def get_recommendations(self, request_data, top_k=None, min_count=10):
if not top_k:
top_k = 20
mapped = map(util.map_dict_value_as_array, request_data['ratings'])
reduced = reduce(util.merge_two_dicts, mapped)
nb_ratings = len(reduced['movieId'])
reduced['userId'] = [999999 for x in range(nb_ratings)]
reduced = gl.SFrame(reduced)
recommendations = self.model.recommend(users=[999999], k=5000, new_observation_data=reduced)
recommendations = self.dataset.movies.join(recommendations, on='movieId')
recommendations = recommendations[recommendations['ratingCount'] > min_count]
if 'filter' in request_data:
filtered = recommendations\
.flat_map(['genres', 'movieId'],
lambda x: [[g, x['movieId']] for g in x['genres'] for i in range(0, len(x['genres']))])\
.filter_by(request_data['filter'], 'genres')
recommendations = recommendations.filter_by(filtered['movieId'], 'movieId')
recommendations = recommendations.remove_column('userId')
recommendations = recommendations.sort('rank')
return util.sframe_to_list(recommendations, top_k)
def get_random(self, top_k):
if not top_k:
top_k = 3
sampled_movies = self.movies.sample(0.01)
# hack to convert SFrame to list of dict
return util.sframe_to_list(sampled_movies, top_k)
def find_top_rated(self, min_count=50, top_k=20):
if not top_k:
top_k = 3
top_rated = self.movies[self.movies['ratingCount'] > min_count].sort('ratingAvg', False)
# hack to convert SFrame to list of dict
return util.sframe_to_list(top_rated, top_k)
def search(self, query, top_k):
if not top_k:
top_k = 3
"""not sure whether this is a good idea .."""
if query is '':
query = 'somestringthatmatchesnothing'
query_re = '.*{}.*'.format(query)
p = re.compile(query_re, re.IGNORECASE)
found_movies = self.movies[self.movies['title'].apply(lambda t: 1 if p.match(t) else 0)]
# hack to convert SFrame to list of dict
return util.sframe_to_list(found_movies, top_k)
not_yet_enhanced_movies, dropped_movies = not_yet_enhanced_movies.dropna_split(columns="tmdbId")
# create client & view
rc = Client()
dv = rc[:] # use all engines
v = rc.load_balanced_view()
with dv.sync_imports():
import tmdbsimple
tmdbsimple.API_KEY = tmdbsimple._get_env_key("TMDB_API_KEY")
import unicodedata
for chunk in chunks(not_yet_enhanced_movies, 16):
print "chunk starting with movieId %d" % chunk[0]["movieId"]
chunk = util.sframe_to_list(chunk)
list_tmdb_info = dv.map_sync(tmdb_util.get_tmdb_info, chunk)
mapped_tmdb_info = map(util.map_dict_list, list_tmdb_info)
dict_tmdb_info = reduce(util.merge_two_dicts, mapped_tmdb_info)
if len(dict_tmdb_info) == 0:
break
sf_tmdb_info = gl.SFrame(dict_tmdb_info)
movie_enhanced = original.movies.join(sf_tmdb_info, on="movieId")
enhanced = enhanced.append(movie_enhanced)
enhanced.export_csv("datasets\\ml-latest-enhanced\\movies.csv")
time.sleep(3)
movies_have_year, movies_no_year = enhanced.dropna_split(columns="year")
for movie in movies_no_year:
info = tmdbsimple.Movies(movie["tmdbId"]).info()
year = dateutil.parser.parse(info["release_date"]).year