COVID-19 Prediction Model condLSTM-Q

Overview

A LSTM-based prediction model for daily COVID-19 death counts.

Usage

Set config.json accordingly.

1. date_generated: Dummy in effect.

2. start_train: The first date of the input data to be trained. null uses the earliest possible date.

3. end_train: The last date of the input data to be trained.

4. start_date: The first date of prediction.

5. end_date: The last date of prediction. null defaults to 2-week prediction.

6. hparam: One can pass custom hyperparameters to the model in the form of parameter name:value object.

   Currently supported:

   history_size: size of the history window

   NUM_CELLS: number of cells of LSTM layer

   lr: learning rate

   dp_ctg: dropout rate on categorical inputs

   dp_ts: dropout rate on timeseries inputs

   EPOCHS: training epochs

7. out_files: Path to output forecast (in csv format).

Run main.py.

One can pass command line arguments.

python main.py <version name, optional> <path/to/config.json, optional>