/BGCL

source code of BGCL

Primary LanguagePython

Files in the folder

BGCL/
├── datasets/
│   └── wsdrean/
│   	├── cold: sample data for cold start 
│       │   ├── rt: data of response time
│       │   └── tp: data of response throughput
│       └── sparsity: sample data for sparsity    
│           ├── rt: data of response time
│           └── tp: data of response throughput
├── utils/
│   ├── attention.py: aggregating the feature of neighbors
│   ├── batchutils.py: some utils
│   ├── combiner.py: implementation of the combiner
│   └── encoder.py: together with aggregator to form the decomposer
│
├── bgcl.py: implementation of the bgcl
├── run.py: run the model
├── train.py: train the model
├── test.py: test the model
└── README.md

Environment Settings

  • Python == 3.6.9
  • torchvision == 0.4.2
  • numpy == 1.17.3
  • scikit-learn == 0.21.3

Parameter Settings

  • epochs: the number of epochs to train
  • test_epochs: the number of epochs to test
  • lr: learning rate
  • optimizer: the optimizer
  • momentum: the momentum of optimizer
  • lr: learning rate
  • embed_dim: embedding dimension
  • droprate: dropout rate in layer
  • temp: the temp
  • batch_size: batch size for training
  • test_batch_size: batch size for testing
  • n_size: the size of neighbor
  • dataset: the prefix of dataset
  • density: the density of dataset
  • cold_start_density: the density of cold_start
  • p: the dropout rate
  • a: the similiarity setting
  • mode: the mode of experiment (sparsity or cold)
  • coldStartType: the cold start type
  • cold_start_density: the density of cold_start

Basic Usage

python run.py

Example of running

unzip the dataset first

example: unzip the rt_user_allData_0.8_0.7_0.500_0.005.7z to get rt_user_allData_0.8_0.7_0.500_0.005.p for running

Dataset: wsdream
Density: 0.005
-------------------- Hyperparams --------------------
dropout rate: 0.5
dimension of embedding: 128
type of optimizer: Adam
learning rate: 0.001
p: 0.8
a: 0.7
datatype: rt
mode: cold
coldStartType: user
cold_start_density: 0.500

...

2022-12-06 21:10:51.528380 Training: [200 epoch,  50 batch] loss: 4.67270, the best RMSE/MAE: 1.85315 / 0.71128
2022-12-06 21:10:52.077927 Training: [200 epoch,  60 batch] loss: 4.13545, the best RMSE/MAE: 1.85315 / 0.71128
2022-12-06 21:10:52.606055 Training: [200 epoch,  70 batch] loss: 4.26554, the best RMSE/MAE: 1.85315 / 0.71128
<Test> RMSE: 1.86849, MAE: 0.71469 
The best RMSE/MAE: 1.85315 / 0.71128