None official implementation of this paper.
- The study deals with a time-series forecasting task using Transformer.
- Experiments were conducted using influenza epidemic data provided by the US CDC.
- Predicted epidemics one week ahead with better accuracy than existing methods such as LSTM, RNN, ARIMA, etc.
- get state-by-state influenza case count data from the web site of CDC
- run
bash recipes/set_data.sh path_to_downloaded_data
- Data of each state will be as below.
REAGION | YEAR | WEEK | ili_ratio | |
---|---|---|---|---|
... | ... | ... | ... | ... |
30 | oregon | 2020 | 32 | 0.5 |
... | ... | ... | ... | ... |
bash recipes/run_train_single_state.sh 0 200 0.0001 mse
See here for details.
bash recipes/run_train_multi_state.sh 0 1 200 0.0001 mse
See here for details
One step ahead prediction trained with all states.