def check_features(allfeatures, smallset, star, classifier):
test_set = list(smallset.index)[-test_size:]
true_values = [vote(classifier(a)) for a in test_set]
pres_value = -1
res = dict()
for feature in allfeatures:
a = enable_feature(star, feature, smallset)
train_star = {
apn: fea
for apn, fea in a.items()
if apn in list(smallset.index)[:-test_size]
}
vn = create_voting_net(
gamma=0.7,
apns=list(train_star.keys()),
distance=lambda x, y: mjaccard(x, y, train_star),
classifier=classifier)
#TP, FP, TN, FN
#e0, e1, e3, e2 = eval_net(net=vn, test_set=test_set, distance=lambda x,y: mjaccard(x,y, star), classifier=nc)
predictions = predict(net=vn,
test_set=test_set,
distance=lambda x, y: mjaccard(x, y, a))
r1 = f1_score(y_true=true_values, y_pred=predictions)
if r1 > pres_value:
print(f"new value detected {r1} {pres_value}")
pres_value = r1
res[feature] = r1
return res
def exp(funcs, labels, test_size=10):
nets = dict()
gamma = 0
train, test = train_test_split(funcs, test_size=test_size, random_state=42)
test.to_csv('/tmp/test.csv')
classifier = lambda x: [[0, 1], [1, 0]][int(labels.loc[x]['malware_label'])
]
d = lambda x, y: distance(x, y, funcs)
intervals = 18
#for gamma in tqdm([0, 1, 2, 4, 8, 16, 32, 64, 72, 80, 88, 96, 104, 110, 128, 164, 180, 192]):
for gamma in tqdm([x * 1 / intervals for x in range(0, intervals + 1)]):
print(f"Current {gamma=}")
mv = make_and_merge(train, labels, gamma)
print("Creating reference voting netwrok")
start = time.time()
reference_voting = create_voting_net(gamma=gamma,
apns=train.index,
distance=d,
classifier=classifier)
end = time.time()
print(f"\tElapsed: {end-start}")
nets[gamma] = [dict(mv), dict(reference_voting)]
if gamma == 0:
gamma = 1
else:
gamma *= 2
print(f"Anchor points: {len(mv.keys())}")
if len(mv.keys()) == 1:
break
# save nets:
save_nets(nets=nets, name=f"{len(train)}-jaccard-votingnets")
def rr(x, gamma, classifier):
start = time.time()
net = create_voting_net(gamma=gamma,
apns=x.keys(),
distance=lambda x1, y1: distance(x1, y1, x),
classifier=classifier)
end = time.time()
print(f"Creating voting network...{gamma=} {len(x)} Elapsed: {end-start}")
return net
def run_euclid(train, test, distance,classifier):
res = dict()
nets = dict()
distance = lambda x,y: adf(x, y, afs)
for gamma in tqdm([0, 1, 2, 4, 8, 16, 32, 64, 128]):
mv = create_voting_net(gamma=gamma, apns=train, distance=distance, classifier=classifier)
false_negative, false_positives = evaluate_voting_net(apns=test, classifier=classifier, net=mv, distance=distance)
res[gamma] = [false_negative, false_positives]
nets[gamma] = mv.copy()
sizes = {g:len(n.keys()) for g, n in nets.items()}
perfs_di = pd.DataFrame.from_dict(res, orient='index', columns=['fp', 'fn'])
szs_di = pd.DataFrame.from_dict(sizes, orient='index', columns=['sizes'])
gj_euc = perfs_di.join(szs_di)
part_size = len(train)
gj_euc['compression'] = gj_euc.sizes/part_size
gj_euc.to_csv(f"res/euc{part_size}.csv")
def run_jaccard(train, test, distance, classifier):
res = dict()
nets = dict()
for gamma in tqdm([0, 0.1, 0.2, 0.4, 0.5, 0.8, 0.85, 0.9, 1.0]):
mv = create_voting_net(gamma=gamma, apns=train, distance=distance, classifier=classifier)
false_negative, false_positives = evaluate_voting_net(apns=test, classifier=classifier, net=mv, distance=distance)
res[gamma] = [false_negative, false_positives]
nets[gamma] = mv.copy()
sizes = {g:len(n.keys()) for g, n in nets.items()}
perfs = pd.DataFrame.from_dict(res, orient='index', columns=['fp', 'fn'])
szs = pd.DataFrame.from_dict(sizes, orient='index', columns=['sizes'])
gj = perfs.join(szs)
part_size = len(train)
gj['compression'] = gj.sizes/part_size
gj.to_csv(f"res/jac{part_size}.csv")