config.yml to reproduce the experiments in the paper

[Hayashi-Yudai]

I heard your excellent presentation at RecSys the other day and would like to try to reproduce the ANS experiment on my own.

As a start, I have been trying to reproduce the results presented in the paper using the Gowalla dataset used in the paper, but have not been successful. If you don't mind, could you please tell me what config.yml you use for your experiments during training?

I am trying a LightGCN based implementation and am using a config.yml like this

epochs: 100
stopping_step: 100
embedding_size: 64
n_layers: 3
eps: 0.1
gamma: 0.1
reg_weight: 0.0001
require_pow: False

load_col:
    inter: [user_id, item_id]
eval_args:
  group_by: user
  split: {'RS': [0.8,0.1,0.1]}
  mode: full

learner: adam
learning_rate: 0.001
train_batch_size: 2048
valid_batch_size: 2048
eval_batch_size: 2048
train_neg_sample_args:
  candidate_num: 32
metrics: ['Recall', 'MRR', 'NDCG', 'Hit', 'Precision']
topk: 20
valid_metric: Recall@20

reference information

Recall@20 red line: LightGCN blue line: LightGCN w/ ANS

[Asa9aoTK]

Thanks for your question, I'm sorry I didn't attach the complete config and log files. Because we haven't sorted out the relevant data yet, I can't reply to you directly on github. My email is asa9ao@hrbeu.edu.cn. Can you send an email to my email address? I will reply to your email with the complete training setup.