def __load(self):
test_pids = mpdutils.get_pids_from_dataset_json(self.test)
for file in [f for f in listdir(self.mpd) if f.endswith(".json")]:
print("Loading gtp data from %s" % file)
playlists = mpdutils.read_dataset_json(join(self.mpd, file))
for p in playlists:
if p["pid"] in test_pids:
continue
for t in p["tracks"]:
self.track_frequencies[t["track_uri"]] += 1
def _count_tracks(self):
for f in [
file for file in listdir(self.mpd) if file.endswith(".json")
]:
print("Counting track frequencies %s" % f)
playlists = mpdutils.read_dataset_json(join(self.mpd, f))
for p in playlists:
tmp = set()
for t in p["tracks"]:
track_uri = t["track_uri"]
if track_uri not in tmp:
self.current_frequencies[track_uri] += 1
tmp.add(track_uri)
def __load(self):
test_pids = mpdutils.get_pids_from_dataset_json(self.test)
for file in [f for f in listdir(self.mpd) if f.endswith(".json")]:
print("Loading ntm data from %s" % file)
playlists = mpdutils.read_dataset_json(join(self.mpd, file))
for p in playlists:
if p["pid"] in test_pids:
continue
nt = normalize(p["name"])
for t in p["tracks"]:
self.title_memberships[nt][t["track_uri"]] += 1
self.track_frequencies[t["track_uri"]] += 1
def run(self):
print("Generating recommendations...")
recommendations = dict()
playlists = mpdutils.read_dataset_json(self.test)
cnt = 0
for p in playlists:
recs = self.model.recommend(p, self.n)
recommendations[p["pid"]] = recs
cnt += 1
if cnt % 100 == 0:
print(cnt)
self._export(recommendations)
print("Recommendations are produced %s" % self.out)
def serialize(self):
test_pids = mpdutils.get_pids_from_dataset_json(self.test)
norm = LookupNormalization()
data = defaultdict(lambda: [])
for file in [f for f in listdir(self.mpd) if f.endswith(".json")]:
print("Converting playlists from %s" % file)
playlists = mpdutils.read_dataset_json(join(self.mpd, file))
for p in playlists:
if p["pid"] in test_pids:
continue
lu = norm.lookup(p["name"])
tracks = [t["track_uri"] for t in p["tracks"]]
data[lu].extend(tracks)
with open(self.out, "w", encoding="utf-8") as o:
for k, v in data.items():
o.write(k)
o.write("\t")
o.write(" ".join(v))
o.write("\n")
def analyze(self):
if self.test.endswith(".json"):
print("Analyzing %s" % self.test)
summary, stats = [], defaultdict(lambda: dict(
instances=0, num_tracks=[], num_samples=[], num_holdouts=[]))
for p in mpdutils.read_dataset_json(self.test):
cid = p["category"]
stats[cid]["instances"] += 1
stats[cid]["num_tracks"].append(p["num_tracks"])
stats[cid]["num_samples"].append(p["num_samples"])
stats[cid]["num_holdouts"].append(p["num_holdouts"])
total, all_tracks, all_samples, all_holdouts = 0, [], [], []
for k, v in sorted(stats.items()):
summary.append([
k, v["instances"],
statistics.mean(v["num_tracks"]),
statistics.mean(v["num_samples"]),
statistics.mean(v["num_holdouts"])
])
total += v["instances"]
all_tracks.extend(v["num_tracks"])
all_samples.extend(v["num_samples"])
all_holdouts.extend(v["num_holdouts"])
summary.append([
"overall", total,
statistics.mean(all_tracks),
statistics.mean(all_samples),
statistics.mean(all_holdouts)
])
print(tabulate(summary, headers=self.headers))
print()
def _search(self, limit=100):
low, up = mpdutils.COLD_START["lower"], mpdutils.COLD_START["upper"]
files = [f for f in listdir(self.mpd) if f.endswith(".json")]
random.shuffle(files)
while True:
file = files.pop()
print("Searching for candidate playlists %s" % file)
playlists = mpdutils.read_dataset_json(join(self.mpd, file))
random.shuffle(playlists)
candidates = [p for p in playlists if low <= p["num_tracks"]][:limit] if self.mode == "custom" \
else [p for p in playlists if low <= p["num_tracks"] <= up][:limit]
for c in candidates:
if self._is_applicable(c):
pjson = dict(pid=c["pid"],
name=c["name"],
category=mpdutils.COLD_START["id"],
num_tracks=c["num_tracks"],
num_samples=mpdutils.COLD_START["seeds"],
num_holdouts=c["num_tracks"],
tracks=[],
holdouts=sorted(c["tracks"],
key=lambda x: x["pos"],
reverse=False))
self.reservations.append(pjson)
for t in c["tracks"]:
track_uri = t["track_uri"]
self.current_frequencies[track_uri] -= 1
if len(self.reservations) >= self.size:
break
def __load_holdouts(test):
data = dict()
playlists = mpdutils.read_dataset_json(test)
for p in playlists:
data[p["pid"]] = [t["track_uri"] for t in p["holdouts"]]
return data
