An artificial intelligence utilities package built to remove the delays of machine learning research.


License
MIT
Install
pip install paiutils==3.0.1

Documentation

PAI-Utils

Programming Artificial Intelligence Utilities is a package that aims to make artificial intelligence and machine learning programming easier through abstractions of extensive APIs, research paper implementations, and data manipulation.

Modules in this package are used extensively in this playlist. The reinforcement modules are used here.

Package Features

  • Analytics
    • Plotting of data through embedding algorithms, such as Isomap and TSNE
  • Audio
    • Recording and playing
    • Volume, speed, and pitch manipulation
    • Trimming and Splitting
    • Nonoverlapping Spectrogram creation
    • Audio file conversions
  • Image
    • Simplified OpenCV Interface
  • Autoencoder
    • Trainner and Predictor
    • Basic network architecture creation
  • Evolution Algorithm
    • One dimensional evolution algorithm
    • Hyperparameter tuner
  • GAN Network
    • Trainner and Predictor
  • Neural Network
    • Trainner and Predictor
    • Dense layers that combine batch norm
    • Convolution layers that combine batch norm, max pooling, upsampling, and transposing
  • Reinforcement
    • OpenAI Gym wrapper
    • Multi-agent adverserial environment
    • Greedy, ascetic, and stochastic policies
    • Noise Policies
    • Exponential, linear, and constant decay
    • Ring and normal memory
    • Agents
      • QAgent: Q-learning with a table
      • DQNAgent Q-learning with a neural network model
      • PGAgent: State to action neural network model (Actor) trained with policy gradients
      • DDPGAgent: State to continous action space neural network model trained with deterministic policy gradients
  • Reinforcement Agents
    • DQNPGAgent: A combination of a DQN and PG agent into one agent
    • A2CAgent: Advantage Actor Critic agent
    • PPOAgent: Proximal Policy Optimization agent
    • TD3Agent: Twin Delayed DDPG Agent