TensorFlow implementation of Heterogeneous Multiple Mini-Graphs Neural Network
The overall architecture of our proposal is displayed here.
The left part illustrates kNN-based hyper-graphs generation given normal node features and their connected relation.
We concatenate the feature matrix of the generated hyper-nodes from different sub-graphs to the input feature matrix
to form the final feature representation that is fed into neural networks.
The prediction is made based on the learned hidden state learned from the middle procedure.
Different colors of nodes and edges indicate different types of nodes and relationships.
tensorflow (>=1.12)
pandas
numpy
python HMGNN.py
The data used in quick-start is the Cora dataset.
The Cora dataset consists of 2708 scientific publications classified into one of seven classes.
The Cora dataset has saved as .npy in dir ./data
The parameters are defined in hparam.py
. Main parameters conclude:
We compare our proposal, HMGNN, with GCN, one of the classic graph convolutional network based approach.
The training dataset is used to learn the model while the model selection
is based on the performance on the validation dataset.
The accuracy measure is considered. Our proposed method achieves the better performance.
HMGNN | GCN | |
---|---|---|
train_acc | 0.908 | 0.860 |
val_acc | 0.867 | 0.854 |
The table above is showed the accuracy of HMGNN and GCN. The pictures show the detailed loss and accuracy curve on training and validation dataset.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。
1. 开源生态
2. 协作、人、软件
3. 评估模型