xptree / deepinf Goto Github PK

Hey! I was interested into why are you using tanh here:

attn_src = torch.matmul(F.tanh(h_prime), self.a_src) # bs x n_head x n x 1

in BatchMultiHeadGraphAttention, get_layers.py. Did it stabilize the training? Is it some form of feature normalization?

Hello, I am really interested in your research.

May I ask the parameters setting of your code, which can get the same best results in your paper 'DeepInf: Social Influence Prediction with Deep Learning'. Especially on your twitter dataset.

Many thanks.

Thank you for your work. In your paper you mentioned that you constructed 6 vertex features, a 64d embedding feature, and 3 ego features. However, after downloading the datasets you provided, I found that in the vertex_feature.ny file, we got 7 features. What are these 7 features then?

Thank you

您好：
感谢您分享的代码，关于多头attention计算的部分有两处问题想请教，希望能得到您的回复:
1. Line84请问这一行的self.w是相当于paper中的参数c吗?
2. Line85,86,87中h_prime分别与self.a_src和self.a_dst相乘再相加，是对应paper中[Wh_j||Wh_i]的部分吗?

希望能得到您的回复，非常感谢！
祝好！

您好，请问您能提供数据集预处理的代码吗？我想尝试将您的方法运用在我自己的数据集上。感谢！

the number of sub-network formed by active neighbors / the number of sub-network formed by neighbors ?

Hello, I'm going to test your implementation using a custom dataset similar to Twitter. Do you have scripts to pre-process raw-data?

I have a question about the input data: I am wondering whether the last element of each row in the"vertex_id.npy" is the ID for ego node?

I am also curious whether you the data processing script is being released or not?