def import_imdb_weighted_graph(self, weights):
weights_dict = {'movies_edges': weights[0], 'labels_edges': weights[1]}
dict_paths = {'cast': 'data_set/IMDb title_principals.csv', 'genre': 'data_set/IMDb movies.csv'}
imdb = MoviesGraph(dict_paths, self.args.graph_percentage)
gnx = imdb.create_graph()
labels = imdb.labels2int(gnx)
knowledge_gnx, knowledge_data = imdb.create_knowledge_graph(labels, float(self.kg_jacard_similarity_threshold))
weighted_graph = imdb.weighted_graph(gnx, knowledge_gnx, labels, weights_dict)
return weighted_graph
def import_imdb_weighted_graph(weights):
from IMDb_data_preparation_E2V import MoviesGraph
weights_dict = {'movies_edges': weights[0], 'labels_edges': weights[1]}
dict_paths = {
'cast': 'data_set/IMDb title_principals.csv',
'genre': 'data_set/IMDb movies.csv'
}
imdb = MoviesGraph(dict_paths)
gnx = imdb.create_graph()
labels = imdb.labels2int(gnx)
knowledge_gnx, knowledge_data = imdb.create_knowledge_graph(labels)
weighted_graph = imdb.weighted_graph(gnx, knowledge_gnx, labels,
weights_dict)
return weighted_graph
def import_imdb_multi_graph(self, weights):
"""
Make our_imdb multi graph using class
:param weights:
:return:
"""
weights_dict = {'movies_edges': weights[0], 'labels_edges': weights[1]}
dict_paths = {'cast': 'data_set/IMDb title_principals.csv', 'genre': 'data_set/IMDb movies.csv'}
imdb = MoviesGraph(dict_paths, self.args.graph_percentage)
gnx = imdb.create_graph()
labels = imdb.labels2int(gnx)
knowledge_gnx, knowledge_data = imdb.create_knowledge_graph(labels, self.kg_jacard_similarity_threshold)
multi_gnx = imdb.weighted_multi_graph(gnx, knowledge_gnx, labels, weights_dict)
return multi_gnx
def import_imdb_multi_graph(weights):
"""
Make our_imdb multi graph using class
:param weights:
:return:
"""
from IMDb_data_preparation_E2V import MoviesGraph
weights_dict = {'movies_edges': weights[0], 'labels_edges': weights[1]}
dict_paths = {
'cast': 'data_set/IMDb title_principals.csv',
'genre': 'data_set/IMDb movies.csv'
}
imdb = MoviesGraph(dict_paths)
gnx = imdb.create_graph()
labels = imdb.labels2int(gnx)
knowledge_gnx, knowledge_data = imdb.create_knowledge_graph(labels)
multi_gnx = imdb.weighted_multi_graph(gnx, knowledge_gnx, labels,
weights_dict)
return multi_gnx
def create_graph(self):
weights_dict = {
'classes_edges': self.class_edges_weights,
'labels_edges': self.labels_edges_weights
}
if self.args.dataset == "our_imdb":
weights_dict["movies_edges"] = weights_dict["classes_edges"]
dict_paths = {
'cast': 'data_set/IMDb title_principals.csv',
'genre': 'data_set/IMDb movies.csv'
}
imdb = MoviesGraph(dict_paths, self.args.images_nodes_percentage)
image_graph = imdb.create_graph()
labels = imdb.labels2int(image_graph)
labels_graph = imdb.create_labels_graph(labels)
kg, knowledge_data = imdb.create_knowledge_graph(
labels, float(0.3))
final_graph = imdb.weighted_graph(image_graph, kg, labels,
weights_dict)
dict_true_edges = self.label_edges_classes_ordered(
list(labels_graph.edges))
classes = list(dict_true_edges.keys())
for i, k in enumerate(
sorted(dict_true_edges,
key=lambda x: len(dict_true_edges[x]),
reverse=True)):
classes[i] = k
seen_classes = classes[:int(0.8 * len(classes))]
unseen_classes = classes[int(0.8 * len(classes)):]
elif self.args.dataset == "cub" or self.args.dataset == "lad":
dict_paths, radius = define_graph_args(self.args.dataset)
graph_preparation = ImagesEmbeddings(dict_paths, self.args)
embeds_matrix, dict_image_embed, dict_image_class = graph_preparation.images_embed_calculator(
)
dict_idx_image_class = {
i: dict_image_class[image]
for i, image in enumerate(list(dict_image_class.keys()))
}
final_graph_creator = FinalGraphCreator(
dict_paths, embeds_matrix, dict_image_embed,
dict_idx_image_class, self.args.images_nodes_percentage,
self.args)
image_graph = final_graph_creator.create_image_graph(radius)
kg, dict_class_nodes_translation = final_graph_creator.imagenet_knowledge_graph(
)
kg = final_graph_creator.attributed_graph(
kg, dict_class_nodes_translation, self.att_weight, radius)
seen_classes, unseen_classes = final_graph_creator.seen_classes, final_graph_creator.unseen_classes
seen_classes = [
dict_class_nodes_translation[c] for c in seen_classes
]
unseen_classes = [
dict_class_nodes_translation[c] for c in unseen_classes
]
labels_graph = final_graph_creator.create_labels_graph(
dict_class_nodes_translation)
final_graph = final_graph_creator.weighted_graph(
image_graph, kg, labels_graph, weights_dict)
else:
raise ValueError("Wrong dataset name: replace with awa2/cub/lad")
dict_class_identification = {
**{c: 0
for c in seen_classes},
**{c: 1
for c in unseen_classes}
}
print("image graph edges & nodes: ", len(image_graph.edges), "&",
len(image_graph.nodes))
print("knowledge graph edges & nodes: ", len(kg.edges), "&",
len(kg.nodes))
print("labels graph edges & nodes: ", len(labels_graph.edges), "&",
len(labels_graph.nodes))
print("final graph edges & nodes: ", len(final_graph.edges), "&",
len(final_graph.nodes))
return final_graph, labels_graph, dict_class_identification