In [109]:
import sys
sys.path.append("../")
In [110]:
import numpy as np
import pandas as pd
In [ ]:
from keras.models import Model
from keras.applications.vgg16 import VGG16
from keras.applications.inception_v3 import InceptionV3
from keras.applications.resnet50 import ResNet50
from keras.models import Model
from os import listdir
from keras.preprocessing.image import load_img, img_to_array
In [111]:
base_model = VGG16(weights="imagenet", include_top=False)
In [125]:
base_model.summary()
In [113]:
def preprocess_image(x):
x /= 255.
x -= 0.5
x *= 2.
return x
In [114]:
images = []
for name in listdir("/tf/notebooks/data/data/posters/"):
images.append(name)
In [115]:
selected_images = images[1000:1500] # Let's learn features on a subset of the images
In [116]:
image = load_img("/tf/notebooks/data/data/posters/"+selected_images[0], target_size=(299,299))
In [117]:
image = img_to_array(image)
In [118]:
image = np.expand_dims(image, axis=0)
In [119]:
image = preprocess_image(image)
In [120]:
feature = base_model.predict(image).ravel()
In [121]:
feature
Out[121]:
In [ ]:
In [122]:
feature.shape
Out[122]:
In [123]:
def load_photos_predict(directory):
images = []
for name in selected_images:
filename = directory + '/' + name
image = load_img(filename, target_size=(299, 299))
# convert the image pixels to a numpy array
image = img_to_array(image)
# reshape data for the model
image = np.expand_dims(image, axis=0)
# prepare the image for the model
image = preprocess_image(image)
feature = base_model.predict(image).ravel()
images.append(feature)
return images
In [124]:
%%time
poster_features = load_photos_predict("/tf/notebooks/data/data/posters/")
Nearest Neighbor¶
In [126]:
from reco.recommend import get_similar
from sklearn.neighbors import NearestNeighbors, VALID_METRICS
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import PIL
from sklearn.neighbors import NearestNeighbors
from sklearn.metrics.pairwise import cosine_distances, cosine_similarity
In [127]:
def get_similar(embedding, k):
model_similar_items = NearestNeighbors(n_neighbors=k, metric="cosine",
algorithm="brute").fit(embedding)
distances, indices = model_similar_items.kneighbors(embedding)
return distances, indices
In [128]:
%%time
item_distances, item_similar_indices = get_similar(poster_features[:100], 5)
In [129]:
item_similar_indices[:5]
Out[129]:
In [130]:
def show_similar(item_index, item_similar_indices):
movie_ids = item_similar_indices[item_index]
#movie_ids = item_encoder.inverse_transform(s)
images = []
for movie_id in movie_ids:
img_path = '/tf/notebooks/data/data/posters/' + str(movie_id+1) + '.jpg'
images.append(mpimg.imread(img_path))
plt.figure(figsize=(20,10))
columns = 5
for i, image in enumerate(images):
plt.subplot(len(images) / columns + 1, columns, i + 1)
plt.axis('off')
plt.imshow(image)
In [132]:
show_similar(2, item_similar_indices)
Text Features¶
In [134]:
item_features = pd.read_csv("/tf/notebooks/data/data/item_features.csv")
In [135]:
item_features.head()
Out[135]:
In [137]:
item_features["overview"][0]
Out[137]:
In [141]:
item_features.overview.isna().sum()
Out[141]:
In [142]:
# fill the missing value with some dummy text
In [143]:
item_features.overview.fillna("Missing value", inplace=True)
Get the sentence vector¶
In [144]:
import spacy
In [145]:
nlp = spacy.load("en_core_web_lg")
In [146]:
doc = nlp(item_features["overview"][0])
In [149]:
doc.vector.shape
Out[149]:
In [151]:
def word_vec(sentence):
doc = nlp(sentence)
return doc.vector
In [152]:
%%time
overview_embedding = item_features["overview"].apply(word_vec)
In [155]:
overview_embedding[:2]
Out[155]:
In [156]:
# Converting overview_embedding into a pandas dataframe
In [157]:
overview_embedding_list = []
for vec in overview_embedding:
overview_embedding_list.append(vec.tolist())
overview_embedding_df = pd.DataFrame(overview_embedding_list)
overview_embedding_df.head()
Out[157]:
In [158]:
%%time
item_distances, item_similar_indices = get_similar(overview_embedding_df, 5)
In [159]:
item_similar_indices
Out[159]:
In [161]:
show_similar(0, item_similar_indices)
In [ ]: