forecast_rnn
Description
A simple rnn to predict weather from multiple features. Current latest version 0.5 uses RNN and produces 1500-2300 MSE. Use ipynb or python script to run. ipynb directly runs on github.
Scores
All scores are TRUE MSE values. This means that they have been upscaled or denormalised to true values of y rather than kept between 0-1. The current method to do this is to multiply all predictions by a constant found originally as np.max(y_input).
| RNN | CNN | Lin Reg | Support Vector | |
|---|---|---|---|---|
| Round 1 (Last 55 rows) | 1600 😏 | ~4000 😟 | ~3200 😌 | 53000!! 😵 |