chronocratic-models 0.1.0a1

Ready-to-use time series models implemented in PyTorch and Lightning.

chronocratic-models

Self-supervised time-series representation learning models built with PyTorch and Lightning.

Note: The PyPI package name uses a hyphen (chronocratic-models), but the import uses the chronocratic.models namespace.

Installation

pip install chronocratic-models

Quick Start

import torch
from chronocratic.models import TS2Vec, TS2VecModelParameters

# Create model with default parameters
model = TS2Vec(TS2VecModelParameters(input_dim=1))
model.eval()

# Encode a synthetic time series (batch, channels, seq_len)
synthetic_data = torch.randn(1, 1, 100)

# Get multi-scale representations
with torch.no_grad():
    representations = model.encode(synthetic_data)
    print(representations.shape)  # (1, channels, hidden_dim)

Models

Convolutional (Dilated)

Model	Description
TS2Vec	Multi-scale hierarchical representation learning via dilated convolutions with hierarchical clustering. Code source: zhihanyue/ts2vec
CoST	Decomposition-based contrastive self-supervised learning with trend-seasonal decomposition and contrastive objectives. Code source: salesforce/CoST
AutoTCL	Automatic temporal contrastive learning with a trainable augmentation module for self-supervised time-series encoding. Code source: AslanDing/AutoTCL

Convolutional (Standard)

Model	Description
Series2Vec	Self-supervised pretraining via contrastive learning on augmented time-series segments.
TSTCC	Temporal and contextual contrastive pretraining for time-series representation learning.
FCN	Fully convolutional encoder designed for Mixup Contrastive Learning (MCL) objectives.

Transformer

Model	Description
TST	Time Series Transformer with masked-reconstruction-based self-supervised pretraining.

Recurrent

Model	Description
TimeNet	Recurrent encoder-decoder architecture for time-series representation learning.

Generative

Model	Description
TimeVAE	Variational autoencoder for time-series data with latent representation encoding and generation.

Features

• Polymorphic augmentation producer contract — models accept any augmentation through a unified interface, eliminating enum-based branching.
• Lightning integration — all models are built on PyTorch Lightning for clean training loops and extensibility.
• Self-supervised representation learning — pre-trained encoders ready for downstream tasks without labeled data.
• Pre-configured model parameters — each model ships with tested default configuration dataclasses.
• NumPy and PyTorch tensor support — flexible input handling for both frameworks.

Documentation

For full API reference, guides, and examples, visit chronocratic-models.readthedocs.io.

License

This project is licensed under the BSD 3-Clause License — see the LICENSE file for details.