Skip to content

README.md

Latest commit

 

History

History
55 lines (48 loc) · 3.78 KB

README.md

File metadata and controls

55 lines (48 loc) · 3.78 KB

LM

Code for the LM project collaboration

It may be helpful to install this git repo for the maze code: https://github.com/MattChanTK/gym-maze.

Required Packages: Quandl, Gym, Pygame, Pandas, Sklearn, Torchvision, Pytorch, Tqdm, Scipy

Optional Packages: Comet_ml

Initial Stock Experiments

  • stockSynthesis.py - simple LSTM close price prediction given the corresponding open price
  • predictOpen.py - predict the next open price given the previous one
  • LSTM_Sequential.py - predict an open price sequence given a sequence of previous open prices
  • stockDataset.py - manipulate kaggle stock data into sequences of open prices
  • stockAgent.py - reinforcement learning agent to learn a multiplier to produce the next open price
  • stockRL.py - runner code for the Stock Agent to predict open prices
  • pytorchSequentialPredict.py - LSTM_Sequential.py but implemented in pytorch instead of keras

Q Learning and Feudal RL Maze Experiments

  • maze_2d_q_learning.py - solve a maze using Q learning
  • maze_env.py - creates the base maze environment for these experiments
  • maze_view_2D.py - helper for maze_env.py
  • maze_2d_dqn.py - solve a maze using Deep Q learning
  • maze_multiscale_qlearning.py - solve a maze using Feudal Q learning
  • multi_maze.py - creates the environment for the feudal learning experiments
  • maze2d_002.npy - maze file for the manager (2x2 maze)
  • maze2d_003.npy - maze file for the worker (4x4 maze)
  • maze_qmodels.py - contains code for the worker and manager networks for maze_multiscale_qlearning.py
  • maze_multiscale_dqn.py - solve a maze using Feudal Deep Q Networks
  • maze_dqn_models.py - contains code for the worker and manager networks for maze_multiscale_dqn.py

Stock Agent Portfolio Experiments

  • stock_qlearning.py - portfolio experiment with a Q learning agent
  • stock_env.py - stock market environment for stock_qlearning.py
  • stock_dqn.py - portfolio experiment with a DQN agent
  • stock_env_dqn.py - stock market environment for stock_dqn.py
  • multiscale_q_stock.py - portfolio experiment with feudal Q learning
  • multi_stock_env.py - stock market environment for multiscale_q_stock.py
  • stock_qmodles.py - contains the worker and manager code for multiscale_q_stock.py
  • stock_multi_dqn.py - portfolio experiment with feudal DQNs
  • multi_stock_env_dqn.py - stock market environment for stock_multi_dqn.py
  • stock_dqn_models.py - contains the worker, manager, and other helper classes for multi_stock_env_dqn.py
  • multi_transaction_feudal.py - feudal portfolio experiment where the manager chooses from several workers with different levels of temporal abstraction
  • multi_trans_stock_env.py - stock market environment for multi_transaction_feudal.py
  • hard_coded_qLearning.py - portfolio experiment with a hard coded agent as a baseline
  • dynamicM_hardcodeW.py - feudal portfolio experiment where the manager chooses from several hard coded workers
  • hard_coded_agents.py - contains the hard coded agents for dynamicM_hardcodeW.py

Steering Angle and TSNE Prediction

  • mutAngTSNE.py - use the TSNE embedding as the subroutine ID when predicting steering angles; also capable of predicting multiple steering angles out
  • predZKomanda.py - use a modified version of the winning udacity steering angle challenge network to predict steering angles and subroutine IDs
  • tsnePredict.py - predict the TSNE centroids using images (poor results)
  • udacityData.py - dataset class to process the udacity driving dataset
  • komanda.py - predict steering angles using the winning udacity steering angle challenge network
  • tsnePrevDataPred.py - predict the next TSNE centroid given the input data for the previous centroid (poor results)