pytorch_net

This repository provides classes for

• Efficient construction of neural networks with PyTorch, including multilayer perceptron (MLP), LSTM, CNN, WideResNet, model ensembles, etc.
• Easy manipulation of networks, including addition and removal of layers and neurons, training of networks, simplification of networks, loading and saving models as a dictionary. See tutorial.ipynb for a simple demonstration.

Requirements:

• PyTorch >= 0.4.1
• scikit-learn
• sympy >= 1.3 for symbolic layers

Some projects using this library:

• LAMP: ICLR 2023 spotlight
• LE-PDE: NeurIPS 2022
• AI Physicist: Physical Review E
• GIB: NeurIPS 2020
• Causal learning: ICML 2019 Time Series Workshop, best poster award
• Meta-learning autoencoder