def random_testing(dataset, language="java_python", n_folds=10):
LOGGER.info("Random testing with '%d' number of folds" % n_folds)
folds = get_cross_val(dataset, n_folds, as_dict=True)
base_folder = lib.get_clusters_folder(dataset)
LOGGER.info("Loading pickle ...")
cluster_path = get_cluster_path(dataset, language)
clusters = cache.load_pickle(cluster_path)
for index, fold in enumerate(folds):
file_name = os.path.join(base_folder, "random_testing",
"fold_%d" % index, "distances.pkl")
cluster_distances = {}
for label, functions in clusters.items():
if label == -1:
continue
similarity_map = defaultdict(dict)
for i in range(len(functions) - 1):
for j in range(i + 1, len(functions)):
assert i != j
f_i, f_j = functions[i], functions[j]
distance = clusterer.execution_distance(
fold[f_i.name], fold[f_j.name])
similarity_map[f_i.name][f_j.name] = distance
similarity_map[f_j.name][f_i.name] = distance
cluster_distances[label] = similarity_map
cache.save_pickle(file_name, cluster_distances)
def compute_similarity(dataset, language=None, functions=None, base_folder=None, file_name=None,
skip_singles=False, update_clone_meta=False, clustering_error=0.01, cluster_suffix="base"):
if not functions:
if language == "java":
functions = load_functions(dataset, update_clone_meta=update_clone_meta)
elif language == "python":
functions = load_py_functions(dataset)
elif language == "java_python":
functions = load_functions(dataset, update_clone_meta=update_clone_meta) + load_py_functions(dataset)
else:
raise RuntimeError("Invalid language: %s" % language)
# if dataset not in ["codejam", "introclass"]:
# raise RuntimeError("Invalid dataset: %s" % dataset)
LOGGER.info("Clustering ... ")
if file_name is None:
file_name = language or "clusters"
LOGGER.warning("A @file_name is not provided. Reverting file name to '%s'" % file_name)
if base_folder is None:
base_folder = lib.get_clusters_folder(dataset)
clusters_txt_file = os.path.join(base_folder, "%s.txt" % file_name)
clusters_pkl_file = os.path.join(base_folder, "%s.pkl" % file_name)
clusters_report_file = os.path.join(base_folder, "%s.md" % file_name)
clusters = get_clusterer()(functions).cluster(clusters_txt_file, skip_singles=skip_singles, clustering_error=clustering_error)
cache.save_pickle(clusters_pkl_file, clusters)
clusterer.save_clusters_to_db(dataset, clusters, cluster_suffix)
n_clusters = len(clusters)
sizes = [len(cluster_funcs) for label, cluster_funcs in clusters.items() if label != -1]
meta_data = "## Cluster sizes\n"
meta_data += "* Number of clusters: %d\n" % n_clusters
meta_data += "* Number of functions clustered: %d\n" % sum(sizes)
meta_data += "* Number of functions not clustered: %d\n\n" % (len(functions) - sum(sizes))
meta_data += "## REPORT\n"
meta_data += Stat(sizes).report()
cache.write_file(clusters_report_file, meta_data)
def save_only_mixed_clusters(dataset, mixed_file_base_name):
"""
Save only mixed functions
:param dataset: Name of dataset
:param mixed_file_base_name: Type of language eg. java_python
:return:
"""
clusters_base_folder = os.path.join(lib.get_clusters_folder(dataset), "cluster_testing")
for folder in sorted(cache.list_dir(clusters_base_folder, is_absolute=False)):
LOGGER.info("Processing '%s' ..." % folder)
folder_path = os.path.join(clusters_base_folder, folder)
base_clusters_file = os.path.join(folder_path, "%s.pkl" % mixed_file_base_name)
base_clusters = cache.load_pickle(base_clusters_file)
mixed_clusters = {}
for label, functions in base_clusters.items():
if label == -1 or len(functions) == 1: continue
sources = set()
for func in functions:
sources.add(func.source)
if len(sources) > 1:
mixed_clusters[label] = functions
LOGGER.info("For folder = %s, # of mixed clusters = %d" % (folder, len(mixed_clusters)))
file_path = os.path.join(folder_path, "only_mixed.txt")
pkl_path = os.path.join(folder_path, "only_mixed.pkl")
file_contents = []
for label, functions in mixed_clusters.items():
file_contents.append("\n\n****** Cluster %d ******" % label)
for func in functions:
file_contents.append(func.body)
cache.write_file(file_path, "\n".join(file_contents))
cache.save_pickle(pkl_path, mixed_clusters)
def fetch_statements(language,
force=False,
do_save=False,
limit=None,
as_list=False):
pkl_file = get_executed_stmts_pkl(language)
if not force and cache.file_exists(pkl_file):
LOGGER.info("Retrieving existing '%s' statements!" % language)
if as_list:
return cache.load_pickle(pkl_file).values()
return cache.load_pickle(pkl_file)
LOGGER.info("Reprocessing '%s' statements!" % language)
store = mongo_driver.MongoStore(props.DATASET)
stmts = {}
mongo_stmts = store.load_stmts(language=language,
is_valid=True,
has_output=True,
limit=limit).items()
n_stmts = len(mongo_stmts)
for i, (key, mongo_stmt) in enumerate(mongo_stmts):
LOGGER.info("Processing %d / %d .... " % (i + 1, n_stmts))
stmt = Statement(mongo_id=mongo_stmt["_id"],
snippet=mongo_stmt["snippet"],
variables=mongo_stmt["variables"],
language=language,
outputs=format_outputs(mongo_stmt["outputs"]))
stmts[stmt.mongo_id] = stmt
if do_save:
LOGGER.info("Saving statements .... ")
cache.save_pickle(pkl_file, stmts)
if as_list:
return stmts.values()
return stmts
def cluster(clustering_error):
functions = load_functions(PD_FUNCTIONS_PATH, "pandas")
functions.update(load_functions(R_FUNCTIONS_PATH, "R"))
file_name = "clusters"
folder = os.path.join(BASE_CLUSTER_FOLDER, "%0.02f" % clustering_error)
cache.mkdir(folder)
clusters_txt_file = os.path.join(folder, "%s.txt" % file_name)
clusters_pkl_file = os.path.join(folder, "%s.pkl" % file_name)
clusters_report_file = os.path.join(folder, "%s.md" % file_name)
clusterer = RepresentativeClusterer(functions.values(),
distance_function=execution_distance)
clusters = clusterer.cluster(clusters_txt_file,
skip_singles=True,
clustering_error=clustering_error)
cache.save_pickle(clusters_pkl_file, clusters)
n_clusters = len(clusters)
sizes = [
len(cluster_funcs) for label, cluster_funcs in clusters.items()
if label != -1
]
meta_data = "## Cluster sizes\n"
meta_data += "* Number of clusters: %d\n" % n_clusters
meta_data += "* Number of functions clustered: %d\n" % sum(sizes)
meta_data += "* Number of functions not clustered: %d\n\n" % (
len(functions) - sum(sizes))
meta_data += "## REPORT\n"
meta_data += stat.Stat(sizes).report()
cache.write_file(clusters_report_file, meta_data)
def remove_overlapping_clusters(dataset, language="java_python"):
# TODO: Think about how to remove syntactic equivalence
store = mongo_store.FunctionStore(dataset)
base_file = os.path.join(properties.META_RESULTS_FOLDER, dataset,
"clusters", "%s.pkl" % language)
clusters = cache.load_pickle(base_file)
non_overlapping_clusters = {}
for label, functions in clusters.items():
if label == -1 or len(functions) == 1: continue
non_overlapping_funcs = []
metas = {}
for func in functions:
meta = store.load_metadata({"name": func.base_name})
metas[func.base_name] = meta
if len(non_overlapping_funcs) == 0:
non_overlapping_funcs.append(func)
continue
is_non_overlapping_funcs_updated = False
for i, existing_func in enumerate(non_overlapping_funcs):
existing_meta = metas[existing_func.base_name]
if overlaps(meta, existing_meta):
is_non_overlapping_funcs_updated = True
if is_more_succinct(meta, existing_meta):
non_overlapping_funcs[i] = func
break
if not is_non_overlapping_funcs_updated:
non_overlapping_funcs.append(func)
if len(non_overlapping_funcs) > 1:
non_overlapping_clusters[label] = non_overlapping_funcs
write_folder = os.path.join(properties.META_RESULTS_FOLDER, dataset,
"clusters", "non_overlapping")
cache.mkdir(write_folder)
clusters_txt_file = os.path.join(write_folder, "%s.txt" % language)
clusters_pkl_file = os.path.join(write_folder, "%s.pkl" % language)
clusters_report_file = os.path.join(write_folder, "%s.md" % language)
cache.save_pickle(clusters_pkl_file, non_overlapping_clusters)
clusterer.save_clusters_to_db(dataset, non_overlapping_clusters,
"non_overlapping")
clusterer.save_clusters_to_txt(non_overlapping_clusters, clusters_txt_file)
sizes = [
len(cluster_funcs)
for label, cluster_funcs in non_overlapping_clusters.items()
if label != -1
]
meta_data = "## Cluster sizes\n"
meta_data += "* Number of clusters: %d\n" % len(non_overlapping_clusters)
meta_data += "* Number of functions clustered: %d\n" % sum(sizes)
meta_data += "## REPORT\n"
meta_data += Stat(sizes).report()
cache.write_file(clusters_report_file, meta_data)
def save_only_target_functions(dataset, mixed_file_base_name, target_language):
"""
Save only java functions from a mixture of java and python clusters
:param dataset: Name of dataset
:param mixed_file_base_name: Type of language eg. java_python
:param target_language: Target Language
:return:
"""
clusters_base_folder = os.path.join(lib.get_clusters_folder(dataset),
"cluster_testing")
for folder in sorted(
cache.list_dir(clusters_base_folder, is_absolute=False)):
LOGGER.info("Processing '%s' ..." % folder)
folder_path = os.path.join(clusters_base_folder, folder)
cache.mkdir(folder_path)
base_clusters_file = os.path.join(folder_path,
"%s.pkl" % mixed_file_base_name)
base_clusters = cache.load_pickle(base_clusters_file)
target_clusters = {}
for label, functions in base_clusters.items():
if label == -1 or len(functions) == 1: continue
contains_target = False
contains_other = False
for func in functions:
if func.source == target_language:
contains_target = True
else:
contains_other = True
if contains_target and not contains_other:
target_clusters[label] = functions
LOGGER.info("For folder = %s, # of '%s' clusters = %d" %
(folder, target_language, len(target_clusters)))
file_path = os.path.join(folder_path, "only_%s.txt" % target_language)
pkl_path = os.path.join(folder_path, "only_%s.pkl" % target_language)
file_contents = []
for label, functions in target_clusters.items():
file_contents.append("\n\n****** Cluster %d ******" % label)
for func in functions:
file_contents.append(func.body)
cache.write_file(file_path, "\n".join(file_contents))
cache.save_pickle(pkl_path, target_clusters)
def save_function(func_data):
saved_funcs = cache.load_pickle(FUNCTION_STORE)
if not saved_funcs:
saved_funcs = {}
saved_funcs[func_data["name"]] = func_data
cache.save_pickle(FUNCTION_STORE, saved_funcs)
def store_args(key, args):
data_store = cache.load_pickle(STORE_PATH)
if not data_store:
data_store = {}
data_store[key] = args
cache.save_pickle(STORE_PATH, data_store)
