def max_synergistic_nudge(old_X: dit.Distribution,
YgivenX: np.ndarray,
eps: float = 0.01):
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
rvs = old_X.get_rv_names()
outcomes = old_X.outcomes
if len(rvs) < 3:
return max_global_nudge(old_X, YgivenX, eps)
nudge, _ = max_nudge(old_X.copy('linear'),
YgivenX,
eps=eps,
nudge_type='synergistic_old')
# print("synergistic eps",sum(abs(nudge)), eps, old_X.outcome_length())
if base == 'linear':
perform_nudge(new_X, nudge)
else:
log_nudge, sign = np.log(np.abs(nudge)), np.sign(nudge)
perform_log_nudge(new_X, log_nudge, sign)
dct = {o: new_X[o] if o in new_X.outcomes else 0.0 for o in outcomes}
#print(outcomes, dct)
new_X = dit.Distribution(dct)
new_X.set_rv_names(rvs)
return new_X
def max_global_nudge(old_X: dit.Distribution,
YgivenX: np.ndarray,
eps: float = 0.01):
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
rvs = old_X.get_rv_names()
outcomes = old_X.outcomes
nudge, _ = max_nudge(old_X.copy('linear'),
YgivenX,
eps=eps,
nudge_type='global')
# print("global eps",sum(abs(nudge)), eps, old_X.outcome_length())
if base == 'linear':
perform_nudge(new_X, nudge)
else:
# print(nudge)
log_nudge, sign = np.log(np.abs(nudge)), np.sign(nudge)
# print(log_nudge, sign)
# log_nudge[log_nudge == -np.inf] = 0
# print("converted to log nudge",nudge, log_nudge, sign)
perform_log_nudge(new_X, log_nudge, sign)
dct = {o: new_X[o] if o in new_X.outcomes else 0.0 for o in outcomes}
#print(outcomes, dct)
new_X = dit.Distribution(dct)
new_X.set_rv_names(rvs)
return new_X
def local_nudge2(old_X: dit.Distribution,
eps: float = 0.01) -> dit.Distribution:
mask = old_X._mask
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
rvs = list(old_X.get_rv_names())
random.shuffle(rvs)
new_Xs = np.zeros((len(rvs), len(old_X)))
for i in range(len(rvs)):
rvs_other = rvs[:i] + rvs[i + 1:]
#print(new_X.get_rv_names())
new_X = individual_nudge(new_X, eps / len(rvs), rvs_other=rvs_other)
return new_X
def global_nudge(old_X: dit.Distribution,
eps: float = 0.01) -> dit.Distribution:
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
outcomes = old_X.outcomes
nudge_size = len(old_X)
rvs = old_X.get_rv_names()
if base == 'linear':
nudge = generate_nudge(nudge_size, eps)
perform_nudge(new_X, nudge)
else:
nudge, sign = generate_log_nudge(nudge_size, eps)
perform_log_nudge(new_X, nudge, sign)
new_X = dit.Distribution(
{o: new_X[o] if o in new_X.outcomes else 0
for o in outcomes})
new_X.set_rv_names(rvs)
return new_X
def max_local_nudge2(old_X: dit.Distribution,
YgivenX: np.ndarray,
eps: float = 0.01):
if old_X.outcome_length() == 1:
return max_global_nudge(old_X, YgivenX, eps)
mask = old_X._mask
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
rvs = old_X.get_rv_names()
sorted_rvs = np.argsort([
entropy(old_X.marginal([rv], rv_mode='indices').pmf)
for rv in range(len(rvs))
])
oldshape = len(old_X)
outcomes = old_X.outcomes
# print("before", new_X.pmf.shape)
for i, rv in enumerate(sorted_rvs):
nudges, _ = max_nudge(new_X.copy('linear'),
YgivenX,
eps=(eps / len(sorted_rvs)),
nudge_type='individual',
minimal_entropy_idx=rv)
# print("local eps",sum([sum(abs(nudge)) for nudge in nudges]), eps, old_X.outcome_length())
new_X = do_max_individual_nudge(new_X, nudges, rv, True)
# print("after {}".format(i), new_X.pmf.shape)
new_X.make_dense()
newshape = len(new_X)
# if oldshape != newshape:
# print(nudges)
# print("after {} and making dense".format(i), new_X.pmf.shape)
dct = {o: new_X[o] if o in new_X.outcomes else 0.0 for o in outcomes}
#print(outcomes, dct)
new_X = dit.Distribution(dct)
new_X.set_rv_names(rvs)
new_X._mask = mask
return new_X
def max_local_nudge1(old_X: dit.Distribution,
YgivenX: np.ndarray,
eps: float = 0.01):
if old_X.outcome_length() == 1:
return max_global_nudge(old_X, YgivenX, eps)
mask = old_X._mask
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
outcomes = old_X.outcomes
rvs = old_X.get_rv_names()
individual_nudges, _ = max_nudge(old_X.copy('linear'),
YgivenX,
eps=eps,
nudge_type='local')
new_Xs = np.zeros((old_X.outcome_length(), len(old_X)))
for i, nudges in enumerate(individual_nudges):
tmp = do_max_individual_nudge(old_X, nudges, i)
# print(i, tmp.pmf, old_X.pmf, new_Xs)
new_Xs[i, :] = tmp.pmf - old_X.pmf
nudge = new_Xs.sum(axis=0)
nudge = eps * nudge / (abs(nudge).sum())
if base == 'linear':
perform_nudge(new_X, nudge)
else:
log_nudge, sign = np.log(np.abs(nudge)), np.sign(nudge)
perform_log_nudge(new_X, log_nudge, sign)
dct = {o: new_X[o] if o in new_X.outcomes else 0.0 for o in outcomes}
#print(outcomes, dct)
new_X = dit.Distribution(dct)
new_X.set_rv_names(rvs)
new_X._mask = mask
return new_X
def local_nudge1(old_X: dit.Distribution,
eps: float = 0.01) -> dit.Distribution:
mask = old_X._mask
base = old_X.get_base()
new_X = old_X.copy(base=base)
old_X.make_dense()
outcomes = old_X.outcomes
rvs = list(old_X.get_rv_names())
random.shuffle(rvs)
#print(rvs)
new_Xs = np.zeros((len(rvs), len(old_X)))
for i in range(len(rvs)):
rvs_other = rvs[:i] + rvs[i + 1:]
tmp = individual_nudge(old_X, eps, rvs_other=rvs_other)
#print("tmp",tmp)
tmp.make_dense()
old_X.make_dense()
if base == 'linear':
new_Xs[i, :] = tmp.pmf - old_X.pmf
else:
new_Xs[i] = tmp.copy(base='linear').pmf - old_X.copy(
base='linear').pmf
#old_X.make_sparse()
nudge = new_Xs.sum(axis=0)
nudge = eps * nudge / (abs(nudge).sum())
if base == 'linear':
perform_nudge(new_X, nudge)
else:
perform_log_nudge(new_X, np.log(np.abs(nudge)), np.sign(nudge))
new_X = dit.Distribution(
{o: new_X[o] if o in new_X.outcomes else 0
for o in outcomes})
new_X.set_rv_names(rvs)
new_X._mask = mask
return new_X
