Concrete Dense Network for Long-Sequence Time Series Clustering

This repository contains the official source code for the paper: "Concrete Dense Network for Long-Sequence Time Series Clustering"

---

📦 Requirements

• Python 3.9

It is strongly recommended to use a virtual environment or container.

Install dependencies:

pip install -r requirements.txt

---

📁 Datasets

UCR Time Series Dataset

1. Download the dataset from UCR Time Series Archive

2. Extract the contents into the datasets folder:

   datasets/
     └── UCRArchive_2018/
   

M5 Forecasting Dataset

1. Download the dataset from the M5 Forecasting - Accuracy competition on Kaggle

2. Extract the contents into the datasets folder:

   unzip m5-forecasting-accuracy.zip -d datasets
   rm -rf datasets/__MACOSX
   rm -f datasets/m5-forecasting-accuracy/sample_submission.csv

3. Run preprocessing script:

   python M5_EDA.py

4. Open and run the notebook for additional preprocessing:

   • M5_EDA_cat_store_id.ipynb

---

🚀 Training & Evaluation

Each model has its own folder containing a run.sh script to launch training for each dataset.

To train a model:

./run.sh

You can customize hyperparameters and set random seeds within the script.

To compute average performance metrics (across different seeds):

python results.py