Mallet data alphabet keep growing when using a CRF

[bethard]

Original issue 408 created by ClearTK on 2014-08-17T03:31:59.000Z:

What steps will reproduce the problem?

1. Train a Mallet CRF
2. Use it to tag new text.

What is the expected output?

The data alphabet of the transducer should not change once the training is over.

What do you see instead?

Each feature that is seen when tagging new documents are added to the alphabet, leaking memory for long running tagging processes.

What version of the product are you using? On what operating system?

Version 1.4.1 and 2.0.0.

I attached a patch with a fix and a unit.

[bethard]

Comment #1 originally posted by ClearTK on 2014-08-20T21:43:29.000Z:

<empty>