Felix660 / neural-ode-weather-forecast

How to train a neural ODE for time series/weather forecasting

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Neural ODE weather forecasting

Companion repo to my blog article Forecasting the weather with neural ODEs. Contains code for training the neural ODE and plotting the fit.

Data

The dataset used is available on Kaggle. It is released under CC0: Public domain and included in this repo's data folder for convenience.

Reproducing the experiments

Make sure you have Julia installed (tested with Julia 1.8). cd to the project root folder and run julia --project to enter the Julia shell using the project environment. Run using Pkg; Pkg.instantiate(), to pull down the dependencies, followed by include("scripts/fit_model.jl") to fit the model and generate the plots and animations seen in the blog article. Don't be surprised if creating the training animation takes at least as long as fitting the model.

Figures

Model training animation

Training animation

Model extrapolation

Model extrapolation

About

How to train a neural ODE for time series/weather forecasting


Languages

Language:Julia 100.0%