This repository contains demo notebooks for serving online models using the lume-model and lume-epics. A simple demo is included outlining an implementation of a minimal model using the tools, as well as an example of an more complicated surrogate model.

A conda installation is required for setting up the required packages and the code is executed inside jupyter notebooks.

$ git clone https://github.com/jacquelinegarrahan/lume-model-server-demo.git

$ cd lume-model-server-demo

Create the environment using conda and the environment.yml file included with the repository:

$ conda env create -f local-environment.yml

Activate the environment:

$ conda activate lume-model-server-demo

$ python -m ipykernel install --user --name=lume-model-server-demo

$ jupyter notebook

In two tabs, open the SimpleServer and SimpleClient notebooks. Begin by following the code outlined in the SimpleServer notebook, and then execute the code in the SimpleClient to render the application. Finally, terminate the server using the Server.stop() method in the SimpleServer notebook.

The surrogate model demo uses the MyModel python class defined in the MySurrogateModel.py. The MyModel class uses a custom implementation to load a Keras model in init and executes the model in the evaluate method. The model is loaded using an hdf5 file containing the model weights and architecture.

In two tabs, open the SurrogateModelServer and SurrogateModelClient notebooks. Begin by following the code outlined in the SurrogateModelServer notebook, and then execute the code in the SurrogateModelClient to render the application. Finally, terminate the server using the Server.stop() method in the SurrogateModelServer notebook.