File size: 3,763 Bytes
eb90a8a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 | """
Program for computing MAP at R metric
using matrix of predicted similarity strength for all sample pairs
Input data are expected to be in the form of two pickled files:
- problem_indices.pcl -- specifies indices of tested problem
for each similarity test experiment
- similarity_probabilities.pcl -- specifies matrix of similarity test
experiment results
Each matrix column describes results
of comparing one problem solution with
solutions of all other problems
"""
import sys
import os
import argparse
import pickle
import numpy as np
def map_at_r(sim, pids):
"""
Function for computing MAP at R metric
Parameter:
- sim -- 2D numpy array of predicted similarity measures
for all pairs of samples
- pids -- 1D numpy array of problem ids corresponding
to columns of matrix of predicted similarity measures
Returns: computed MAP at R metric
"""
#Count number of source code solutions of each problem
r = np.bincount(pids) - 1
max_r = r.max()
#Mask for ignoring the similarity predictions lying
#beyond the number of solutions of checked problem
mask = np.arange(max_r)[np.newaxis, :] < r[pids][:, np.newaxis]
np.fill_diagonal(sim,-np.inf)
#Select and sort top predictions
result = np.argpartition(-sim,
range(max_r + 1), axis=1)[:, : max_r]
#Get correct similarity predictions
tp = pids[result] == pids[:, np.newaxis]
#Remove all predictions beyond the number of
#solutions of tested problem
tp[~mask] = False
#Get only tested problem
valid = r[pids] > 0
#Compute cumulative probability of correct predictions
p = np.cumsum(tp, axis=1,
dtype = np.float32) / np.arange(1, max_r+1,
dtype = np.float32)[np.newaxis, :]
#average across similarity prediction for each tested problem
ap = (p * tp).sum(axis=1)[valid] / r[pids][valid]
val = np.mean(ap).item()
return val
def main(args):
"""
Main function of program for computing MAP at R metric
Arguments are descibed below
"""
if not os.path.exists(args.similarities):
sys.exit(f"Directory {args.similarities} with similarity analysis does not exist")
with open(f"{args.similarities}/problem_indices.pcl", "rb") as _f:
pids = pickle.load(_f)
with open(f"{args.similarities}/similarity_probabilities.pcl", "rb") as _f:
sim = pickle.load(_f)
n_problem_solutions = pids.shape[0]
if sim.shape[0] != n_problem_solutions * n_problem_solutions:
sys.exit(
f"Number of similarity samples {n_problem_solutions.shape[0]} ",
f" is not square of number of problem solutions {n_problem_solutions}")
map_r = map_at_r(sim.reshape(n_problem_solutions, n_problem_solutions),
pids)
print("Map@R is ", map_r)
#######################################################################
# Command line arguments of are described below
#######################################################################
if __name__ == '__main__':
print("\nComputation of MAP at R metric of similarity analysis")
#Command-line arguments
parser = argparse.ArgumentParser(
description = "Computation of MAP at R metric")
parser.add_argument('similarities', type=str,
help='Directory with similarity results')
args = parser.parse_args()
print("Parameter settings used:")
for k,v in sorted(vars(args).items()):
print("{}: {}".format(k,v))
main(args)
|