nycdsa_p5_cf.py

df_id = pd.read_csv('links.csv', sep=',')

idx_to_movie = {}
for row in df_id.itertuples():
    idx_to_movie[row[1]-1] = row[2]  
    
total_movies = 9000

movies = [0]*total_movies
for i in range(len(movies)):
    if i in idx_to_movie.keys() and len(str(idx_to_movie[i])) == 6:
        movies[i] = (idx_to_movie[i]) 
movies = filter(lambda imdb: imdb != 0, movies)
total_movies  = len(movies)

URL = [0]*total_movies 
IMDB = [0]*total_movies 
URL_IMDB = {"url":[],"imdb":[]}
i = 0
for movie in movies:
    (URL[i], IMDB[i]) = get_poster(movie, base_url)
    if URL[i] != base_url+"":
        URL_IMDB["url"].append(URL[i])
        URL_IMDB["imdb"].append(IMDB[i])
    i += 1 
# URL = filter(lambda url: url != base_url+"", URL)
df = pd.DataFrame(data=URL_IMDB) 

total_movies = len(df)  

import urllib

poster_path = "/Users/wannjiun/Desktop/nycdsa/project_5_recommender/posters/"
for i in range(total_movies):
    urllib.urlretrieve(df.url[i], poster_path + str(i) + ".jpg")
    
from keras.applications import VGG16
from keras.applications.vgg16 import preprocess_input
from keras.preprocessing import image as kimage

image = [0]*total_movies
x = [0]*total_movies
for i in range(total_movies):
    image[i] = kimage.load_img(poster_path + str(i) + ".jpg", target_size=(224, 224))
    x[i] = kimage.img_to_array(image[i])
    x[i] = np.expand_dims(x[i], axis=0)
    x[i] = preprocess_input(x[i])   
    
model = VGG16(include_top=False, weights='imagenet')

prediction = [0]*total_movies
matrix_res = np.zeros([total_movies,25088])
for i in range(total_movies):
    prediction[i] = model.predict(x[i]).ravel()
    matrix_res[i,:] = prediction[i] 
    
similarity_deep = matrix_res.dot(matrix_res.T)
norms = np.array([np.sqrt(np.diagonal(similarity_deep))])
similarity_deep = similarity_deep / norms / norms.T