The code to run experiments with the DESYNC-MHP model is shipped here. Below are instructions to run the example code inside a docker container.
The lib
folder contains all the internal code used in the paper (written in Python). Examples of code usage are provided in the notebooks
folder as jupyter notebooks.
The code can be run over Docker. These instructions assume that Docker Desktop is installed on your computer and that a docker deamon is running.
-
We first need to build the docker image (this may take a few minutes to install and compile all dependencies).
docker build -t desync-mhp .
This creates the docker image
desync-mhp
with all the necessary dependencies. Notice: it may take a few minutes to build the image. -
Now that the image is built, we can create a container to start a
jupyter
server on thedocker run -p 8888:8888 desync-mhp
This runs a container and exposes a jupyter
server on port 8888
.
The following instructions assume that the previous installation step was performed and that the desync-mhp
container is running. A jupyter
server can then be accessed by opening the following adress in a web browser:
http://0.0.0.0:8888/?token=dummydummy
There are two notebooks illustrating the contributions on the paper.
The first notebook 1. DESYNC-MHP MLE on a toy example
takes the toy example used in the paper and applies both the classic maximum likelihood estimation and our DESYNC-MHP MLE approach to accurately recover the parameters of the model.
The second notebook 2. Effect of synchronization noise on the classic ML estimator
reproduces Figure 1b from the paper by varying the synchronization noise on a simple toy example to demonstrate the destructive effect of synchronization noise on the classic maximum likelihood estimation.