Causal Representation Learning for Out-of-Distribution Recommendation

This is the pytorch implementation of our paper at WWW 2022:

Causal Representation Learning for Out-of-Distribution Recommendation

Wenjie Wang, Xinyu Lin, Fuli Feng, Xiangnan He, Min Lin, Tat-Seng Chua

Environment

• Anaconda 3
• python 3.7.3
• pytorch 1.4.0
• numpy 1.16.4

Usage

Data

The experimental data are in './data' folder, including Synthetic Data, Meituan and Yelp. Due to the large size, 'item_feature.npy' of Yelp is uploaded to Google drive.

Training

python main.py --model_name=$1 --dataset=$2 --mlp_dims=$3 --mlp_p1_1_dims=$4 --mlp_p1_2_dims=$5 --mlp_p2_dims=$6 --mlp_p3_dims=$7 --lr=$8 --wd=$9 --batch_size=$10 --epochs=$11 --total_anneal_steps=$12 --anneal_cap=$13 --CI=$14 --dropout=$15 --Z1_hidden_size=$16 --E2_hidden_size=$17 --Z2_hidden_size=$18 --bn=$19 --sample_freq=$20 --regs=$21 --act_function=$22 --log_name=$23 --gpu=$24 --cuda

or use run.sh

sh run.sh model_name dataset mlp_dims mlp_p1_1_dims mlp_p1_2_dims mlp_p2_dims mlp_p3_dims lr wd batch_size epochs total_anneal_steps anneal_cap CI dropout Z1_hidden_size E2_hidden_size Z2_hidden_size bn sample_freq regs act_function log_name gpu_id

• The log file will be in the './code/log/' folder.
• The explanation of hyper-parameters can be found in './code/main.py'.
• The default hyper-parameter settings are detailed in './code/hyper-parameters.txt'.

Inference

Get the results of COR over iid and ood data where only user features are drifted by running inference.py:

python inference.py --dataset=$1 --ckpt=$2 --cuda

Fine-tuneing

python main.py --model_name=$1 --dataset=$2 --X=$3 --lr=$4 --wd=$5 --batch_size=$6 --epochs=$7 --total_anneal_steps=$8 --anneal_cap=$9 --CI=$10 --dropout=$11 --bn=$12 --sample_freq=$13 --regs=$14 --log_name=$15 --ckpt=$16 --gpu=$17 --ood_finetune --cuda

or use finetune.sh

sh finetune.sh model_name dataset X lr wd batch_size epochs total_anneal_steps anneal_cap CI dropout bn sample_freq regs log_name <pre-trained model directory> gpu_id

• The log file will be in the './code/log/finetune/' folder.

Examples

1. Train COR on iid meituan:

cd ./code
sh run.sh COR meituan [3000] [] [1] [] [] 1e-3 0 500 300 0 0.1 1 0.5 500 1000 200 0 1 0 tanh log 0

2. Inference on synthetic data:

cd ./code
python inference.py --dataset synthetic --ckpt <pre-trained model directory> --cuda

3. Fine-tuning COR on ood yelp:

cd ./code
sh finetune.sh COR yelp 0 0.0001 0.03 500 0 0.5 1 0.4 0 1 0 log <pre-trained model directory> 0

License

NUS © NExT++