def recommendations_from_seed(database, rule_index, seed_song):
# Find rules, that affect this seed_song:
associated = list(rule_index.lookup(seed_song))
# No rules? Just use the seed_song as starting point...
# One day, we will have rules. Shortcut for now.
if not associated:
bfs = sorted_breadth_first_search(seed_song)
# Throw away first song:
next(bfs)
for recom in bfs:
yield recom
else:
# Create an iterator for each seed_song, in each associated rule:
breadth_first_iters = deque()
# Now populate the list of breadth first iterators:
for left, right, *_ in associated:
# The maximum number that a single seed_song in this rule may
# deliver (at least 1 - himself, therefore the ceil)
bulk = right if seed_song in left else left
# We take the songs in the opposite set of the rule:
for bulk_song in bulk:
breadth_first = sorted_breadth_first_search(bulk_song)
breadth_first_iters.append(breadth_first)
# The result set will be build by half
base_half = sorted_breadth_first_search(seed_song)
try:
next(base_half)
except StopIteration:
pass
# Yield the base half first:
songs_set = set([seed_song])
# Now build the final result set by filling one half original songs,
# and one half songs that were pointed to by rules.
for recom in roundrobin(base_half, roundrobin(*breadth_first_iters)):
# We have this already in the result set:
if recom in songs_set:
continue
else:
songs_set.add(recom)
yield recom
def recommendations_from_seed(database, rule_index, seed_song):
# Find rules, that affect this seed_song:
associated = list(rule_index.lookup(seed_song))
# No rules? Just use the seed_song as starting point...
# One day, we will have rules. Shortcut for now.
if not associated:
bfs = sorted_breadth_first_search(seed_song)
# Throw away first song:
next(bfs)
for recom in bfs:
yield recom
else:
# Create an iterator for each seed_song, in each associated rule:
breadth_first_iters = deque()
# Now populate the list of breadth first iterators:
for left, right, *_ in associated:
# The maximum number that a single seed_song in this rule may
# deliver (at least 1 - himself, therefore the ceil)
bulk = right if seed_song in left else left
# We take the songs in the opposite set of the rule:
for bulk_song in bulk:
breadth_first = sorted_breadth_first_search(bulk_song)
breadth_first_iters.append(breadth_first)
# The result set will be build by half
base_half = sorted_breadth_first_search(seed_song)
try:
next(base_half)
except StopIteration:
pass
# Yield the base half first:
songs_set = set([seed_song])
# Now build the final result set by filling one half original songs,
# and one half songs that were pointed to by rules.
for recom in roundrobin(base_half, roundrobin(*breadth_first_iters)):
# We have this already in the result set:
if recom in songs_set:
continue
else:
songs_set.add(recom)
yield recom
def recommendations_from_attributes(subset,
database,
rule_index,
max_seeds=10,
max_numeric_offset=None):
chosen_songs = list(
islice(database.find_matching_attributes(subset, max_numeric_offset),
max_seeds))
if not chosen_songs:
return []
seed_iterables = [
recommendations_from_seed(database, rule_index, song)
for song in chosen_songs
]
return chain(chosen_songs, roundrobin(*seed_iterables))
def recommendations_from_attributes(
subset, database, rule_index,
max_seeds=10, max_numeric_offset=None
):
chosen_songs = list(islice(
database.find_matching_attributes(subset, max_numeric_offset),
max_seeds
))
if not chosen_songs:
return []
seed_iterables = [
recommendations_from_seed(database, rule_index, song)
for song in chosen_songs
]
return chain(
chosen_songs,
roundrobin(*seed_iterables)
)
