kaimaoge/ignnk

ignnk's Issues

HELP, There are errors when running the `kpmf.py` in baselines folder

When I directly run the kpmf.py file in the baselines, I faced these errors as following.

1. one error like following

   ERROR 1
   Traceback (most recent call last):
   File "kpmf.py", line 187, in
   U, V, RMSE,time = kpmf_sgd(trainSet, mask_train, D, K_u_inv, K_v_inv, sigma_r, eta, valSet, mask_val)
   File "kpmf.py", line 120, in kpmf_sgd
   diag_K_v_inv_V=np.multiply(np.diag(K_v_inv),V)
   ValueError: operands could not be broadcast together with shapes (1500,) (1500,1000)

2. After I tried to fix ERROR 1, another error showed up.

   ERROR 2
   Traceback (most recent call last):
   File "kpmf.py", line 187, in
   U, V, RMSE,time = kpmf_sgd(trainSet, mask_train, D, K_u_inv, K_v_inv, sigma_r, eta, valSet, mask_val)
   File "kpmf.py", line 123, in kpmf_sgd
   gd_u=-2/(sigma_r**2)*(np.multiply
   ValueError: operands could not be broadcast together with shapes (1,3990) (399,1000)

I am feeling helpless here, could you please fix the errors after you uploaded the baseline files. Thanks you so much. 🙏

About the Reproductive Results of Your Model

Hi, Thanks for your great work.

I have several problems with reproductive results.

First, in the Demo_METR-LA.ipynb, the best results of all matrices are MAE: 6.944404237195023, RMSE: 10.242227312998402, MAPE: 0.1962963276318442, which are poor than the results you reported in the paper.

Also, I adopted the pre-trained model IGNNK_metr_10iter. However, the results are 13.195990007471547 25.898476875369784 0.3367085701777233, which are even worse than the Jupiter results.

So, how can I solve this problem. Can you release your pre-trained model that is slightly similar to your results in the paper. Or I need to retrain the model myself.

Thanks. :)

About the performance

A great work, and thanks the authors for sharing the code.
However, I have some questions about the model performance.
I trained this model using the code and got the following results:
on the METR-LA dataset, RMSE: 10.376, MAE: 6.829, R2: 0.806
on the NREL dataset, RMSE: 5.697, MAE: 3.975, R2: 0.729
on the USHCN dataset, RMSE: 3.494, MAE: 2.123, R2: 0.736
on the PeMS-BAY dataset, RMSE: 6.608, MAE: 3.826, R2: 0.525, MAPE: 8.9%
Those results are much worse than the results reported in the paper.
Even in the Demo_METR-LA.ipynb provided by the authors, the best results are RMSE: [10.40374277], MAE: [6.7628467], R2: [0.80512759], which are much worse than the paper reported.
While in the SeData dataset, I got the results as RMSE: 6.664, MAE: 4.222, R2: 0.720, which are similar and even better than the results in the paper.
So, how to train this model on the METR-LA, NREL, USHCN, and PeMS-BAY datasets to achieve similar results as reported in the paper.

spatial location problems

In section 3.2, "As we can not exactly know the spatial locations of test nodes, Wsample is also randomly generated to make the learning samples generalizable to more cases." was given.
However, I found the adjacent matrix is all node's spatial location in testing. For example, the dims for adj matrix of METR is 207*207. The spatial locations of test nodes are given?

questions

thanks for authors,
your method "IGNNK" is good, I want to conduct research about kriging problems, can you provide me the baseline code?

Thank you very much!

Setting sigma for adjacency matrix

Hello. I am YJ Hong.

In appendix material in IGNNK paper, there are different sigma settings for different datasets respectively when constructing corresponding adjacency matrix.

Can you let me know how you set sigma ?

For example of NREL dataset, sigma for adjacency matrix is set to 7.5km and I cannot figure out why this value came out.

Thanks in advance.

Regards, YJ Hong.