Performance benchmark of cond_rnn for predicting the daily temperature in Amsterdam.
Cond_rnn is compared against; an ARMA model, an ARMAX model and a pure autoregressive LSTM.
The results are as follows;
- ARMA model: Mean Absolute Error (MAE) of 1.47 degrees
- ARMAX model with 30 cities: MAE of 1.25 degrees
- Pure autoregressive LSTM model: MAE of 1.46 degrees
- Cond_rnn with 30 cities: MAE 0.87 degrees
Extract city_temperature.csv.zip and convert python to notebook,
e.g. jupytext --to notebook lstm.py.