run_greedy.py

# Runs the greedy algorithm.
# Command line arguments: first argument is path to data file. Second argument is path to save output. Third argument (*optional*) is path where the node suspiciousness values are stored.
import time
start_time = time.time()
from greedy import *
import sys

M = readData(sys.argv[1])
# The example data is a 500 x 500 matrix with an injected dense block among the first 20 nodes
print(("finished reading data: shape = %d, %d @ %d" % (M.shape[0], M.shape[1], time.time() - start_time)))

if len(sys.argv) > 3: # node suspiciousness present
	print("using node suspiciousness")
	rowSusp = np.loadtxt("%s.rows" % (sys.argv[3], ))
	colSusp = np.loadtxt("%s.cols" % (sys.argv[3], ))
	lwRes = logWeightedAveDegree(M, (rowSusp, colSusp))
else:
	lwRes = logWeightedAveDegree(M)

print(lwRes)
np.savetxt("%s.rows" % (sys.argv[2], ), np.array(list(lwRes[0][0])), fmt='%d')
np.savetxt("%s.cols" % (sys.argv[2], ), np.array(list(lwRes[0][1])), fmt='%d')
print(("score obtained is %f" % (lwRes[1],)))
print(("done @ %f" % (time.time() - start_time,)))

# When no node suspiciousness values are passed, we detect the injected block. However, when adding large suspiciousness values to the first 10 rows and columns, the new detected block becomes the first 10 rows and columns. Intuitively, the first 10 rows and columns are now suspiciousness enough that it no longer becomes worth it for the algorithm to include the remaining 10 rows and columns.