Configure SMTP scrolling with Mallet CRF

A question for everyone who used the Java Mallet SimpleTagger library for conditional random fields (CRF). Suppose I already use the multi-threaded option for the maximum number of CPUs available to me (this is so): where to start, and I would try if I need it to work faster?

A related question is, is there a way to do something similar to the Stochastic Gradient Descent, which will speed up the learning process?

The type of training I want to do is simple:

Input:
Feature1 ... FeatureN SequenceLabel
...

Test Data:
Feature1 ... FeatureN
...

Output:

Feature1 ... FeatureN SequenceLabel
...

(Where the functions are the result of the processing I did for the data in my own code.)

, - CRF-, Mallet, , , , .

+5
2

, , L-BFGS, Mallet. CRFSuite, SGD, L-BFGS. Léon Bottou SGD-, .

, CRF ++ CRF. L-BFGS, , , .

CRFSuite CRF ++ .

: , . CRFSuite -n- - (n-1) - , , , .

+4

All Articles