def __init__(self, dataset, epsilon, numClasses, numFeatures, batchSize):
print("initializing client")
data = utils.load_dataset(dataset, npy=True)
X = data['X']
y = data['y']
(n, d) = X.shape
X_train = X[range(0, n / 2), :]
y_train = y[range(0, n / 2)]
X_test = X[range(n / 2, n), :]
y_test = y[range(n / 2, n)]
self.ytest = y_test
self.batchSize = batchSize
self.softmax_model_obj = softmax_model_obj.SoftMaxModel(
X_train, y_train, epsilon, numClasses)
self.softmax_model_test = softmax_model_test.SoftMaxModelTest(
X_test, y_test, X_train, y_train, numClasses, numFeatures)
models = []
for i in range(numClasses):
# Try a little more IID
models.append(dataPath + str(i)) # + str((i + 1) % 10) + str((i
# + 2) % 10))
for attack in argv[2:]:
attack_delim = attack.split("_")
sybil_set_size = attack_delim[0]
from_class = attack_delim[1]
to_class = attack_delim[2]
for i in range(int(sybil_set_size)):
models.append(dataPath + "_bad_" + from_class + "_" + to_class)
softmax_test = softmax_model_test.SoftMaxModelTest(dataset, numClasses,
numFeatures)
# Hard code poisoners in a 2_x_x attack
threshholds = [0.29, 0.28, 0.27, 0.26, 0.25]
# threshholds = [0.5, 0.25, 0.1, 0.01, 0.001]
num_trials = 5
eval_data = np.zeros((num_trials * len(threshholds), 3))
for sim_i in range(len(threshholds)):
max_similarity = threshholds[sim_i]
for eval_i in range(num_trials):
print("Evaluating " + str(eval_i) + "th iteration of " +
str(max_similarity))
topk_prop = 0.05
weights, ratio = non_iid(max_similarity,
Xtest,
