Final project for UNC COMP 590: Geospatial Store
We use the uwsgi-nginx-flask Docker image to deploy the application, along with Docker Compose to build and configure our services. We have one service for Flask and one service for Nginx.
Setup
First, clone this repository to your local machine and cd into it:
git clone https://github.com/akan72/comp590
cd comp590
Next, we must set up the Flask development environment. This repository utilizes the dotenv package to manage app settings.
Starting off, we will use development mode and thus Flask requires us to define two environment variables.
echo "FLASK_APP=src/app/__init__.py" >> flask/.env
echo "FLASK_ENV=development" >> flask/.env
Run the script create_db.py to initialize the SQLAlchemy database on your local machine.
python flask/create_db.py
We use SQLAlchemy to create a light lightweight Object-Relational Mapping (ORM) to create a virtual Sqlite3 database for model prediction results that lives in file. To view these results after running several models and creating the database, you can run:
sqlite3 flask/src/app/database.db
SELECT * FROM prediction;
Build and Test
To build our services, we will now begin using Docker Compose. After starting up the Docker daemon on your machine, run:
docker-compose build
docker-compose up
The following achieves the same result:
docker-compose up --build
The app may now be viewed by visiting http://127.0.0.1/ or http://localhost/ in a web browser. You must rebuild the image every time changes are made, but if you wish to restart the application without having made changes, only need to run:
docker-compose up
For rapid development and testing, the Flask application can be started without the webserver or container by runningflask run within the flask directory and navigating to http://127.0.0.1:5000/.
Dashboard
From the index, we have also included a Dashboard where users can see the historical results of trained models. Every request to the API is logged within a SQLite database that stores the user_id of the requester, the time of the request, as well as the type of model and its result.
The dashboard may be filtered by user_id or by model_type, and users can view the original image submitted as well as new images outputted by models performing tasks like segmentation.
Data Sources
For sample satellite images one source is the Planet Amazon Dataset from Kaggle. After logging in with a Kaggle account, the various .jpg datasets may be installed through the Kaggle CLI or directly downloaded to your machine in a compressed format.
Project Directory Organization
├── LICENSE
├── README.md
├── docker-compose.yml <- Configuration file for Docker Compose
├── flask
│ ├── Dockerfile
│ ├── data
│ │ ├── raw <- Directory of sample images for the various models
│ ├── main.py <- Entrypoint for the `Flask` application
│ ├── models <- Directory of Trained and serialized models used for making predictions
│ ├── notebooks <- Directory of Jupyter notebooks for exploration and model testing
│ ├── requirements.txt <- The requirements file for reproducing the analysis environment
│ ├── src
│ │ ├── app
│ │ │ ├── static
│ │ │ │ ├── uploads <- Temporary location for all images uploaded to our application
│ │ │ ├── templates <- Directory of templates for the `Flask` application
│ │ │ ├── models.py <- Contains the schema for our API's prediction results
│ │ │ └── views.py <- Backend logic for the `Flask` application
│ │ ├── models <- Scripts to train and serialize models
│ │ └── visualization <- Scripts to perform exploratory data analysis and visualization
│ ├── test_requests.py <- Example of how to use our API using Python's `requests` package
| ├── create_db.py <- Script that initializes the SQLAlchemy database
│ └── uwsgi.ini <- uWSGI config file
├── nginx
│ ├── Dockerfile
│ └── nginx.conf <- `Nginx` configuration file
└── reports <- Generated analysis as HTML, PDF, LaTeX, etc.
└── figures <- Generated graphics and figures to be used in reporting
Future Work
- Upload the container to Docker Hub
- Add aggregate counts by date to the dashboard, show examples for computing aggregated statistics about different models
- Create pipeline to ingest Google Earth Engine Polygons for other GEE models
- Define spec and create example for others to submit additional models