EasyHybrid.jl

Caution

Work in progress

EasyHybrid.jl provides a simple and flexible framework for hybrid modeling, enabling the integration of neural networks with mechanistic (physics-based) models. This approach can be expressed as:

$$ \hat{y} = \mathcal{M}(h(x;\theta), z; \phi) $$

where $\hat{y}$ denotes the predicted output of the hybrid model, $h(x;\theta)$ is a neural network with inputs $x$ and learnable parameters $\theta$, $z$ denotes additional inputs passed directly to the mechanistic model $\mathcal{M}(\cdot, z;, \phi)$, which is parameterized by $\phi$. The parameters $\phi$ may be known from first principles or learned from data.

Installation

Clone the repository

git clone https://github.com/EarthyScience/EasyHybrid.jl.git

and start using it by opening one of the env in projects, i.e. Q10.jl. There executing the first 4 lines should get you all needed dependencies. shift + enter.

If you want to start adding new functionality then do

EasyHybrid $ julia # call julia in the EasyHybrid directory

julia> ] # ']' should be pressed, this is the pkg mode

pkg > activate . # activate this project

install dependencies

pkg > instantiate

and now you are good to go!

using EasyHybrid