This is the official implement of

This environment consists of three containers: main, PostgreSQL, and MLflow.

The main container is in which you execute experiments and analyze data.
The MLflow container is the endpoint to which you sent the experiment results and the server running the MLflow Web UI.
The PostgreSQL container is the database used by MLflow and Prefect. You can also use this container for other purposes.

nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04

3.11.5

If you want to use other version, please change the python version in Dockerfile.

We provide a example of .env, so you copy this file as template, please.

cp .env.example .env

Plaase change it if necessary.
Details of .env are given below.

IMAGE_TAG="{user_name}/{project_name}:{build date yyyymmdd}" <- Image tag ex) "user_name/project_name:yyyymmdd"
CONTAINER_NAME=""                                            <- container name to distinguish it from other users' containers.
POSTGRES_PASSWORD=""                                         <- Password of PostgreSQL server. This phrase automatically sets the PostgreSQL password when you first start the container. 
                                                                The PostgreSQL username will be the username of the user who built the image.
PORT_SSH=""                                                  <- SSH port of main container. Please select port number from dynamic, private or ephemeral ports.
PORT_SSH_MLFLOWSERVER=""                                     <- SSH port of MLFlow container. Please select port number from dynamic, private or ephemeral ports.
HOST_NAME="localhost"                                        <- No need to change it.
CUDA_LAUNCH_BLOCKING=1                                       <- This variable and its value allow for a detailed display of CUDA errors.

If you want to add an apt package, you have to write the package name to Dockerfile directly.
./enviroments/Dockerfile

RUN apt install -y build-essential ca-certificates locales-all \
    libncursesw5-dev libssl-dev libsqlite3-dev libgdbm-dev libbz2-dev \
    liblzma-dev zlib1g zlib1g-dev uuid-dev libffi-dev libdb-dev bzip2 checkinstall libreadline-gplv2-dev tk-dev \
    software-properties-common git emacs vim curl unzip htop openssh-server wget procps sudo nodejs tmux postgresql-client <- Please add the package behind here.

For Python packages, please add the library name and version to requirements.txt. ./environments/python/requirements.txt

library==ver.sion

Please execute the below "make" command to build your image.

make build

Follow the script to set the key pair and password to log in to the container.
The key pair will be created in the following folder: ./environments/etc/home/.ssh.
If you save this private key on your PC, you can use an SSH connection with exchange key encryption.
If you want to connect without exchanging key ciphers, you can use the password you set here.

make run

A user with the same username as the user who built the docker container is prepared inside the container.
Please connect to the container using your username.

make ssh 

or

ssh {user_name}@127.0.0.1 -p {PORT_SSH}

This container cannot be directly connected from the outside.
Therefore, you have to set up a connection to the container via the host.

The default ports for Prefect and MLflow are 4200 and 5000.
Please connect via http://127.0.0.1:4200 and http://127.0.0.1:5000 using SSH port forwarding.

We obtained a health checkup dataset actually at a healthcare center. Therefore, we cannot be made public.
The code used in the experiment runs as follows:

1. Run the prefect server and prefect workpool by executing following commands.

   make prefect
   make prefect_workpool
   

2. Run the following commands.

   cd src/multi_task/
   python main.py --config_path args/conditions/*.yaml --gpus 0 --use_prefect
   

We used the following scripts for preprocessing the dataset.

./src/preprocess/main.py