locuslab/TCN
TCN
TCN is a deep learning sequence framework and library for building temporal convolutional networks. It provides a toolkit for implementing purely convolutional architectures to model sequential data as an alternative to recurrent neural networks.
The project includes a sequence modeling benchmark suite designed to evaluate the accuracy and processing speed of architectures. This suite utilizes standardized tasks, including memory problems, digit classification, music, and language tasks, to quantify performance.
The framework covers a range of structural components for sequence processing, such as causal convolutional layers to prevent information leakage and dilated kernels to increase the receptive field. It also provides residual connection blocks for gradient stability and data pipelines for transforming raw time-series data into tensors.
Features
- Sequence Modeling Frameworks - Serves as a complete toolkit for implementing and testing time-series processing patterns using convolutional layers.
- Dilated Convolutions - Utilizes dilated convolutions to exponentially increase the receptive field without increasing the parameter count.
- Causal Convolutions - Implements causal convolutional layers that prevent information leakage from future to past in temporal sequences.
- Sequence Learning Models - Provides capabilities for modeling sequential data across various time scales using convolutional layers.
- Convolutional Sequence Models - Replaces recurrent neural networks with a stack of convolutional layers to allow parallel processing of sequences.
- Temporal Convolutional Networks - Provides a dedicated library for building temporal convolutional networks as an alternative to recurrent neural networks.
- Model Performance Benchmarking - Implements a performance evaluation framework to test model accuracy and speed across various sequence tasks.
- Model Benchmarking Suites - Includes a collection of standardized tasks and datasets for evaluating the accuracy and efficiency of sequence models.
- Residual Connection Implementations - Includes residual connection blocks to stabilize gradients and support deeper network architectures.
- Time Series Forecasting - Provides convolutional structures designed to model temporal dependencies and predict future values in sequence data.
- Model Evaluation and Benchmarking - Provides a benchmark suite to validate model accuracy and processing speed across standardized sequence tasks.
- Time Series Tensor Pipelines - Ships pipelines that transform raw sequential data into formatted tensors for 1D convolutional processing.
- Language Modeling - Convolutional and recurrent networks for sequence modeling.
- Model Implementations - Temporal convolutional networks implementation.
