Research Fall 2019

• FuzzyWuzzy: pip install fuzzywuzzy
• py_stringmatching: pip install py_stringmatching
• Pytorch: conda install -c pytorch pytorch
• AllenNLP: pip install allennlp

• create_sets.py: Takes raw data and converts it into training and testing labeled sets. It creates a full version of testing and training data and a reduced version for local testing (prefix 's_').
• logistic_binary_match.py: Each quantity entry is compared to each database option and is classified as match or not. The inputs are encoded using a BOW.
• logistic_binary_str_comparison.py: Each quantity entry is compared to each database option and is classified as match or not. The input for the logistic regression are string matching values.
• logistic_multiclass.py: Each database option is a class and every input is classified to one of them.
• main.py: Retrieves files from Dr. Korpusik desktop using retrieveSSH.py, saves data by meal and concats all data, performs exact string matching from string_matching.py.
• model_sum.xlsx: Results from all experiments using NN:
  • ReLu & Loss function
  • Batch size
  • Patience
  • Optimizer
  • 2 Linear & Dot product vs 2 Linear & Bilinear vs 1 Linear
  • RNNs
  • Stacked
• Neural_Network_test.py: Image classification NN, used to make sure pytorch was working and to practice making a NN. Not used in research.
• NN_conv_allennlp.py: Convolusional NN model used to classify the user quantities. Still in development and testing.
• NN_match_with_allennlp.py: Simple FNN used to classify the user quantities.
• NN_test_allennlp.py: Text classification NN using allennlp, used to make sure allennlp was working and to practice using the library. Not used in research.
• organize_food_units.py: Organizes food_units.csv by adding column names, sorting, and removing nbd letter prefix
• readCSV.py: Helper file to read and process CSV files.
• retrieveSSH.py: Helper file to retrieve data using SSH.
• string_matching.py: Performs string matching using FuzzyWuzzy (Levenshtein distance).
• tmp.xlsx: Temp results from exepriments.

You can find a diagram of the basic NN architecture in Lucidchart here (Login required).