Feedback

.github/ISSUE_TEMPLATE/bug_template.md

@@ -0,0 +1,17 @@
+---
+name: Bug Report
+about: 버그 리포트할 때 사용하는 템플릿
+title: "[BUG] "
+labels: bug
+
+---
+
+## Description
+
+## How to reproduce
+
+1.
+2.
+3.
+
+## Solution

.github/ISSUE_TEMPLATE/feature_template.md

@@ -0,0 +1,16 @@
+---
+name: Feature Request
+about: 새로운 기능을 추가할 때 사용하는 템플릿
+title: "[FEAT] "
+labels: feature
+assignees: ''
+
+---
+
+## Description
+
+
+## Todo
+- [ ] 
+
+## ETC

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,12 @@
+## Overview
+-
+
+## Change Log
+-
+
+## To Reviewer
+-
+
+## Issue Tags
+- Closed | Fixed: # 
+- See also: #

.gitignore

@@ -0,0 +1,8 @@
+output
+wandb
+__pycache__
+saved
+submit
+log
+log_tensorboard
+data

README.md

@@ -0,0 +1,136 @@
+# Movie Recommendation
+
+## OverView
+본 프로젝트는 'Movie Recommendation'으로, 사용자의 영화 시청 이력 데이터를 토대로 사용자가 다음에 시청할 영화 및 좋아할 영화를 예측하는 기존 대회들과는 다르게 사용자의 영화 시청 이력에서 10개를 랜덤으로 제거한 후 제거된 시청 이력이 무엇인지를 맞춰야 하는 대회이다.
+
+모든 사용자에게 10개의 영화를 추천하며, 이때 추천 리스트의 정확성(Recall@10)을 평가 기준으로 삼는다.
+
+평가를 위한 정답(ground-truth) 데이터는 Sequential Recommendation 시나리오를 바탕으로 사용자의 Time-Ordered Sequence에서 일부 Item이 누락(dropout)된 상황을 상정한다
+<br><br>
+
+## Component
+
+### 프로젝트 디렉토리 구조 
+```
+📦level2-movierecommendation-recsys-03-main
+ ┗ 📂code
+   ┣ 📂EASER
+   ┣ 📂EDA
+   ┣ 📂bert4rec
+   ┣ 📂custom
+   ┣ 📂multi
+   ┣ 📂recbole
+   ┃ ┗ 📂configs - .yaml
+   ┃   ┣ ADMMSLIM, DIFFRec, EASE, deepfM
+   ┃   ┣ fm, lightgcn, ract, recvae, 
+   ┃   ┗ seq, slim
+   ┣ 📂s3rec
+   ┗ 📂voting
+```
+### 데이터셋 구조
+```
+📦level2-movierecommendation-recsys-03-main
+ ┗ 📂train
+	 ┣ 📜Ml_item2attributes.json
+	 ┣ 📜directors.tsv
+	 ┣ 📜genres.tsv
+	 ┣ 📜titles.tsv
+	 ┣ 📜train_ratings.csv
+	 ┣ 📜writers.tsv
+	 ┗ 📜years.tsv
+```
+
+
+## Team
+<br>
+<table align="left">
+  <tr height="155px">
+    <td align="center" width="150px">
+      <a href="https://github.com/ksb3966"><img src="https://github.com/ksb3966.png" width="100px;" alt=""/></a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/SiwooPark00"><img src="https://github.com/SiwooPark00.png" width="100px;" alt=""/></a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/arctic890"><img src="https://github.com/arctic890.png" width="100px;" alt=""/></a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/JaeGwon-Lee"><img src="https://github.com/JaeGwon-Lee.png" width="100px;" alt=""/></a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/jinmin111"><img src="https://github.com/jinmin111.png" width="100px;" alt=""/></a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/chris3427"><img src="https://github.com/chris3427.png" width="100px;" alt=""/></a>
+    </td>
+  </tr>
+  <tr height="80px">
+    <td align="center" width="150px">
+      <a href="https://github.com/ksb3966">김수빈_T6021</a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/SiwooPark00">박시우_T6060</a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/arctic890">백승빈_T6075</a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/JaeGwon-Lee">이재권_T6131</a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/jinmin111">이진민_T6139</a>
+    </td>
+    <td align="center" width="150px">
+      <a href="https://github.com/chris3427">장재원_T6149</a>
+    </td>
+  </tr>
+</table>
+&nbsp;
+<br>
+
+## Role
+
+| 이름 | 역할 |
+| --- | --- |
+| 김수빈 | EDA, 모델 선정 및 튜닝, EASER, SASRec, Bert4Rec 실험 수행   |
+| 박시우 | EDA, s3rec baseline 정리, recbole baseline 구축, RecVAE, ADMMSLIM, soft voting 앙상블 구현 |
+| 백승빈 | EDA, EASE, SLIMElastic, DiffRec 모델 실험 및 튜닝, hard voting기반 앙상블 구현 |
+| 이재권 | EDA, LightGCN 모델 실험 및 튜닝 |
+| 이진민 | EDA, Multi-DAE, Multi-VAE 코드 모듈화 및 실험 |
+| 장재원 | EDA, Sequential models, Type based model, 모델들 성능 비교 및 분석 |
+<br>
+
+## Experiment Result
+
+### Single Model Result
+|  | Public Recall@10 | Private Recall@10 |
+| --- | --- | --- |
+| Popular item rule based model | 0.0673 | 0.0671 |
+| Genre rule based model | 0.0619 | 0.0626 |
+| Type based model | 0.0687 | 0.0696 |
+| GRU4Rec | 0.0970 | 0.0809 |
+| SASRec | 0.0884 | 0.0833 |
+| S3Rec(Pretrained) | 0.0829 | 0.0743 |
+| BERT4Rec | 0.0687 | 0.0676 |
+| LightGCN | 0.1302 | 0.1316 |
+| DiffRec | 0.1413 | 0.1431 |
+| RecVAE | 0.1349 | 0.1362 |
+| Multi-VAE | 0.1394 | 0.1377   |
+| Multi-DAE (with side information) | 0.1427 | 0.1413 |
+| EASE | 0.1566 | 0.1565 |
+| ADMMSLIM | 0.1524 | 0.1541 |
+| SLIMElastic | 0.1562 | 0.1562  |
+| EASER | 0.1612 | 0.1603 |
+
+### Ensemble Result
+|  | Private Recall@10 | Public Recall@10 |
+| --- | --- | --- |
+| EASER, 앙상블 모델*에 type별 추천 적용 | 0.1614 | 0.1605 |
+| 앙상블 모델*(EASE, ADMMSLIM, RecVAE) | 0.1613 | 0.1611 |
+
+최종적으로는 가장 높은 Public score 값에 해당하는 두 결과 값을 제출하였다.
+<br><br>
+
+## Wrap-Up Report
+[MovieRec Wrap-up Report - Suggestify.pdf](./Movie_Recommendation_Wrap-up_Report-Suggestify.pdf)
+<br>

code/EASER/dataloader.py

@@ -0,0 +1,77 @@
+import pandas as pd
+import os
+from pytz import timezone
+from train_args import parse_args
+from collections import defaultdict
+import numpy as np
+
+import warnings
+warnings.filterwarnings(action="ignore")
+
+class Dataloader():
+    def __init__(self, config):
+        self.config = config
+        self.df = pd.read_csv(os.path.join(self.config.data_path, 'train_ratings.csv'))
+
+        self.users = self.df.user.unique()
+        self.items = self.df.item.unique()
+
+        self.item_encoder, self.item_decoder = self.generate_encoder_decoder('item')
+        self.user_encoder, self.user_decoder = self.generate_encoder_decoder('user')
+        self.num_item, self.num_user = len(self.item_encoder), len(self.user_encoder)
+
+        self.df['item_idx'] = self.df['item'].apply(lambda x : self.item_encoder[x])
+        self.df['user_idx'] = self.df['user'].apply(lambda x : self.user_encoder[x])
+
+        self.user_train, self.user_valid = self.generate_sequence_data()
+
+    def dataloader(self):
+        return self.df, self.users, self.items
+
+    def generate_encoder_decoder(self, col : str) -> dict:
+        """
+        encoder, decoder 생성
+
+        Args:
+            col (str): 생성할 columns 명
+        Returns:
+            dict: 생성된 user encoder, decoder
+        """
+
+        encoder = {}
+        decoder = {}
+        ids = self.df[col].unique()
+
+        for idx, _id in enumerate(ids):
+            encoder[_id] = idx
+            decoder[idx] = _id
+
+        return encoder, decoder
+
+    def generate_sequence_data(self) -> dict:
+        """
+        sequence_data 생성
+
+        Returns:
+            dict: train user sequence / valid user sequence
+        """
+        users = defaultdict(list)
+        user_train = {}
+        user_valid = {}
+        for user, item, time in zip(self.df['user_idx'], self.df['item_idx'], self.df['time']):
+            users[user].append(item)
+
+        for user in users:
+            np.random.seed(self.config.seed)
+
+            user_total = users[user]
+            valid = np.random.choice(user_total, size = self.config.valid_samples, replace = False).tolist()
+            train = list(set(user_total) - set(valid))
+
+            user_train[user] = train
+            user_valid[user] = valid # valid_samples 개수 만큼 검증에 활용 (현재 Task와 가장 유사하게)
+
+        return user_train, user_valid
+
+    def get_train_valid_data(self):
+        return self.user_train, self.user_valid