def recommendations():
rec = Recommender()
rec.import_cosine_sim_from_pkl()
movies_enhanched = []
msg = ''
if request.method == "POST":
tconst_id = request.form['tconst']
dbhandler = DbHandler()
movie_title = dbhandler.exec_sql_cmd(
f"SELECT primaryTitle FROM title_basics WHERE tconst = '{tconst_id}'"
)
msg = f"Recommendations for '{movie_title[0][0]}'"
print(msg)
movies = rec.get_recommendation_titles_from_tconst(tconst=tconst_id,
limit=20)
for movie in movies[1:]:
movies_enhanched.append({
"tconst":
movie[0],
"title":
movie[1],
"imdburl":
f"https://www.imdb.com/title/{movie[0]}/"
})
return render_template("recommendations.html",
name="Apostolis",
movies=movies_enhanched,
result_message=msg)
def test_get_titles_from_tconst(self):
"""
Test the get_titles_from_tconst_list function
Grab the recommendation titles for a random movie and crosscheck with the titles in the title_basics table
:return:
"""
rec = Recommender()
rec.import_cosine_sim_from_pkl()
sample_tconst = random.choice(list(
rec.cosine_sim.tconst.values)) # grab a random movie to test
sample_recommendations_tconst = rec.get_recommendation_from_tconst(
sample_tconst)
sample_titles = get_titles_from_tconst_list(
sample_recommendations_tconst)
recommendation_titles = [result[1] for result in sample_titles]
# Make sure the titles returned are a subset of the titles in title_basics table
dbhandler = DbHandler()
dbhandler.connect()
# Use a join to limit the results only on those that interest us
all_titles = dbhandler.conn.execute(
text(
"SELECT tconst,primaryTitle from title_basics NATURAL JOIN title_keywords"
))
# Convert the result of the query to a df
all_titles_df = pd.DataFrame(data=[row for row in all_titles],
columns=['tconst', 'primaryTitle'])
# Find the titles that exist in both the df from the db and the results from the recommender
same_titles = all_titles_df.loc[all_titles_df['tconst'].isin(
sample_recommendations_tconst)]['primaryTitle']
# Make sure we found all the movie titles
assert len(recommendation_titles) == same_titles.size
def insert_to_db(self):
myDbHandler = DbHandler()
myDbHandler.connect()
self.filtered_df.to_sql("title_basics",
myDbHandler.conn,
if_exists='append',
index=False)
return
def test_execute_select_from_sql(self):
handler = DbHandler()
handler.connect()
try:
results = handler.exec_select_sql_from_file(os.path.join(DATA_PATH, "sql/select_actors.sql"))
print([r['tconst'] for r in results][:2])
except Exception as e:
assert False, e
return
def __init__(self):
self.dbhandler = DbHandler()
self.dbhandler.connect()
data = self.dbhandler.exec_select_sql_from_file(
os.path.join(DATA_PATH, "sql/select_soup.sql"))
self.df = pd.DataFrame(data=data, columns=['tconst', 'soup'])
self.cosine_sim = None
self.cosine_sim_csv_path = os.path.join(DATA_PATH,
'cosine_sim/cosine_sim.csv')
return
def filter_foreign_keys(self):
myDbHandler = DbHandler()
myDbHandler.connect()
nconst_ids = [
row["nconst"] for row in myDbHandler.conn.execute(
text("SELECT DISTINCT nconst \ "
"FROM title_principals"))
]
self.filtered_df = self.filtered_df[self.filtered_df.nconst.isin(
nconst_ids)]
return
def test_execute_query(self):
"""
Run a sample query to make sure the connection is indeed correct and the schema is created
:return:
"""
handler = DbHandler()
handler.connect()
try:
handler.conn.execute(text("SELECT table_name FROM information_schema.tables \
WHERE table_schema = 'movie_recommender';"))
except Exception as e:
assert False, e
def insert_to_db(self):
"""
Insert the filtered_df in the db, tablename: title_ratings
:return:
"""
myDbHandler = DbHandler()
myDbHandler.connect()
self.filtered_df.to_sql("title_ratings",
myDbHandler.conn,
if_exists='append',
index=False)
return
def __init__(self):
myDbHandler = DbHandler()
keywords_list = myDbHandler.exec_select_sql_from_file(
os.path.join(DATA_PATH, 'sql/select_genres_keywords.sql'))
self.keywords_df = pd.DataFrame(
keywords_list, columns=['tconst', 'genres', 'keywords'])
print(self.keywords_df.head())
actors_list = myDbHandler.exec_select_sql_from_file(
os.path.join(DATA_PATH, 'sql/select_actors.sql'))
self.actors_df = pd.DataFrame(actors_list,
columns=['tconst', 'PrimaryName'])
print(self.actors_df.head())
self.soup_df = pd.DataFrame(columns=['tconst', 'soup'])
def filter_foreign_keys(self):
"""
Keep only the tconst ids that are in the title_basics table
:return:
"""
myDbHandler = DbHandler()
myDbHandler.connect()
tconst_ids = [
row["tconst"] for row in myDbHandler.conn.execute(
text("SELECT tconst FROM title_basics"))
]
self.filtered_df = self.filtered_df[self.filtered_df.tconst.isin(
tconst_ids)]
return
def filter_foreign_keys(self):
myDbHandler = DbHandler()
myDbHandler.connect()
tconst_ids = [
row["tconst"] for row in myDbHandler.conn.execute(
text("SELECT tconst FROM title_basics"))
]
self.filtered_df = self.filtered_df[self.filtered_df.tconst.isin(
tconst_ids)]
self.filtered_df = self.filtered_df.groupby(['tconst', 'nconst'
]).size().reset_index()
columns_to_keep = ['tconst', 'nconst']
self.filtered_df = self.filtered_df[columns_to_keep]
return
def insert_to_db(self) -> None:
"""
Assuming the group_actors function is already called and the self.soup_df df is created, insert it
in the title_soup table
:return:
"""
myDbHandler = DbHandler()
myDbHandler.connect()
self.soup_df.to_sql("title_soup",
myDbHandler.conn,
if_exists='append',
index=False)
print("Successfully inserted values in the db")
return
def get_titles_from_tconst_list(tconst_list: list) -> list:
"""
:param tconst_list: list of tconst ids, normally generated via get_recommendation_from_tconst
:return: list of tuples, where first field is the tconst and second field is the title from the title_basics table
"""
dbhandler = DbHandler()
dbhandler.connect()
all_titles = [
row for row in dbhandler.conn.execute(
sql_text(f"SELECT tconst, primaryTitle FROM "
f"title_basics NATURAL JOIN title_soup"))
]
all_titles_df = pd.DataFrame(data=all_titles,
columns=['tconst', 'primaryTitle'])
return [(i, all_titles_df.loc[all_titles_df['tconst'] == i]
['primaryTitle'].values[0]) for i in tconst_list]
def index():
enhanced_movies = []
if request.method == "GET":
msg = "Provide a title to search for a movie"
if request.method == "POST":
movie_param = request.form["title"]
msg = f"Results for '{movie_param}'"
if movie_param.strip(): # if movie_param is not whitespace
dbhandler = DbHandler()
sql_cmd = f"select tconst, startYear,primaryTitle, genres FROM title_basics " \
f"natural join title_keywords WHERE primaryTitle LIKE '%{movie_param}%' LIMIT 20"
movies = dbhandler.exec_sql_cmd(sql_cmd=sql_cmd)
for movie in movies:
movie_dict = {field: movie[field] for field in movie._fields}
movie_dict[
'imdburl'] = f"https://www.imdb.com/title/{movie['tconst']}/"
enhanced_movies.append(movie_dict)
if not enhanced_movies:
msg = "No results where found, please try again"
return render_template("index.html",
movies=enhanced_movies,
result_message=msg)
def test_handler_init(self):
handler = DbHandler()
assert handler.host is not None and handler.password is not None and handler.username is not None \
and handler.db is not None
class Recommender:
def __init__(self):
self.dbhandler = DbHandler()
self.dbhandler.connect()
data = self.dbhandler.exec_select_sql_from_file(
os.path.join(DATA_PATH, "sql/select_soup.sql"))
self.df = pd.DataFrame(data=data, columns=['tconst', 'soup'])
self.cosine_sim = None
self.cosine_sim_csv_path = os.path.join(DATA_PATH,
'cosine_sim/cosine_sim.csv')
return
def create_cosine_sim(self) -> None:
"""
Creates the cosine_sim matrix from the self.df dataframe
This is the method that does all the raw calculation, and this only needs to be done
when the dataset changes or the cosine_csv.csv file is missing for some reason
Otherwise it is better to just import the cosine_csv.csv file that is generated by an earlier
use of this method
:return:
"""
cv = CountVectorizer()
count_matrix = cv.fit_transform(self.df['soup'])
cosine_sim = cosine_similarity(count_matrix)
tconst_array = self.df['tconst'].values
cosine_sim_df = pd.DataFrame(data=cosine_sim)
cosine_sim_df.insert(0, column='tconst', value=tconst_array)
self.cosine_sim = cosine_sim_df
return
def export_cosine_sim_to_pkl(self,
pkl_path=os.path.join(
DATA_PATH, 'cosine_sim/cosine_sim.pkl')):
"""
Exports the self.cosine_sim dataframe to the specified path for later imports
:param pkl_path: the path of the pkl file to export the dataframe
:return:
"""
# Initialize the self.cosine_sim df
self.create_cosine_sim()
print(self.cosine_sim.head())
print("Starting to write...")
start_time = time.time()
self.cosine_sim.to_pickle(pkl_path)
print("Finished writing...")
end_time = time.time()
print(f"Elapsed {end_time - start_time} s")
return
def import_cosine_sim_from_pkl(self,
pkl_path=os.path.join(
DATA_PATH, 'cosine_sim/cosine_sim.pkl'),
auto_create=False):
if not os.path.exists(pkl_path):
if not auto_create: # Throw exception, we shouldn't create the file that doesn't exist
print(f"Can't find file with path {pkl_path}, exiting")
raise FileNotFoundError(errno.ENOENT,
os.strerror(errno.ENOENT), pkl_path)
else: # we should call export_cosine_sim_to_pkl to create the file
self.create_cosine_sim()
self.export_cosine_sim_to_pkl(pkl_path=pkl_path)
self.cosine_sim = pd.read_pickle(pkl_path)
return
def get_tconst_from_idx(self, idx: int) -> str:
return self.df[self.df.index == idx]["tconst"].values[0]
def get_index_from_tconst(self, tconst) -> int:
"""
Returns the index of the row the tconst param corresponds to in the self.df dataframe
Throws an Exception if the tconst value is not found
:param tconst: str the tconst to look for in the df
:return: int the index of the row found
"""
if not (tconst in self.cosine_sim['tconst'].values):
raise Exception(
f"tconst can't be found in tconst values of self.df, {tconst}")
return self.cosine_sim[self.cosine_sim['tconst'] ==
tconst].index.values[0]
def get_recommendation_from_tconst(self, tconst: str, limit=10) -> list:
"""
Assumes the self.cosine_sim is set (imported via import_cosine_sim_from_csv or generated via
create_cosine_sim)
:param tconst: str the tconst of the value we are looking for
:param limit: int optional the number of results to be included, default is 10
:return: sorted_tconst: list of tconst id recommendations in a sorted order, starting from most similar
"""
tconst_idx = self.get_index_from_tconst(
tconst) # get the index of the movie
# print(tconst_idx)
movie_recommendations = list(enumerate(self.cosine_sim[tconst_idx]))
# sort the values by the similarity in desc order
sorted_movie_recommendations = sorted(movie_recommendations,
key=lambda x: x[1],
reverse=True)
sorted_tconst = [
self.get_tconst_from_idx(movie[0])
for movie in sorted_movie_recommendations[:limit]
]
# print(sorted_tconst)
return sorted_tconst
def get_recommendation_titles_from_tconst(self, tconst: str, limit=10):
return get_titles_from_tconst_list(
self.get_recommendation_from_tconst(tconst, limit))
def test_handler_connection(self):
handler = DbHandler()
handler.connect()
assert handler.conn is not None