only do trending calculations near the tip of the chain

This commit is contained in:
Alex Grintsvayg 2019-10-11 10:51:36 -04:00
parent eb8a88dbd6
commit 065215cc27
No known key found for this signature in database
GPG key ID: AEB3F089F86A22B5
2 changed files with 13 additions and 9 deletions

View file

@ -1,7 +1,11 @@
from math import sqrt from math import sqrt
TRENDING_WINDOW = 134 # number of blocks in ~6hr period (21600 seconds / 161 seconds per block) # TRENDING_WINDOW is the number of blocks in ~6hr period (21600 seconds / 161 seconds per block)
TRENDING_DATA_POINTS = 7 # WINDOW * DATA_POINTS = ~1 week worth of trending data TRENDING_WINDOW = 134
# TRENDING_DATA_POINTS says how many samples to use for the trending algorithm
# i.e. only consider claims from the most recent (TRENDING_WINDOW * TRENDING_DATA_POINTS) blocks
TRENDING_DATA_POINTS = 7
CREATE_TREND_TABLE = """ CREATE_TREND_TABLE = """
create table if not exists trend ( create table if not exists trend (
@ -26,7 +30,7 @@ class ZScore:
if self.last is not None: if self.last is not None:
self.count += 1 self.count += 1
self.total += self.last self.total += self.last
self.power += self.last**2 self.power += self.last ** 2
self.last = value self.last = value
@property @property
@ -35,7 +39,7 @@ class ZScore:
@property @property
def standard_deviation(self): def standard_deviation(self):
return sqrt((self.power / self.count) - self.mean**2) return sqrt((self.power / self.count) - self.mean ** 2)
def finalize(self): def finalize(self):
if self.count == 0: if self.count == 0:
@ -47,19 +51,19 @@ def register_trending_functions(connection):
connection.create_aggregate("zscore", 1, ZScore) connection.create_aggregate("zscore", 1, ZScore)
def calculate_trending(db, height, is_first_sync, final_height): def calculate_trending(db, height, final_height):
# don't start tracking until we're at the end of initial sync # don't start tracking until we're at the end of initial sync
if is_first_sync and height < (final_height - (TRENDING_WINDOW*TRENDING_DATA_POINTS)): if height < (final_height - (TRENDING_WINDOW * TRENDING_DATA_POINTS)):
return return
if height % TRENDING_WINDOW != 0: if height % TRENDING_WINDOW != 0:
return return
db.execute(f""" db.execute(f"""
DELETE FROM trend WHERE height < {height-(TRENDING_WINDOW*TRENDING_DATA_POINTS)} DELETE FROM trend WHERE height < {height - (TRENDING_WINDOW * TRENDING_DATA_POINTS)}
""") """)
start = (height-TRENDING_WINDOW)+1 start = (height - TRENDING_WINDOW) + 1
db.execute(f""" db.execute(f"""
INSERT OR IGNORE INTO trend (claim_hash, height, amount) INSERT OR IGNORE INTO trend (claim_hash, height, amount)
SELECT claim_hash, {start}, COALESCE( SELECT claim_hash, {start}, COALESCE(

View file

@ -699,7 +699,7 @@ class SQLDB:
update_claims, delete_claim_hashes, affected_channels, forward_timer=True) update_claims, delete_claim_hashes, affected_channels, forward_timer=True)
r(self.insert_supports, insert_supports) r(self.insert_supports, insert_supports)
r(self.update_claimtrie, height, recalculate_claim_hashes, deleted_claim_names, forward_timer=True) r(self.update_claimtrie, height, recalculate_claim_hashes, deleted_claim_names, forward_timer=True)
r(calculate_trending, self.db, height, self.main.first_sync, daemon_height) r(calculate_trending, self.db, height, daemon_height)
class LBRYDB(DB): class LBRYDB(DB):