In this project, we are matching a set of reciepts to a set of users, each of whom has a vendor name, total cost, vendor address, and purchase date.

wjoijfoqwjeoif wehfkqjwef qwkqeflkqwef

All data is in our Google Drive folder: https://drive.google.com/drive/u/0/folders/1MzXCKFLMYq2LaYryNc79-pFDkfExN4kb It was given to us by Bill.com for this specific competition.

• Copy the img and ocr folders into data/original
• Copy Users.csv and test_transactions.csv into data/original as well
• Also be sure to install all libraries in requirements.txt
  • Use the command "pip install -r requirements.txt"

• To clean the data, run scripts/clean_data.py
• Then, to add OCR data from receipts that did not have them given, run scripts/gen_ocr.py
• Then, to find the best parameters for converting these metrics to matchings, run all the cells in lven_vname.ipynb and machine_learning.ipynb
  • The model's accuracy was around 89.5%

• data/
  • final/ the directory for processed and final data
  • interim/ the directory cleaned and manipulated data goes into
  • original/ the directory to put the data from drive into
• modules/ all Python scripts to be imported
  • cvread.py reads receipts and generated OCRs and confidence intervals
  • lavenshtein.py contains all functions for lavenshtein-distance related activities
  • rev_helper.py helps create the array of metrics from the cleaned data
• scripts/ all Python scripts that the user can run
  • add_confidences.py an experimental script to determine the OCR confidence in each word
  • clean_data.py cleaning the Users.csv, test_transactions.csv, and existing OCR data files
  • gen_ocr.py generates OCR csv files formatted the same way as the ones given, but for receipts without them
  • optimal_matching.py finds the optimal matching using the final data
• laven_vname.ipynb
• machine_learning.ipynb
• ocr_cv.ipynb
• README.md
• requirements.txt