Two DP algorithms were implemented; one done by Roy and the other by Pratyush. There are also binary implementations of the DP algorithm and they can be found with suffix _Binary: DP_Binary and DP_Binary_split. If the suffix _split exists in the name of the directory, it means the objective function changes based on the simulation run in the files Policy Comparison Notebook (zoomed in transaction costs) or Policy Comparison Notebook (regular).
This can be seen under the files DQN_Binary and DQN_Binary_split, where _split denotes that the objective function changes based on the function you are trying to anaylze in the simulation. If you are looking for a continuous action space implementation of DQN, please see the file DQN-Old.
As mentioned previously, the notebooks Policy Comparison Notebook (zoomed in transaction costs) or Policy Comparison Notebook (regular) run a simulation against other policies and splits the objective function into gains (Sharpe ratio) and losses (cost turnover) to analyze each of the policies. They both have the same structure except the notebook ending in _10 zoomes in on low transaction costs between 0 and 10 bps.
If you are interested in looking for optimal rebalance zone (or equivalently zone of no rebalance), please see the notebooks VisualizeResults_Split and VisualizeResults for the simulation and for the general algorithm respectively.
In the archives, you will find old implementations of the algorithm, as well as a stab at different models like DDPG (continuous action and state space), DQN-11 (multi-period implementation), and an implementation by Dr. Yang DP-Rebal-Prof, which also includes a stab at signal change.