Question regarding the choice of repair parameters.

[zhe-tao]

The interface exposes two repair parameters, namely n and num_core:

REASSURE/ICLR/Experiments/MNIST/PNN_MNIST.py

Line 12 in 4781c8e

def PNN_MNIST(repair_num, n, num_core, remove_redundant_constraint=False):

Regarding n, how would the user determine it? Is n=0.5 as used in PNN_MNIST.py always a reasonable choice?

REASSURE/ICLR/Experiments/MNIST/PNN_MNIST.py

Line 59 in 4781c8e

PNN_MNIST(num, 0.5, num_core=10, remove_redundant_constraint=False)

Regarding num_core, considering it actually controls the number of buggy inputs to be repaired in parallel, would os.cpu_count() always be a reasonable choice?

REASSURE/ICLR/tools/build_PNN.py

Lines 234 to 244 in 4781c8e

	pool = multiprocessing.Pool(core_num)
	if self.buggy_points is not None:
	arg_list = [[self.buggy_points[i], self.P[i], self.ql[i], self.qu[i], is_gurobi] for i in
	range(len(self.buggy_points))]
	res = pool.starmap(self.PatchForOnePoint, arg_list)
	else:
	arg_list = [[self.linear_regions[i][0], self.linear_regions[i][1], self.linear_regions[i][2], self.P[i],
	self.ql[i], self.qu[i]] for i in range(len(self.linear_regions))]
	res = pool.starmap(self.PatchForOneLinearRegion, arg_list)
	pool.close()
	g_list, cd_list = [res_g[0] for res_g in res], [res_g[1] for res_g in res]