Running

• Download glove dataset for neural word embedding http://nlp.stanford.edu
• Create a database in MySQL server. You do NOT need to create a table.
• Edit settings in settings.py.
• Edit glove_path in word2vec.py

Databases

• mbzdb: clone from https://github.com/elliotchance/mbzdb and run ./init.pl after editing settings
• msd: run ./getMSD.sh after editing settings.py
• mxm: run ./getMXM.sh

Tensorboard

• install bazel 0.20.0
• git submodule update --init --recursive
• bazel run //greeter_tensorboard --incompatible_remove_native_http_archive=false -- --logdir=/tmp/greeter_demo

Project Structure

• train.py
  • Runs all the scripts needed to setup the neural word embedding dictionary, requests subset of data from SQL server, and trains the neural network.
• frontend:
  • Uses Tensorboard as front end with check boxes to select desired features for training data. Python backend parses input and requests from SQL server a subset of data based on the specified criteria. The subset of data is then forwarded through a fully-connected neural net. The neural network trains and learns to predict the year in which an input song was written.
• Neural net:
  • Input 1: subset of training data about song information
  • Output 1: a trained neural network model
  • Input 2: subset of test data about song information
  • Output 2: classification (predicted year), and a log used for visualization with Tensorboard.