bybatkhuu / tsc_vrae

Time series clustering VRAE.

TSC_VRAE Module

Time series clustering VRAE.

Features

• Time-series clustering
• Variational recurrent auto-encoders
• Pytorch

---

Prerequisites

• Python (>= v3.7.11)
• For NVIDIA GPU:
  • NVIDIA GPU driver (at least >= v418.39) - https://github.com/bybatkhuu/wiki/blob/main/manuals/installs/nvidia-driver-linux.md
  • NVIDIA CUDA (v10.1) and cuDNN (v7.6.5) - https://github.com/bybatkhuu/wiki/blob/main/manuals/installs/cuda-linux.md

Getting started

Clone the project

git clone https://github.com/bybatkhuu/tsc_vrae.git
cd tsc_vrae

Install dependencies

# For CPU:
cat requirements.txt | xargs -n 1 -L 1 pip3 install --timeout 60 --no-cache-dir

# For GPU:
cat requirements.gpu.txt | xargs -n 1 -L 1 pip3 install --timeout 60 --no-cache-dir

Usage/Examples

import os
import numpy as np
from tsc_vrae import TscVRAE

model_name = 'model_name'
model_dir = f"{os.getcwd()}/models"
vrae_kwargs = {
    'hidden_size': 200,
    'hidden_layer_depth': 2,
    'latent_length': 10,
    'batch_size': 16,
    'learning_rate': 1e-5,
    'dropout_rate': 0.2,
    'n_epochs': 200,
    'optimizer': 'Adam',
    'cuda': True,
    'print_every': 30,
    'clip': True,
    'max_grad_norm': 5,
    'loss': 'MSELoss',
    'block': 'LSTM'
}

X = np.array([[[1, 1], [1, 1]],
              [[2, 1], [1, 1]],
              [[3, 3], [3, 3]],
              [[3, 3], [3, 3]],
              [[10, 2], [1, 2]],
              [[10, 2], [1, 2]],
              [[10, 2], [1, 2]],
              [[3, 3], [3, 4.5]],
              [[1, 2], [1, 2]],
              [[1, 2], [1, 2]]])

tsc_vrae = TscVRAE(model_name=model_name, model_dir=model_dir, vrae_kwargs=vrae_kwargs)
if not tsc_vrae.is_trained:
    tsc_vrae.train(X)

cluster_ids = tsc_vrae.cluster(X)
print(f"Cluster IDs: {cluster_ids}")

---

Running Tests

To run tests, run the following command:

python -m tests/test*.py

Environment Variables

You can use the following environment variables to your .env file:

ENV=development
DEBUG=true
APP_NAME=tsc_vrae
LOGS_DIR="/var/log/app"

Documentation

• Hyper-parameters

Roadmap

• Support Tensorflow
• Add module API documentation
• Add tests
• Add more integrations

---

References

• https://github.com/tejaslodaya/timeseries-clustering-vae
• https://github.com/RobRomijnders/AE_ts
• https://arxiv.org/pdf/1412.6581.pdf