search

geopanag / pandemic_tgnn

MIT License
44 stars 14 forks source link

Could you share the rest of the code? #2

Closed childe-wuhen closed 3 years ago

childe-wuhen commented 3 years ago

Could you share the rest of the code? I can't find the code of MPNN_TL and the code of the function 'generate_graphs_by_day()' and 'generate_graphs_britain()' in the file 'utils'.

geopanag commented 3 years ago

These functions were made to manage the initial data, which we can not share due to the FB data license agreement. Instead I cropped the data such that we can share it and you can read it with generate_graphs_tmp. MPNN_TL is given by the metalearn.py.

childe-wuhen commented 3 years ago

Thanks a lot! But why don't you use the mean square error (MSE) or mean absolute error (MAE) to evaluate the model? by the way, how can I run "error_case_maps.R"? I know nothing about R language.

childe-wuhen commented 3 years ago

Thanks a lot! But why don't you use the mean square error (MSE) or mean absolute error (MAE) to evaluate the model? by the way, how can I run "error_case_maps.R"? I know nothing about R language.

geopanag commented 3 years ago

The training is done based on MSE (experiments.py line 30). The final predictions are evaluated on MAE (line 299) because it indicates the number of cases lost so it s more interpretable. The error_case_maps.R is used for the plotting so it is not required for the method :) . I uploaded it because people were asking me about the map plots of the paper.

catcatwang commented 3 years ago

Hello, author. I still don't understand. I can't find generate graphs britain, generate graphs by_ Where are these two functions? How to call it？looking forward to your reply

geopanag commented 3 years ago

Hello, as I said above, these functions were used in the initial version with the FB data. Since I can not share these data I created an aggregated version that you can load with generate_graphs_tmp. That is why these functions are commented out.

catcatwang commented 3 years ago

Thank you very much for your reply. But I still can't debug the program successfully. I've changed the generate graphs TMP replaces generate graphs by_ Day, or display error. Do you know why, or can you release the final version?

geopanag commented 3 years ago

I have removed all instances and calls to generate_graphs_by_day from all scripts and fixed the path names. I tested experiments.py and metalearn and they are working. Please re-download the code and try again.

catcatwang commented 3 years ago

Thank you very much. I have successfully transferred the code

catcatwang commented 3 years ago

Hello, I'm sorry to disturb you again. I'd like to ask why i started training Loss and val Loss will be very big, reaching hundreds or even thousands,Is it because I use CPU instead of GPU?

geopanag commented 3 years ago

Hello, the error I print is the cumulative error. You have to divide it by the number of regions in the country to get the average per region. Also in the first test samples the train set is small (as noted in the paper) so you should expect the error to be big.