A LSTM-based prediction model for daily COVID-19 death counts.
Set config.json
accordingly.
-
date_generated
: Dummy in effect. -
start_train
: The first date of the input data to be trained.null
uses the earliest possible date. -
end_train
: The last date of the input data to be trained. -
start_date
: The first date of prediction. -
end_date
: The last date of prediction.null
defaults to 2-week prediction. -
hparam
: One can pass custom hyperparameters to the model in the form ofparameter name
:value
object.Currently supported:
history_size
: size of the history windowNUM_CELLS
: number of cells of LSTM layerlr
: learning ratedp_ctg
: dropout rate on categorical inputsdp_ts
: dropout rate on timeseries inputsEPOCHS
: training epochs -
out_files
: Path to output forecast (incsv
format).
Run main.py
.
One can pass command line arguments.
python main.py <version name, optional> <path/to/config.json, optional>