Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Optimization the base _filter_seen method #26

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Optimization the base _filter_seen method #26

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 10 additions & 32 deletions replay/models/base_rec.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,6 @@
filter_cold,
get_unique_entities,
get_top_k,
get_top_k_recs,
return_recs,
vector_euclidean_distance_similarity,
vector_dot,
Expand Down Expand Up @@ -457,33 +456,6 @@ def _filter_seen(
"""
users_log = log.join(users, on="user_idx")
self._cache_model_temp_view(users_log, "filter_seen_users_log")
num_seen = users_log.groupBy("user_idx").agg(
Copy link
Collaborator

@monkey0head monkey0head Mar 22, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

t seems those steps are not redundant.
some models (a lot of models) do not filter seen items inside _predict(). So if we want to predict with filter_seen_items=True, we

  1. predict more items than needed (k + max_seen),
  2. filter out seed items (thus we will get number of recommendations in interval [k, k + max_seen]
  3. get top-k most relevant

So, ether you placed seen filtering inside _predict method of each model and it is not in the current pr, or you just removed some necessary functionality. Lets discuss

sf.count("item_idx").alias("seen_count")
)
self._cache_model_temp_view(num_seen, "filter_seen_num_seen")

# count maximal number of items seen by users
max_seen = 0
if num_seen.count() > 0:
max_seen = num_seen.select(sf.max("seen_count")).collect()[0][0]

# crop recommendations to first k + max_seen items for each user
recs = recs.withColumn(
"temp_rank",
sf.row_number().over(
Window.partitionBy("user_idx").orderBy(
sf.col("relevance").desc()
)
),
).filter(sf.col("temp_rank") <= sf.lit(max_seen + k))

# leave k + number of items seen by user recommendations in recs
recs = (
recs.join(num_seen, on="user_idx", how="left")
.fillna(0)
.filter(sf.col("temp_rank") <= sf.col("seen_count") + sf.lit(k))
.drop("temp_rank", "seen_count")
)

# filter recommendations presented in interactions log
recs = recs.join(
Expand All @@ -495,6 +467,16 @@ def _filter_seen(
how="anti",
).drop("user", "item")

# crop recommendations to first k + max_seen items for each user
recs = recs.withColumn(
"temp_rank",
sf.row_number().over(
Window.partitionBy("user_idx").orderBy(
sf.col("relevance").desc()
)
),
).filter(sf.col("temp_rank") <= sf.lit(k))

return recs

# pylint: disable=too-many-arguments
Expand Down Expand Up @@ -557,10 +539,6 @@ def _predict_wrap(
if filter_seen_items and log:
recs = self._filter_seen(recs=recs, log=log, users=users, k=k)

recs = get_top_k_recs(recs, k=k).select(
"user_idx", "item_idx", "relevance"
)

output = return_recs(recs, recs_file_path)
self._clear_model_temp_view("filter_seen_users_log")
self._clear_model_temp_view("filter_seen_num_seen")
Expand Down