Deep learning has the power to potentially save millions of dollars (and more importantly, lives) in places like California where the annual "fire season" arrives every Fall.
We built this API to show how the technology can fight this and other crises, and inspire our students to do the same.
To classify your own images, you can use the live API: use the link here to read the documentation and send requests.
You only need to use ONE of the following options:
You can download this repository and run it using Docker:
$ docker compose up
Then head over to http://localhost:8000/docs or http://localhost:8000/redoc in the browser.
Alternatively, you can also make a virtual environment. This is recommended, as it will allow you to also run the automated tests as well (discussed below). Here are the commands to install the dependencies locally:
$ python3 -m venv env
$ source env/bin/activate
(env) $ python -m pip install -r requirements.txt
And then run the app using uvicorn in the Command Line:
(env) $ uvicorn app.main:app --reload
Then head over to http://localhost:8000/docs or http://localhost:8000/redoc in the browser.
To run the tests, you will first need to set up a Python virtual environment to run this project locally (see above). Then you can run the automated tests from the root directory, using the command line:
(env) $ pytest
If there are no failures, then you should be good to go!
You can inspect the code for the tests in app/main_test.py if you wish.
The image dataset and neural network model used for the production API will be documented on the Releases page of this repository.
TBD
TBD
In this project, we've worked with different tools like Tensorflow, Docker, FastAPI and Heroku. The next steps would be to two-fold:
- For the modelling engineers: how would you improve the neural networks performance?
- For the MLOps engineers: how would you improve the performance and scalability of the REST API in production?
- This Towards Data Science blog by Youness Mansar will give you a little more detail on how you can build a deployment-driven deep learning project (using the Google Cloud Platform's App Engine).
- Another blog by Shinichi Okada in Towards Data Science will give more details how to deploy FastAPI applications (such as this repo!) on Heroku specifically.
- If you're curious to know why we used
python -m pipin the Using Virtual Environments section, please read this explanation to see how it differs from just usingpip/pip3.