BRIDGE User Guide

1. Requirements

Main Dependency Requirements:

Below are the main dependencies and their version requirements for the project:

• torch == 1.10.1+cu113
• torch-geometric == 2.1.0.post1
• torch-cluster == 1.5.9
• torch-scatter == 2.0.9
• torch-sparse == 0.6.12
• torch-spline-conv == 1.2.1
• scikit-learn == 1.3.2
• numpy == 1.24.4
• pandas == 2.0.3
• matplotlib == 3.7.5
• scipy == 1.10.1
• datasets == 3.1.0
• huggingface-hub == 0.27.1
• wandb == 0.18.7
• tqdm == 4.67.0
• filelock == 3.16.1
• protobuf == 5.28.3
• requests == 2.32.3
• typing_extensions == 4.12.2
• attrs == 24.3.0

To install the complete requiring packages, use the following command at the root directory of the repository:

pip install -r requirements.txt

Dataset download:

https://huggingface.co/datasets/aboutime233/BRIDGE-data/tree/main

2. Quick Start

Unzip the downloaded dataset archive into a folder named data in the root directory of the repository.

Pretraining + Fine-tuning

Navigate to the model-node/scripts directory and execute the following command:

python main.py --dataset Cora --shot_num 1

Fine-tuning

In the same directory as main.py, execute the following command:

python main.py --dataset Cora --model_path pretrain_model.pkl

Parameter Explanation:
--dataset Select the dataset to use.
--shot_num Specify the dataset type for fine-tuning, either 1shot or 5shot.
--model_path The path to the pretrained model parameters saved during the pretraining phase.

3. Results reproduction:

You can reproduce our results by executing the following commands in the scripts directory under the corresponding task type.

Node Classification

1-shot

python main.py --dataset Cora

The dataset parameter can be set from the following Six datasets: Cora, Citeseer, Computers, Pubmed, Reddit, and Photo. You also need to add parameters as shown below:

• --lr == 0.00008094590967608754
• --l2_coef == 0.00004404197581665391
• --hid_units == 256
• --lambda_entropy == 0.20401015296835048
• --dropout_rate == 0.1913510180577923
• --variance_weight == 1521434.9368374627
• --downstreamlr == 0.000962404050084371
• --reg_weight == 1
• --reg_thres == 0.4
• --shot_num == 1 When the dataset is Reddit, the learning rate lr should be set to 0.00001.

5-shot

python main.py --dataset Cora

The dataset parameter can be set from the following Six datasets: Cora, Citeseer, Computers, Pubmed, Reddit, and Photo. You also need to add parameters as shown below:

• --lr == 0.00001
• --l2_coef == 0.00004404197581665391
• --hid_units == 256
• --lambda_entropy == 0.20401015296835048
• --dropout_rate == 0.1913510180577923
• --variance_weight == 1521434.9368374627
• --downstreamlr == 0.000962404050084371
• --reg_weight == 1
• --reg_thres == 0.4
• --shot_num == 5

Graph Classification

1-shot

python main.py --dataset Cora

The dataset parameter can be set from the following Six datasets: Cora, Citeseer, Computers, Pubmed, Reddit, and Photo. You also need to add parameters as shown below:

• --lr == 0.00008094590967608754
• --l2_coef == 0.00004404197581665391
• --hid_units == 256
• --lambda_entropy == 0.07845469891338196
• --dropout_rate == 0.1913510180577923
• --variance_weight == 1521434.9368374627
• --downstreamlr == 0.001
• --reg_weight == 1
• --reg_thres == 0.4
• --shot_num == 1

5-shot

python main.py --dataset Cora

The dataset parameter can be set from the following Six datasets: Cora, Citeseer, Computers, Pubmed, Reddit, and Photo. You also need to add parameters as shown below:

• --lr == 0.00008094590967608754
• --l2_coef == 0.00004404197581665391
• --hid_units == 256
• --lambda_entropy == 0.07845469891338196
• --dropout_rate == 0.1913510180577923
• --variance_weight == 1521434.9368374627
• --downstreamlr == 0.001
• --reg_weight == 1
• --reg_thres == 0.4
• --shot_num == 5