dnouri/skorch
Skorch
Skorch is a deep learning workflow manager and tensor-based model interface. It provides a consistent API for training and predicting with neural networks within standard machine learning workflows, acting as a hyperparameter optimizer for finding optimal network configurations.
The library specializes in wrapping PyTorch neural networks in a scikit-learn compatible interface. This allows tensor-based models to be used within traditional machine learning pipelines and grid search tools, including the mapping of parameter grids to model configurations.
The framework covers training lifecycle management through early stopping, checkpointing, and learning rate schedulers. It also includes capabilities for model parameter control via layer freezing, automatic translation between tensors and NumPy arrays, and real-time training progress monitoring.
Features
- Scikit-learn Wrappers - Wraps PyTorch neural networks in a consistent scikit-learn interface for use in standard machine learning pipelines.
- Deep Learning Frameworks - Provides a framework for managing deep learning training lifecycles with early stopping and checkpointing.
- Workflow Automation - Integrates tensor-based models into automated workflows for training and prediction using a familiar API.
- Training Lifecycle Management - Manages the end-to-end training process using early stopping, checkpointing, and learning rate schedulers.
- Hyperparameter Tuning - Uses grid search and standard scoring functions to find the optimal configuration for neural network parameters.
- Hyperparameter Optimization - Translates scikit-learn parameter grids into PyTorch configurations for automated hyperparameter optimization.
- Neural Network Training Frameworks - Controls the training lifecycle with early stopping, checkpointing, and learning rate schedulers to ensure convergence.
- Hyperparameter Optimizers - Finds optimal network configurations using standard grid search and scoring functions.
- PyTorch Model Wrappers - Wraps PyTorch neural networks in a scikit-learn compatible interface for use within standard ML pipelines.
- Scikit-Learn Implementations - Allows wrapping neural networks in a compatible scikit-learn interface for use in pipelines and grid searches.
- Tensor Model Interfaces - Provides a consistent API for training and predicting with tensor-based networks within standard ML workflows.
- Model Weight Checkpoints - Persists model weights and optimizer states to allow training resumption from the best performing epoch.
- Lifecycle Callbacks - Implements a callback system to manage early stopping, checkpointing, and learning rate adjustments during training.
- Layer Freezing - Disables weight updates in specific neural network layers to control the training process.
- Parameter Freezing - Controls gradient updates by selectively freezing weights of specific neural network layers.
- Tensor to NumPy Conversions - Automatically converts data between PyTorch tensors and NumPy arrays for compatibility with traditional ML workflows.
- Deep Learning - Scikit-learn compatible wrapper for neural network training.
- PyTorch Ecosystem - Scikit-learn compatible wrapper for PyTorch.
- Developer Tools - Scikit-learn compatible wrapper.
- Training and Experimentation - Scikit-learn compatible wrapper for neural networks.
- Training and Utilities - Scikit-learn compatible wrapper for neural networks.
