What is your electricity file modelData.csv comes from?

[hihihihiwsf]

Hello, I'm reproducing your experiment of electricity part. I got the dataset from ttps://archive.ics.uci.edu/ml/datasets/ElectricityLoadDiagrams20112014 , with the file type of txt. I'd like to learn about what content is in your file modelData.csv so that I can change the dataset to that frame. Thank you very much!

[oneday88]

Hi, glad to hear from you. I added an R script to the ec fold. You can easily generate the modelData.csv by running the R script "basicPreprocess.R"

[hihihihiwsf]

Thanks for your the data generator file, and I met another error when I run the bestPointHuber.py and other methods in TCNQuantile. It said that in the line 152 of the file nnModels.py,
Check failed: shape_assign(&(*in_shape)[i], dshape) Incompatible input shape: expected [512,168,0], got [512,336,1]

[hihihihiwsf]

It seems like the dimensions of reshaped x_num and input_store are not the same other than the appointed dim. But when I chenged the function as output = nd.concat(input_store,x_num.reshape((input_store.shape[0],input_store.shape[1],-1)), dim=2) It got an error in the forward shape inconsistency. Can you help you with this? Thank you very much!

[hihihihiwsf]

Besides , what's the difference among nnModels.py, nnModels-Copy1.py and newModels.py ?

[oneday88]

Hi, glad to hear from you again. I update a new file bestPointHuber.ipynb, which taken the input pkl file "feature_prepare.pkl" resulting from file "ec_feature_preprocess.ipynb" (instead for the original new_feature_prepare.pkl)

[hihihihiwsf]

Thank you for your reply ! My result of ECPoint.py and ECPointHuber.py without changing any parameters in them have results of valid ND is 0.42 and 0.49, and valid NRMSE is 2.82 and 2.35 respectively, which are so much different with the result in your paper. Can you help me with this?

[oneday88]

CPU or GPU server?

[hihihihiwsf]

GPU

[oneday88]

Training loss and testing loss logs, please. If ND is about 0.42. It means the predictive result is just straight light and do not learn any patterns. I will re-check the codes next Monday.

[hihihihiwsf]

Do you mean this? loss.zip I drawed the predicted first 400 values and the according true values, the result is:

Thank you very much!

[hihihihiwsf]

Hi, have you checked the code yet? I tried a lot of parameters of the model but have no progress.

[oneday88]

Hi, I upload a new folder ecPointModel for the point forecasting task. Sorry for the mess and thanks a lot for your feedback, I will take some time to refine the total project. For any questions, please let me know immediately.

[oneday88]

Hi, I updated the folder for probabilistic forecasting for ec dataset. You can try it in your GPU server, and for any questions. Please let me know.

[oneday88]

This issue has been solved.