# Torch-Struct: Structured Prediction Library
[](https://coveralls.io/github/harvardnlp/pytorch-struct?branch=master)
A library of tested, GPU implementations of core structured prediction algorithms for deep learning applications.
* HMM / LinearChain-CRF
* HSMM / SemiMarkov-CRF
* Dependency Tree-CRF
* PCFG Binary Tree-CRF
* ...
Designed to be used as efficient batched layers in other PyTorch code.
[Tutorial paper](https://arxiv.org/abs/2002.00876) describing methodology.
## Getting Started
```python
!pip install -qU git+https://github.com/harvardnlp/pytorch-struct
# Optional CUDA kernels for FastLogSemiring
!pip install -qU git+https://github.com/harvardnlp/genbmm
# For plotting.
!pip install -q matplotlib
```
```python
import torch
from torch_struct import DependencyCRF, LinearChainCRF
import matplotlib.pyplot as plt
def show(x): plt.imshow(x.detach())
```
```python
# Make some data.
vals = torch.zeros(2, 10, 10) + 1e-5
vals[:, :5, :5] = torch.rand(5)
vals[:, 5:, 5:] = torch.rand(5)
dist = DependencyCRF(vals.log())
show(dist.log_potentials[0])
```
```python
# Compute marginals
show(dist.marginals[0])
```
```python
# Compute argmax
show(dist.argmax.detach()[0])
```
```python
# Compute scoring and enumeration (forward / inside)
log_partition = dist.partition
max_score = dist.log_prob(dist.argmax)
```
```python
# Compute samples
show(dist.sample((1,)).detach()[0, 0])
```
```python
# Padding/Masking built into library.
dist = DependencyCRF(vals, lengths=torch.tensor([10, 7]))
show(dist.marginals[0])
plt.show()
show(dist.marginals[1])
```
```python
# Many other structured prediction approaches
chain = torch.zeros(2, 10, 10, 10) + 1e-5
chain[:, :, :, :] = vals.unsqueeze(-1).exp()
chain[:, :, :, :] += torch.eye(10, 10).view(1, 1, 10, 10)
chain[:, 0, :, 0] = 1
chain[:, -1,9, :] = 1
chain = chain.log()
dist = LinearChainCRF(chain)
show(dist.marginals.detach()[0].sum(-1))
```
## Library
Full docs: http://nlp.seas.harvard.edu/pytorch-struct/
Current distributions implemented:
* LinearChainCRF
* SemiMarkovCRF
* DependencyCRF
* NonProjectiveDependencyCRF
* TreeCRF
* NeuralPCFG / NeuralHMM
Each distribution includes:
* Argmax, sampling, entropy, partition, masking, log_probs, k-max
Extensions:
* Integration with `torchtext`, `pytorch-transformers`, `dgl`
* Adapters for generative structured models (CFG / HMM / HSMM)
* Common tree structured parameterizations TreeLSTM / SpanLSTM
## Low-level API:
Everything implemented through semiring dynamic programming.
* Log Marginals
* Max and MAP computation
* Sampling through specialized backprop
* Entropy and first-order semirings.
## Examples
* BERT Part-of-Speech
* BERT Dependency Parsing
* Unsupervised Learning
* Structured VAE
## Citation
```
@misc{alex2020torchstruct,
title={Torch-Struct: Deep Structured Prediction Library},
author={Alexander M. Rush},
year={2020},
eprint={2002.00876},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
This work was partially supported by NSF grant IIS-1901030.