class PopularityWinrateDistribution:
def __init__(self, redis, name, max_items=9, bucket_size=5, ttl=600, use_lua=None):
self.name = name
self.max_items = max_items
self.bucket_size = bucket_size
self.observations = RedisPopularityDistribution(
redis,
name="%s_OBSERVATIONS" % self.name,
namespace="POPULARITY",
ttl=ttl,
max_items=self.max_items,
bucket_size=self.bucket_size,
use_lua=use_lua
)
self.wins = RedisPopularityDistribution(
redis,
name="%s_WINS" % self.name,
namespace="POPULARITY",
ttl=ttl,
max_items=self.max_items,
bucket_size=self.bucket_size,
use_lua=use_lua
)
def increment(self, key, win=False, as_of=None):
self.observations.increment(key, as_of=as_of)
if win:
self.wins.increment(key, as_of=as_of)
def distribution(self, start_ts, end_ts):
games = self.observations.distribution(
start_ts=start_ts,
end_ts=end_ts,
)
wins = self.wins.distribution(
start_ts=start_ts,
end_ts=end_ts,
)
result = {}
for key, val in games.items():
result[key] = {
"games": val,
"wins": wins.get(key, 0)
}
return result
def test_redis_popularity_distribution(_mock_lock): r = fakeredis.FakeStrictRedis() distribution = RedisPopularityDistribution(r, "DECKS", namespace="test") actuals = defaultdict(int) for deck in DECKS: actuals[deck] += 1 distribution.increment(deck) assert distribution.size() == 18 assert distribution.observations() == len(DECKS) actual_most_popular = list(sorted(actuals.items(), key=lambda t: t[1], reverse=True)) dist_most_popular = list( sorted(distribution.distribution().items(), key=lambda t: t[1], reverse=True) ) actual_result = actual_most_popular[0][0] expected_result = int(dist_most_popular[0][0]) assert actual_result == expected_result assert distribution.popularity(expected_result) == 32.0
def test_bucket_sizes_and_ttls(_mock_lock):
r = fakeredis.FakeStrictRedis()
distribution = RedisPopularityDistribution(
r,
"DECKS",
ttl=5,
namespace="test",
bucket_size=1
)
# Create t_0 as 5 seconds in the past
current_ts = datetime.utcnow()
td = timedelta(seconds=5, microseconds=current_ts.microsecond)
t_0 = current_ts - td
def t_N(N):
return t_0 + timedelta(seconds=N)
distribution.increment("A", as_of=t_N(1))
distribution.increment("B", as_of=t_N(2))
distribution.increment("A", as_of=t_N(3))
distribution.increment("A", as_of=t_N(4))
# First assert the full distribution exists
expected_distribution = {"A": 3.0, "B": 1.0}
actual_distribution = distribution.distribution()
assert expected_distribution == actual_distribution
# Then assert accessing a partial distribution (t_2, t_3) within the full time range
expected_distribution = {"A": 1.0, "B": 1.0}
actual_distribution = distribution.distribution(start_ts=t_N(2), end_ts=t_N(3))
assert expected_distribution == actual_distribution
# Finally, assert that the first observation of "A" has aged out due to the TTL
time.sleep(1)
expected_distribution = {"A": 2.0, "B": 1}
actual_distribution = distribution.distribution()
assert expected_distribution == actual_distribution