def setup_dataset(self):
"""
Creating train and test split.
"""
self.positive_edges, self.test_positive_edges = train_test_split(self.edges["positive_edges"],
test_size=self.args.test_size)
self.negative_edges, self.test_negative_edges = train_test_split(self.edges["negative_edges"],
test_size=self.args.test_size)
self.ecount = len(self.positive_edges + self.negative_edges)
self.X = setup_features(self.args,
self.positive_edges,
self.negative_edges,
self.edges["ncount"])
self.positive_edges = torch.from_numpy(np.array(self.positive_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
self.negative_edges = torch.from_numpy(np.array(self.negative_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
self.y = np.array([0 if i < int(self.ecount/2) else 1 for i in range(self.ecount)]+[2]*(self.ecount*2))
self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(self.device)
self.X = torch.from_numpy(self.X).float().to(self.device)
def setup_dataset(self):
"""
Creating train and test split.
"""
# self.positive_edges, self.test_positive_edges = train_test_split(self.edges["positive_edges"], test_size = self.args.test_size)
# self.negative_edges, self.test_negative_edges = train_test_split(self.edges["negative_edges"], test_size = self.args.test_size)
self.positive_edges = self.edges["positive_edges"]
self.negative_edges = self.edges["negative_edges"]
self.ecount = len(self.positive_edges + self.negative_edges)
node_count = self.node_count
self.X = setup_features(self.args, self.positive_edges,
self.negative_edges, node_count)
self.positive_edges = torch.from_numpy(
np.array(self.positive_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
self.negative_edges = torch.from_numpy(
np.array(self.negative_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
# self.y = np.array([0 if i< int(self.ecount/2) else 1 for i in range(self.ecount)] +[2]*(self.ecount*2))
# self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(self.device)
self.y = np.array(
[1 if label == -1 else 0 for label in self.node_labels])
self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(
self.device)
if self.args.subgraph_training == True:
self.X = self.X[
self.nodes_dict['subgraph_map_from_original_feature']]
self.X = torch.from_numpy(self.X).float().to(self.device)
def setup_dataset(self):
"""
Creating train and test split.
"""
self.positive_edges, self.test_positive_edges = train_test_split(self.edges["positive_edges"], # postive的点边关系
test_size=self.args.test_size)
self.negative_edges, self.test_negative_edges = train_test_split(self.edges["negative_edges"],
test_size=self.args.test_size)
self.ecount = len(self.positive_edges + self.negative_edges) # 训练样本数量
# SVD分解，邻接矩阵，为节点特征;
self.X = setup_features(self.args,
self.positive_edges, # 训练样本
self.negative_edges,
self.edges["ncount"]) # 点的数量
# positive样本的点边关系转换成torch类型
self.positive_edges = torch.from_numpy(np.array(self.positive_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
# negative样本的点边关系
self.negative_edges = torch.from_numpy(np.array(self.negative_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
# label: [0],[1] 两个部分; [2]是两倍的点边数量-表示不连接的边; !! np.array([0]*len(self.positive_edges) + [1]*len(self.negative_edges) + [2]*(self.ecount*2))
self.y = np.array([0]*len(self.positive_edges[0]) + [1]*len(self.negative_edges[0]) + [2]*(self.ecount*2)) #np.array([0 if i < int(self.ecount/2) else 1 for i in range(self.ecount)]+[2]*(self.ecount*2))
self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(self.device)
self.X = torch.from_numpy(self.X).float().to(self.device) # [5881,64]的input features
def setup_dataset(self):
"""
Creating train and test split.
"""
# self.positive_edges, self.test_positive_edges = train_test_split(self.edges["positive_edges"], test_size = self.args.test_size)
# self.negative_edges, self.test_negative_edges = train_test_split(self.edges["negative_edges"], test_size = self.args.test_size)
self.positive_edges = self.edges["positive_edges"]
self.negative_edges = self.edges["negative_edges"]
self.ecount = len(self.positive_edges + self.negative_edges)
node_count = self.node_count # node_countを補正
self.X = setup_features(self.args, self.positive_edges,
self.negative_edges, node_count)
self.positive_edges = torch.from_numpy(
np.array(self.positive_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
self.negative_edges = torch.from_numpy(
np.array(self.negative_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
# self.y = np.array([0 if i< int(self.ecount/2) else 1 for i in range(self.ecount)] +[2]*(self.ecount*2))
# self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(self.device)
# self.y : ノードのラベル Fraudが正例, Benignが負例
self.y = np.array(
[1 if label == -1 else 0 for label in self.node_labels])
self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(
self.device)
self.X = torch.from_numpy(self.X).float().to(self.device)
def setup_dataset(self):
"""
Creating train and test split.
"""
def train_test_label_split(la_data):
"""
Filter according to the label
"""
tr_pos = la_data[la_data['dataset_type'].isin(['train'])]
te_pos = la_data[la_data['dataset_type'].isin(['test', 'valid'])]
del tr_pos['dataset_type']
del te_pos['dataset_type']
return tr_pos, te_pos
self.positive_edges, self.test_positive_edges = self.edges[
"positive_edges"]['train'], self.edges["positive_edges"]['test']
self.negative_edges, self.test_negative_edges = self.edges[
"negative_edges"]['train'], self.edges["negative_edges"]['test']
self.ecount = len(self.positive_edges + self.negative_edges)
self.X = setup_features(self.args, self.positive_edges,
self.negative_edges, self.edges["ncount"])
self.positive_edges = torch.from_numpy(
np.array(self.positive_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
self.negative_edges = torch.from_numpy(
np.array(self.negative_edges,
dtype=np.int64).T).type(torch.long).to(self.device)
self.y = np.array([0] * (self.positive_edges.shape[1]) + [1] *
(self.negative_edges.shape[1]) + [2] * 2 *
((self.positive_edges.shape[1]) +
(self.negative_edges.shape[1])))
self.y = torch.from_numpy(self.y).type(torch.LongTensor).to(
self.device)
self.X = torch.from_numpy(self.X).float().to(self.device)