Unable to reproduce results for SACN model

JD-AI-Research-Silicon-Valley / SACN

End-to-end Structure-Aware Convolutional Networks for Knowledge Base Completion

MIT License

112 stars 30 forks source link

Unable to reproduce results for SACN model #5

Closed svjan5 closed 5 years ago

svjan5 commented 5 years ago

Hi, Using the default hyperparameters, I am unable to reproduce MRR score .33 for ConvTransE model from the given code. Currently, I am getting around .295 on the test set. Can you share the hyperparameters you used for training ConvTransE?

Thanks in advance

chaoshangcs commented 5 years ago

Hi, thank you so much for your question. The model is sensitive to the parameters. Here are the basic hyperparameters for your reference. Thanks! If you have any problems, please email me. Thanks!

CUDA_VISIBLE_DEVICES=0 python main.py model ConvTransE init_emb_size 100 dropout_rate 0.4 channels 50 lr 0.001 dataset FB15k-237 process True

svjan5 commented 5 years ago

Hi @chaoshangcs, I could reproduce ConvTransE results with my own implementation. However, I could not reproduce SACN results.