrease) % expected_effect(N, causal_relation(X, 'I+', _Y), neg):- negative_value(N, X),!. expected_effect(N, causal_relation(X, 'I-', _Y), neg):- positive_value(N, X),!. expected_effect(N, causal_relation(X, 'P+', _Y), neg):- decreasing(N, X),!. expected_effect(N, causal_relation(X, 'P-', _Y), neg):- increasing(N, X),!. % Value or derivative of X is expected to have % no effect on Y % expected_effect(N, causal_relation(X, 'I+', _Y), none):- zero_value(N, X),!. expected_effect(N, causal_relation(X, 'I-', _Y), none):- zero_value(N, X),!. expected_effect(N, causal_relation(X, 'P+', _Y), none):- steady(N, X),!. expected_effect(N, causal_relation(X, 'P-', _Y), none):- steady(N, X),!. % Value or derivative of X is not expected to have % an effect on Y, because this value or derivative % is unknown. Strictly speaking, this is a slightly % different case than 'expected to have no effect', % but GARP seems to treat them the same way, i.e., % no effect is processed. Therefore, the same status % is given to these dependencies. % expected_effect(N, causal_relation(X, 'I+', _Y), none):- unknown_value(N, X),!. expected_effect(N, causal_relation(X, 'I-', _Y), none):- unknown_value(N, X),!. expected_effect(N, causal_relation(X, 'P+', _Y), none):- unknown_derivative(N, X),!. expected_effect(N, causal_relation(X, 'P-', _Y), none):- unknown_derivative(N, X),!. % rest case - make sure no backtracking can occur, % because this clause will always fire % % expected_effect(_N, CausalParRel, unknown):- % CausalParRel = causal_relation(_X, _Rel, _Y). % actual_effect(N, Y, Effect) % % This predicate finds out whether Y changes or not % (Effect = none) in state N, and if yes, in which direction % (Effect = pos or neg). % actual_effect(N, Y, pos):- increasing(N, Y),!. actual_effect(N, Y, neg):- decreasing(N, Y),!. actual_effect(N, Y, none):- steady(N, Y),!. % rest case - make sure no backtracking can occur, % because this clause will always fire % % actual_effect(_N, _Y, unknown). % causal_chain(StateNr, A, B, Chain, Length) % causal_chain(N, A, B, Chain, L):- path(A, B, causal_graph(N), Chain, L).