KalmanNet / RTSNet_TSP

code for RTSNet

RTSNet

Feb.13, 2023 Update: branch "batched"

Support a batch of sequences being processed simultaneously, leading to dramatic efficiency improvement.

Overview

RTSNet extends KalmanNet from filtering to smoothing. The inside network structure is also upgraded compared with old version in ICASSP2022.

Running code

• Linear kinematic model

python3 main_linear_CA.py

• Linear canonical model (include scaling model dim, scalable trajectory length T, random initial state)

python3 main_linear_canonical.py

• Linear canonical with partial model info

python3 main_linear_estF.py

python3 main_linear_estH.py

• Non-linear Lorenz Attractor model (discrete-time, decimation and non-linear observation function cases)

python3 main_lor_DT.py

python3 main_lor_DT_NLobs.py

python3 main_lor_decimation.py

Parameter settings

• Simulations/model_name/parameters.py

Contain model settings: m, n, f/F, h/H, Q and R.

• Simulations/config.py

Contain dataset size, training parameters and network settings.

• main files

Set flags, paths, etc.