Skip to content
/ feiertag Public

An open-source neural sequence tagging toolkit.

License

Notifications You must be signed in to change notification settings

erip/feiertag

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Feiertag

Build

Feiertag is an open-source neural sequence tagging toolkit built with PyTorch.

Installation

Use the package manager pip to install feiertag.

git clone [email protected]:erip/feiertag.git
cd feiertag
pip install -e .

Usage

Feiertag supports a handful of baseline architectures and datasets with configuration handled by Hydra. An example to train an NER model on CoNLL 2003 formatted data is shown below:

feiertag-train data.train="$DATA_DIR/train.txt" data.valid="$DATA_DIR/valid.txt" data_format=conll2003 trainer.max_epochs=25 embedding.path="$EMBEDDING_DIR/glove.6B.50d.txt" model=bilstm_crf data.loader.batch_size=128 embedding.freeze=false

Model training is handled by PyTorch-Lightning and kwargs passed to the pytorch_lightning.Trainer class can be provided by Hydra override syntax. For example, to emulate pl.Trainer(gpus='0,1', deterministic=True), use the following syntax:

feiertag-train ... +trainer.gpus="0,1" +trainer.deterministic=true

Managing Experiments

By default, Hydra maintains an output directory structure to separate runs. Similarly, PyTorch-Lightning logs checkpoints, hyperparameters, and Tensorboard logfiles with the following structure:

outputs
└── 2020-12-02
    └── 18-02-11
        ├── lightning_logs
        │   └── version_0
        │       ├── checkpoints
        │       │   └── epoch=1.ckpt
        │       ├── events.out.tfevents.1606950132.erip.16601.0
        │       └── hparams.yaml
        └── train.log

To view the above experiment in tensorboard, issue tensorboard --logdir outputs/2020-12-02/18-02-11/lightning_logs/version_0/.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

License

MIT

About

An open-source neural sequence tagging toolkit.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages