runtime error

[chengqb0126]

Hi,dear professor!First,i run the process_data.py,it is no problem.Second,i run the main.py to train the model,it is also no problem.But the last,i run the main.py to test the model,it is wrong.I need your help!Thank you very much!

[pengqianhan]

Thanks for reading our paper. The run_trainval(model, pra_traindata_path=traindata_path, pra_testdata_path='/mnt/DataSets/phan635_1/stfusion-dataset/test_data.pkl') is including training and validation. The test result in our paper is from the validation, because the final test only can be done on the official website, as said by the author of grip++ in "https://github.com/xincoder/GRIP/issues/26". I hope this helps

[chengqb0126]

Thanks for reading our paper. The run_trainval(model, pra_traindata_path=traindata_path, pra_testdata_path='/mnt/DataSets/phan635_1/stfusion-dataset/test_data.pkl') is including training and validation. The test result in our paper is from the validation, because the final test only can be done on the official website, as said by the author of grip++ in "xincoder/GRIP#26". I hope this helps感谢您阅读我们的论文。run_trainval（model，pra_traindata_path=traindata_path，pra_testdata_path ='/mnt/DataSets/phan635_1/stfusion-dataset/test_data. pkl'）包括训练和验证。我们论文中的测试结果来自验证，因为最终的测试只能在官方网站上进行，正如grip++的作者在“ xincoder/GRIP#26 ”中所说。我希望这能帮助

Thank you for your reply!Are you comparing the results of the STF model validation set with those of Grip++and other benchmark models?The official website of Apolloscape has already released the actual test results. I would like to use your model to obtain a predicted test result and evaluate it with the actual results. However, running a predicted test result will result in an error and cannot generate a prediction. txt like Grip++. Your idea is great, and I think it will be as good as your paper suggests in long-term predictions, but I want to implement it.

[pengqianhan]

I'm glad to hear that you're interested in the spatial temporal fusion idea. However, I must admit that the method I used to implement the idea is not very efficient. The dataset processed by data_process.py is quite large, and the training stage requires a significant amount of memory. There is still a lot of work to be done to improve the code. Please feel free to contribute your own code to the repository.

[chengqb0126]

ok!Thank your for your help.

[chengqb0126]

I'm glad to hear that you're interested in the spatial temporal fusion idea. However, I must admit that the method I used to implement the idea is not very efficient. The dataset processed by data_process.py is quite large, and the training stage requires a significant amount of memory. There is still a lot of work to be done to improve the code. Please feel free to contribute your own code to the repository.很高兴你对时空融合的想法感兴趣。但是，我必须承认，我用来实现这个想法的方法不是很有效。data_process.py处理的数据集非常大，训练阶段需要大量内存。还有很多工作要做，以改善代码。请随意将您自己的代码贡献到存储库中。

Generally speaking, the constructed graphs are one by one, and each graph is connected using time series. Your idea is to build a large graph, but this situation leads to a very large amount of data in the constructed graph, almost increasing in geometric form. I think your idea is ahead of time, which impressed me. It is great, but I don't know if this idea is logical. Please forgive my lack of knowledge reserves!