Invoice processing in Appfolio

Using CNN (AlexNet based) and BoW of vendor names and addresses

Run experiments under Python 3.6.5 with following libs:

• pytorch 0.4.1 with CUDA 9.0 support (optional)
• torchvision 0.2.1
• scikit-learn 0.19.2
• scikit-image 0.14.0

and data processing lib including

• google-cloud-storage 1.10.0
• google-cloud-vision 0.33.0
• boto3 1.7.4
• joblib 0.11

Run mongoDB and load database invoice

mkdir ~/workspace
cd ~/workspace
git clone https://github.com/Gandor26/invoice.git
cd invoice
python prepare.py [vhost name] --download --ocr --split --create

Note that there might be a delay while downloading OCR json files from Google cloud after they are generated. Once finished, a random split of training/test dataset will be available in ~/workspace/invoice/data/set

Simply running

python main.py data/set

will load default configurations and start training.

Running

python main.py --help

will show a list of argument to configure the model.

You can press Ctrl-C anytime to stop training and start doing test on the pre-defined test set with the best model evaluated on a randomly generated validation set from training set.

All the output will be available in ~/workspace/invoice/logs/{train,test}.log.