/RNN-ControlwithDelay

Primary LanguagePython

Recurrent neural networks for stochastic control problems with delay in TensorFlow (2.0)

Training

Under the directory src, run the command

python main.py --config_path=configs/polog_lstm.json

Command-line flags:

  • config_path: Config path corresponding to the control problem to solve. There are three control problems implemented so far. See Problems and Configs section below.
  • exp_name: Name of numerical experiment, suffix of numpy file output.

Problems and Configs

equation.py and config files under configs now support the following three problems, corresponding to three examples in Section 4.1, 4.2, and 4.3 of ref [1]:

  • LQ: Linear-quadratic problem with delay (3-dimensional or 10-dimensional state variable).
  • Csmp: Optimal consumption in a delayed financial market.
  • POlog: Portfolio optimization with complete memory and log utility.

Suffix _lstm means using long short-term memory (LSTM) networks and _shff means using feedforward networks with shared parameters.

Dependencies

Reference

[1] Han, J. and Hu, R. Recurrent neural networks for stochastic control problems with delay, (2021) [arXiv]