def main():
parser = argparse.ArgumentParser()
parser.add_argument("project_name", help="Name of project")
parser.add_argument("project_dir", help="Location of project dir")
args = parser.parse_args()
spectra_maker.make_spectra(
args.project_name, args.project_dir, projects.get_version_names(args.project_name), False
)
def compute_features(project_name):
if not os.path.exists(FEATURE_DIR):
os.mkdir(FEATURE_DIR)
for version in projects.get_version_names(project_name):
fname = get_feature_file(project_name, version)
if os.path.exists(fname):
print("V {0} Already exists, skipping".format(version))
continue
print("Computing version {0}".format(version))
spectra_file = spectra_maker.get_spectra_file(project_name, version)
make_feature_file(spectra_file, fname)
def get_total_scores(project_name, ranker_type, filter_type, provider_type,
versions=None):
version_to_rankres = {}
if versions is None:
versions = projects.get_version_names(project_name)
for version in versions:
failing_to_rankres = evaluator.get_ranker_results(project_name,
version,
ranker_type,
filter_type,
provider_type)
version_to_rankres.update(failing_to_rankres)
version_to_score = {ver: rank_res.score for ver, rank_res in
version_to_rankres.items()}
version_to_score = pd.Series(version_to_score)
version_to_score.sort_values(inplace=True, ascending=False)
return version_to_score
def optimize_classifier(project_name):
initial_scores = {}
for version in projects.get_version_names(project_name):
rank_res_d = evaluator.get_ranker_results(project_name,
version,
"intersection",
"none",
TrivialProvider())
assert len(rank_res_d) == 1
rank_res = next(iter(rank_res_d.values()))
initial_scores[version] = rank_res.score
def to_optimize(classify_vector_with_cutoff):
cutoff = classify_vector_with_cutoff[-1]
classify_vector = classify_vector_with_cutoff[:-1]
score = evaluation_fn(project_name, initial_scores,
classify_vector, cutoff)
# minimize negative
return -1 * score
vecsize = 14 + 1
lower = [-1.0 for i in range(vecsize)]
upper = [1.0 for i in range(vecsize)]
x0 = [0.1 for i in range(vecsize)]
res = scipy.optimize.anneal(to_optimize, x0,
#maxiter=1,
maxeval=10,
dwell=5,
full_output=True)
x0 = res[0]
print("Vec is {0}, cutoff {1}".format(x0[:-1], x0[-1]))
print(res)
def evaluation_fn(project_name, initial_scores,
classify_vector, cutoff):
print("Evaluating with cutoff {0} classify {1}".format(cutoff,
classify_vector))
scoresum = 0
ver = None
for version in projects.get_version_names(project_name):
features = feature_computer.get_feature_vecs(project_name, version)
filter_obj = spectra_filter.DotProductFilter(classify_vector,
cutoff,
features)
ranker_obj = tarantula.TarantulaRanker()
rank_res = get_res(project_name, version, ranker_obj, filter_obj)
if rank_res is None:
continue
score = rank_res.score
scorediff = score - initial_scores[version]
scoresum += scorediff
print("Score is {0} for version: {1}".format(scoresum, ver))
return scoresum
