DiffImp

Code for the paper: "DiffImp: Efficient Diffusion Model for Probabilistic Time Series Imputation with Bidirectional Mamba Backbone"

Requirements

Python==3.10.15

torch==2.3.1

cuda==11.8

mamba_ssm==2.2.2

causal_conv1d==1.4.0

Experiments

First modify the config in ./config

train for specific dataset:

python train.py --config path_to_your_config

inference for specific dataset:

python inference.py --config path_to_your_config

If you want to build your own imputer, modify ./imputers according to the imputer files.

For mujoco dataset:

python train.py --config ./config/config_bissm2_mujoco_90_large.json

For ablation studies:

modify line 16 in train.py and line 9 in inference.py to corresponding imputers in .\imputers\