Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
Theheavens 425c99032a | 1 year ago | |
---|---|---|
.. | ||
Reproduce_GPRGNN.sh | 1 year ago | |
gprgnn_trainer.py | 1 year ago | |
process.py | 2 years ago | |
readme.md | 1 year ago | |
test_accuracy | 2 years ago |
Run with following (available dataset: "cora", "citeseer", "pubmed", "computers", "photo", "squirrel", "chameleon", "cornell", "texas")
sh Reproduce_GPRGNN.sh
Cora | Citeseer | PubMed | Computers | Photo | Chameleon | Squirrel | Texas | Cornell | |
---|---|---|---|---|---|---|---|---|---|
Tensorflow | 79.08 | 65.96 | 83.74 | 83.44 | 92.06 | 67.61 | 51.82 | 88.02 | 85.25 |
Paddle | |||||||||
Origin | 79.51 | 67.63 | 85.07 | 82.90 | 91.93 | 67.48 | 49.93 | 92.92 | 91.36 |
GPRConv
In file layers/conv/gpr_conv.py
The basic structure is similar to gcn_conv.py
__init__
method: Add the initialization for learnt weights $\gamma_k$ and define the steps of propagation $K$
reset_parameters
method: Use $\textbf{PPR}$ initializaiton method to reset parametersGPRGNNModel
In file models/gprgnn.py
__init__
method: besides basic dimension information of node features and num of classes, GPR related hyperparameters ($K$, $\textbf{Init}$, $\alpha$, $\text{dprate}$, etc.) are also transfered to the networkforward
propagation:
GPRConv
NormalizeFeatures
In file transforms/normalize_features.py
In file examples/gprgnn/gprgnn_trainer.py
random_planetoid_splits
is defined to randomly split dataset and ensure the numbers of nodes from each classes are the same in training set.In file examples/gprgnn/Reproduce_GPRGNN.sh
To keep the same with experiments in origin paper,5 homophily graph datasets, Cora
, Citeseer
, PubMed
, Computers
and Photo
, adopt sparse split $(2.5%/2.5%/95%)$ for (training/validation/test), and 4 heterophily graph datasets, Chameleon
, Squirrel
, Texas
and Cornell
, adopt dense split $(60%/20%/20%)$ .
Train model with each hyperparameter setting for 10 times and calculate the average accuracy.
Best hyperparameters for different datasets are given in the file.
In file examples/gprgnn/test.accuracy
Cora | Citeseer | PubMed | Computers | Photo | Chameleon | Squirrel | Texas | Cornell | |
---|---|---|---|---|---|---|---|---|---|
Tensorflow | 79.08 | 65.96 | 83.74 | 83.44 | 92.06 | 67.61 | 51.82 | 88.02 | 85.25 |
Paddle | |||||||||
Origin | 79.51 | 67.63 | 85.07 | 82.90 | 91.93 | 67.48 | 49.93 | 92.92 | 91.36 |
Texas
and Cornell
is unstable, fluctuating from 75 ~ 95.No Description
Python Cuda C++ Cython Markdown other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》