Collaborative Filtering with FastAI

Here is a little boilerplate code for collaborative filtering with FastAI in just a few lines of code!

::: {.cell _kg_hide-input=‘true’ _kg_hide-output=‘true’ execution=‘{“iopub.execute_input”:“2023-01-12T05:15:34.597707Z”,“iopub.status.busy”:“2023-01-12T05:15:34.596398Z”,“iopub.status.idle”:“2023-01-12T05:15:34.605714Z”,“shell.execute_reply”:“2023-01-12T05:15:34.604205Z”,“shell.execute_reply.started”:“2023-01-12T05:15:34.597652Z”}’ papermill=‘{“duration”:4.977321,“end_time”:“2022-08-08T09:11:57.701696”,“exception”:false,“start_time”:“2022-08-08T09:11:52.724375”,“status”:“completed”}’ tags=‘[]’ trusted=‘true’ execution_count=23}

import pandas as pd
from datetime import datetime
from fastai import *
from fastai.collab import *
from fastai.tabular import *
from pathlib import Path
pd.set_option('mode.chained_assignment', 'warn')

import warnings
warnings.filterwarnings('ignore')

:::

Data Prep Step

base_path = Path('/kaggle/input/the-movies-dataset')

credits = pd.read_csv(base_path/'credits.csv')
keywords = pd.read_csv(base_path/'keywords.csv')
movies = pd.read_csv(base_path/'movies_metadata.csv').\
                     drop(['belongs_to_collection', 'homepage', 'imdb_id', 'poster_path', 'status', 'title', 'video'], axis=1).\
                     drop([19730, 29503, 35587]) # Incorrect data type

movies['id'] = movies['id'].astype('int64')

df = movies.merge(keywords, on='id').\
    merge(credits, on='id')

df['original_language'] = df['original_language'].fillna('')
df['runtime'] = df['runtime'].fillna(0)
df['tagline'] = df['tagline'].fillna('')

df.dropna(inplace=True)

ratings_df = pd.read_csv(base_path/'ratings_small.csv')

ratings_df['date'] = ratings_df['timestamp'].apply(lambda x: datetime.fromtimestamp(x))
ratings_df.drop('timestamp', axis=1, inplace=True)

ratings_df = ratings_df.merge(df[['id', 'original_title', 'genres', 'overview']], left_on='movieId',right_on='id', how='left')
ratings_df = ratings_df[~ratings_df['id'].isna()]
ratings_df.drop('id', axis=1, inplace=True)
ratings_df.reset_index(drop=True, inplace=True)

ratings_df.head()

movies_df = df[['id', 'original_title']]
movies_df.rename(columns={'id':'movieId'}, inplace=True)
ratings_df.merge(movies_df)

	userId	movieId	rating	date	original_title	genres	overview
0	1	1371	2.5	2009-12-14 02:52:15	Rocky III	[{'id': 18, 'name': 'Drama'}]	Now the world champion, Rocky Balboa is living in luxury and only fighting opponents who pose no threat to him in the ring. His lifestyle of wealth and idleness is shaken when a powerful young fighter known as Clubber Lang challenges him to a bout. After taking a pounding from Lang, the humbled champ turns to former bitter rival Apollo Creed to help him regain his form for a rematch with Lang.
1	4	1371	4.0	2000-02-06 04:11:42	Rocky III	[{'id': 18, 'name': 'Drama'}]	Now the world champion, Rocky Balboa is living in luxury and only fighting opponents who pose no threat to him in the ring. His lifestyle of wealth and idleness is shaken when a powerful young fighter known as Clubber Lang challenges him to a bout. After taking a pounding from Lang, the humbled champ turns to former bitter rival Apollo Creed to help him regain his form for a rematch with Lang.
2	7	1371	3.0	1996-12-29 14:19:20	Rocky III	[{'id': 18, 'name': 'Drama'}]	Now the world champion, Rocky Balboa is living in luxury and only fighting opponents who pose no threat to him in the ring. His lifestyle of wealth and idleness is shaken when a powerful young fighter known as Clubber Lang challenges him to a bout. After taking a pounding from Lang, the humbled champ turns to former bitter rival Apollo Creed to help him regain his form for a rematch with Lang.
3	19	1371	4.0	1997-02-06 01:43:24	Rocky III	[{'id': 18, 'name': 'Drama'}]	Now the world champion, Rocky Balboa is living in luxury and only fighting opponents who pose no threat to him in the ring. His lifestyle of wealth and idleness is shaken when a powerful young fighter known as Clubber Lang challenges him to a bout. After taking a pounding from Lang, the humbled champ turns to former bitter rival Apollo Creed to help him regain his form for a rematch with Lang.
4	21	1371	3.0	1997-01-21 13:11:03	Rocky III	[{'id': 18, 'name': 'Drama'}]	Now the world champion, Rocky Balboa is living in luxury and only fighting opponents who pose no threat to him in the ring. His lifestyle of wealth and idleness is shaken when a powerful young fighter known as Clubber Lang challenges him to a bout. After taking a pounding from Lang, the humbled champ turns to former bitter rival Apollo Creed to help him regain his form for a rematch with Lang.
...	...	...	...	...	...	...	...
45184	652	129009	4.0	2015-09-19 19:27:07	Love Is a Ball	[{'id': 35, 'name': 'Comedy'}, {'id': 10749, 'name': 'Romance'}]	Etienne makes a good living out of marrying off poor but titled young men to rich but untitled young ladies. Millicent is now in his sights on the Riviera, and Grand Duke Gaspar is the bait. But what if Millicent starts to fancy planted chauffeur John instead, and Gaspar takes a shine to Etienne's secretary Janine?
45185	653	2103	3.0	2000-01-18 02:04:26	Solaris	[{'id': 18, 'name': 'Drama'}, {'id': 878, 'name': 'Science Fiction'}, {'id': 9648, 'name': 'Mystery'}, {'id': 10749, 'name': 'Romance'}]	Upon arrival at the space station orbiting an ocean world called Solaris a psychologist discovers that the commander of an expedition to the planet has died mysteriously. Other strange events soon start happening as well, such as the appearance of old acquaintances of the crew, including some who are dead.
45186	659	167	4.0	1996-06-30 12:25:50	K-PAX	[{'id': 18, 'name': 'Drama'}, {'id': 878, 'name': 'Science Fiction'}]	Prot is a patient at a mental hospital who claims to be from a far away Planet. His psychiatrist tries to help him, only to begin to doubt his own explanations.
45187	659	563	3.0	1996-06-13 19:29:47	Starship Troopers	[{'id': 12, 'name': 'Adventure'}, {'id': 28, 'name': 'Action'}, {'id': 53, 'name': 'Thriller'}, {'id': 878, 'name': 'Science Fiction'}]	Set in the future, the story follows a young soldier named Johnny Rico and his exploits in the Mobile Infantry. Rico's military career progresses from recruit to non-commissioned officer and finally to officer against the backdrop of an interstellar war between mankind and an arachnoid species known as "the Bugs".
45188	665	129	3.0	2001-07-15 21:28:48	千と千尋の神隠し	[{'id': 14, 'name': 'Fantasy'}, {'id': 12, 'name': 'Adventure'}, {'id': 16, 'name': 'Animation'}, {'id': 10751, 'name': 'Family'}]	A ten year old girl who wanders away from her parents along a path that leads to a world ruled by strange and unusual monster-like animals. Her parents have been changed into pigs along with others inside a bathhouse full of these creatures. Will she ever see the world how it once was?

45189 rows × 7 columns

DataLoader Creation Step

dls = CollabDataLoaders.from_df(ratings_df, user_name='userId', item_name='original_title', rating_name='rating', bs=64)
dls.show_batch()

	userId	original_title	rating
0	346	The Getaway	1.0
1	615	Yesterday	3.5
2	19	Jaws: The Revenge	1.0
3	452	The Aviator	3.0
4	102	Houseboat	2.0
5	294	Confessions of a Dangerous Mind	4.0
6	357	Pirates of the Caribbean: Dead Man's Chest	5.0
7	518	Stuck on You	4.0
8	279	Crustacés et coquillages	3.0
9	115	バトル・ロワイアル	4.5

Model Creation/Training Step

learn = collab_learner(dls, n_factors=50, y_range=(0, 5.5))
learn.fit_one_cycle(5, 5e-3, wd=.1)

epoch	train_loss	valid_loss	time
0	0.930709	0.902386	00:05
1	0.786654	0.789312	00:05
2	0.572339	0.765915	00:05
3	0.408200	0.764980	00:06
4	0.322286	0.766817	00:05

learn.model

EmbeddingDotBias(
  (u_weight): Embedding(672, 50)
  (i_weight): Embedding(2777, 50)
  (u_bias): Embedding(672, 1)
  (i_bias): Embedding(2777, 1)
)

Prediction

movie_names = list(movies_df.drop_duplicates(subset='movieId', keep='first').original_title)
size = len(movie_names)

def predict_top_movies(userId, count=5):
  query = { 'userId': [userId] * size, 'original_title': movie_names }
  query_df = pd.DataFrame(data=query)
  query_dl = dls.test_dl(query_df)
  preds, y = learn.get_preds(dl=query_dl)
  results = sorted(zip(preds, movie_names), reverse=True)[:count]
  for idx, (score, name) in enumerate(results):
    print("Score: ", round(float(score), 2), " for movie: ", name)

def predict_user_rating(userId, movieName):
  query = { 'userId': [userId] * size, 'original_title': [movieName] * size}
  query_df = pd.DataFrame(data=query)
  query_dl = dls.test_dl(query_df)
  preds, y = learn.get_preds(dl=query_dl)
  results = sorted(zip(preds, [movieName]), reverse=True)
  for idx, (score, name) in enumerate(results):
    print("Score: ", round(float(score), 2), " for movie: ", movieName)

predict_top_movies(123)
predict_user_rating(123, 'Minions')

Score:  4.83  for movie:  Sleepless in Seattle
Score:  4.64  for movie:  Laura
Score:  4.6  for movie:  Galaxy Quest
Score:  4.58  for movie:  Men in Black II
Score:  4.56  for movie:  Lonely Hearts

Score:  3.29  for movie:  Minions

• https://www.kaggle.com/code/seandokko/collaborative-filtering-with-fastai
• https://candidcode.io
• https://linkedin.com/in/dokko1230
• https://github.com/dok