search

chnsh / DCRNN_PyTorch

Diffusion Convolutional Recurrent Neural Network Implementation in PyTorch
MIT License
440 stars 111 forks source link

Testing results during training #4

Open razvanc92 opened 4 years ago

razvanc92 commented 4 years ago

Hello, firstly I would like to thank you for the implementation. I've been trying to use your implementation and I've noticed a big difference, during training when evaluating (fx every 10 steps) you're only reporting the mae (over all 12 time stamps), while DCRNN reports mae/mape/rmse for every time stamp. I would be interested to see those numbers during training, or at least at the end of the training so I can compare it with other models. Do you have any suggestions how I could do this?

AprLie commented 4 years ago

Hello, firstly I would like to thank you for the implementation. I've been trying to use your implementation and I've noticed a big difference, during training when evaluating (fx every 10 steps) you're only reporting the mae (over all 12 time stamps), while DCRNN reports mae/mape/rmse for every time stamp. I would be interested to see those numbers during training, or at least at the end of the training so I can compare it with other models. Do you have any suggestions how I could do this?

you can rewrite evalute() in dcrnn_supervisor. I could provide the code which outputs this three metrics. ps: The distance between nodes (the .csv file) should be float dtype, or you will have dtype mismatch problem.

chnsh commented 4 years ago

@razvanc92 that is a reasonable request - @AprLie thanks for your help. Would @razvanc92 and @AprLie be willing to send in a PR?

Noahprog commented 4 years ago

I would like to use that too if that is possible, @AprLie @chnsh could you forward that code? I'm not sure how though. Thanks in advance

yandou904 commented 2 years ago

Hello, firstly I would like to thank you for the implementation. I've been trying to use your implementation and I've noticed a big difference, during training when evaluating (fx every 10 steps) you're only reporting the mae (over all 12 time stamps), while DCRNN reports mae/mape/rmse for every time stamp. I would be interested to see those numbers during training, or at least at the end of the training so I can compare it with other models. Do you have any suggestions how I could do this?

you can rewrite evalute() in dcrnn_supervisor. I could provide the code which outputs this three metrics. ps: The distance between nodes (the .csv file) should be float dtype, or you will have dtype mismatch problem.

I have met with the same problem, could you also provide me with the same code? thank you!

yandou904 commented 2 years ago

I would like to use that too if that is possible, @AprLie @chnsh could you forward that code? I'm not sure how though. Thanks in advance

Have you got the answer? Could you please provide me with one? Thank you!

Yangzelin99 commented 8 months ago

I would like to use that too if that is possible, @AprLie @chnsh could you forward that code? I'm not sure how though. Thanks in advance

Have you got the answer? Could you please provide me with one? Thank you!

I'm running into this issue as well.Have you got the answer? Could you please provide me with one? Thank you!

User766843 commented 6 months ago

I would like to use that too if that is possible, @AprLie @chnsh could you forward that code? I'm not sure how though. Thanks in advance

Have you got the answer? Could you please provide me with one? Thank you!

I'm running into this issue as well.Have you got the answer? Could you please provide me with one? Thank you!

Have you got the answer? Could you please provide me with one? Thank you!