CRF

A Conditional Random Field Model based Chinese Word Segmentation Project.

Introduction

CRF model takes advantage of contextual information, thus compared to HMM model, CRF improves the accuracy and recall.

Supported Keras Loss and Optimizer

Loss:

1. supports loss function:"mean_squared_error","mean_absolute_error","mean_absolute_percentage_error","mean_squared_logarithmic_error","squared_hinge","hinge","categorical_crossentropy","sparse_categorical_crossentropy","binary_crossentropy","kullback_leibler_divergence","poisson","cosine_proximity"
2. NOT support function:"categorical_hinge","logcosh"
3. Training error function:"sparse_categorical_crossentropy", Optimizer:
4. supports optimizer function:'sgd','rmsprop','adadelta','adam','adamax','nadam'
5. NOT support function:'adagrad',
6. NOT support serialization: 'tfoptimizer'.