ReceiptScanPrediction

This solution trains an linear classifier to predict receipt scan.

Requirement

Python >= 3.8, Pip, Docker

Setup & Run

NOTE: Please run pip install requirements.txt to install dependencies for both training and inferencing.

Inference

1. Open a command line tool, navigate to the project directory. run python3 app.py.
2. Visit localhost:8020, enter a date in MM/DD/YYYY format and click submit.
3. Run Ctrl+C in the same command line window to abort/exit.

Training

1. Prepare your data file, or use the one provided by the project directory.
2. In a command line tool, enter jupyter notebook, it will automatically open a window.
3. In the window, navigate to project folder, double click the file Train.ipynb, this will open the file in a new window.
4. (Optional) In the new window, if you want to experinment with your own file, replace url with your data file's path.
5. Select Run - Run All Cells, this command will train and export model weights along with some performance insights.

Docker

NOTE: Docker version is not usable due to a weird gradio error (not invoking function calls with button click).

1. Open a command line tool, navigate to ./app directory of the project directory.
2. Run docker build -t gefei_fetch_ml ./ to build image.
3. Run docker run -p 8123:8020 gefei_fetch_ml to run the image, visit the site with localhost:8123, you can replace 8123 with any ports you wish to use.
4. Run docker stop gefei_fetch_ml to stop container.
5. Run docker rm gefei_fetch_ml to remove the container.