Setup Instructions:

1. Make sure Venv is setup and you have version 3.6 of Python installed
2. Clone the repo
3. cd into the repo and run virtualenv venv
4. run . venv/bin/activate
5. run pip install -r requirements.txt

Now everything is setup.

To run the the Flask server:

1. cd app
2. export FLASK_APP=app.py
3. flask run
4. Visit localhost:5000/ to see the app running

To run the Jupyter notebook:

1. Make sure you are in the root directory (cd ../ if you're in ./app/)
2. jupyter notebook
3. Visit localhost:8888 to view the notebook. All notebooks are in notebook dir

Usage

There is an API endpoint at localhost:5000/is_hotdog that accepts a POST. In the form data, it simply expects the file to have the key file. Here's an example cURL request:

POST /is_hotdog HTTP/1.1
Host: localhost:5000
Cache-Control: no-cache
Postman-Token: b626461f-9367-76e3-9056-af4e3e9ff18c
Content-Type: multipart/form-data; boundary=----WebKitFormBoundary7MA4YWxkTrZu0gW

------WebKitFormBoundary7MA4YWxkTrZu0gW
Content-Disposition: form-data; name="file"; filename="1.jpg"
Content-Type: image/jpeg


------WebKitFormBoundary7MA4YWxkTrZu0gW--

This endpoint will return a simple JSON object with probability and what values. Probability is simply the value from 0 to 1 that the image is in fact a hotdog. A value closer to 1 means its more sure, closer to 0 means its less sure. On our classifier of hotdog/not hotdog, we use a 20%/80% split, meaning it has to be very sure to classify something as a hotdog instead of a traditional 50%/50% split. Here is an example of the return JSON:

{
    "probability": 0.00001633167266845703,
    "what": "not hotdog"
}