Experimental Design Analysis

The open-resource repository for the paper Unveiling the Secrets: How Masking Strategies Shape Time Series Imputation The code and configurations for reproducing the experimental results in the paper.

Code for the Time Series Imputation Survey

The scripts and configurations used in the work are all put here.

❖ Python Environment Creation

A proper Python environment is necessary to reproduce the results. Please ensure that all the below library requirements are satisfied.

pypots >=0.4
tsdb >=0.2
pygrinder >=0.4

For Linux OS, it is able to create the environment with Conda by running conda create -f pypots.yml. For other OS, library version requirements can also be checked out in pypots.yml.

❖ Datasets Introduction and Generation

Introduction

Air

Air (Beijing Multi-Site Air-Quality) is collected from twelve Beijing monitoring sites hourly in forty-eight months. At each site, eleven air pollution variables (e.g. PM2.5, NO, O₃) are collected. The dataset has 1.6% originally missing data.

PhysioNet2012

PhysioNet2012 (PhysioNet-2012 Mortality Prediction Challenge) includes multivariate clinical time series data collected from 11,988 patients in ICU. Each sample contains thirty-seven measurements (e.g. glucose, heart rate, temperature) recorded in the first forty-eight hours after admission to the ICU. This dataset has 80% values missing.

ETTm1

ETTm1 (Electricity Transformer Temperature) records seven state features, including oil temperature and six power load variables of electricity transformers collected every fifteen minutes for two years. There is no original missingness in this dataset.

Generation

The scripts for generating three datasets used in this work are in the directory data_processing. To generate the preprocessed datasets, please run the shell script generate_datasets.sh or execute the below commands:

# generate PhysioNet2012 dataset
python data_processing/gene_physionet_2012.py

# generate Air dataset
python data_processing/gene_air_quality.py

# generate ETTm1 dataset
python data_processing/gene_ettm1.py

## ❖ Model Training and Results Reproduction
```shell
# reproduce the results on the dataset PhysioNet2012
nohup python train_models_for_physionet2012.py > physionet2012.log&

# reproduce the results on the dataset Air
nohup python train_models_for_air.py > air.log&

# reproduce the results on the dataset ETTm1
nohup python train_models_for_ettm1.py > ettm1.log&

After all execution finished, please check out all logging information in the according .log files.