lancopku / crf-adf

CRF Toolkit based on C#; support ADF (Adaptive stochastic gradient Decent based on Feature-frequency information, ACL 2012)

CRF-ADF Sequential Tagging Toolkit v1.0

This is a general purpose software for sequential tagging (or called sequential labelling, linear-chain structured classification). The CRF (Conditional Random Fields) model is described in (Lafferty et al., 2001) and the ADF (Adaptive stochastic gradient Decent based on Feature-frequency information) fast training algorithm is described in (Sun et al., ACL 2012). [Tutorial]

Main features:

• Developed with C#
• High accuracy (72.3% on Bio-Entity Recognition Task at BioNLP/NLPBA 2004, and 97.5% on Chinese Word Segmentation MSR Task)
• Fast training (faster convergence rate than traditional batch/online training methods, including LBFGS & SGD)
• General purpose (it is task-independent & trainable using your own tagged corpus)
• Support rich edge features (Sun et al., ACL 2012)
• Support various training methods, including ADF training, SGD training, & Limited-memory BFGS training
• Support automatic n-fold cross-validation for tuning hyper-parameters
• Support various evaluation metrics, including token-accuracy, string-accuracy, & F-score