what is TokenPrecision,SpanPrecion....?

Help
Prakash
2011-10-11
2013-04-26
  • Prakash
    Prakash
    2011-10-11

    Hi,

    Can anybody explain result below but I have.. I could not understand any of it. I delete repeated output with ……… to make the question a bit smaller. Hope I will get answer sooner 

    | Task presenting examples to AnnotatorLearner: 1 document(s) in 1.83 sec
    | Task presenting examples to AnnotatorLearner: 2 document(s) in 3.56 sec
    ………………………………………………………………………………………………………..
    ………………………………………………………………………………………………………..
    | Task presenting examples to AnnotatorLearner: 69 document(s) in 166.73 sec
    | Task presenting examples to AnnotatorLearner: 70 document(s) in 167.91 sec

    Task training semi-markov voted-perceptron: 0.29% (1/350 sequences) in 1.56 sec
    ……………………………………………………………………………………………………………………
    ……………………………………………………………………………………………………………………….
    Task training semi-markov voted-perceptron: 19.71% (69/350 sequences) in 137.30 sec
    Task training semi-markov voted-perceptron: 20.00% (70/350 sequences) in 138.30 sec

    Epoch 0: sequenceErr=60 transitionErrors=3335/90433
    | Task training semi-markov voted-perceptron: 20.29% (71/350 sequences) in 139.50 sec
    | Task training semi-markov voted-perceptron: 20.57% (72/350 sequences) in 140.65 sec
    …………………………………………………………………………………………………………………………
    ………………………………………………………………………………………………………………………..
    | Task training semi-markov voted-perceptron: 39.71% (139/350 sequences) in 273.91 sec
    | Task training semi-markov voted-perceptron: 40.00% (140/350 sequences) in 274.92 sec

    Epoch 1: sequenceErr=52 transitionErrors=1161/90433

    Task training semi-markov voted-perceptron: 40.29% (141/350 sequences) in 276.14 sec
    …………………………………………………………………………………………………………………………
    ………………………………………………………………………………………………………………………..
    Task training semi-markov voted-perceptron: 59.71% (209/350 sequences) in 410.58 sec
    Task training semi-markov voted-perceptron: 60.00% (210/350 sequences) in 411.61 sec
    Epoch 2: sequenceErr=40 transitionErrors=740/90433
    Task training semi-markov voted-perceptron: 60.29% (211/350 sequences) in 412.82 sec
    …………………………………………………………………………………………………………………………
    ………………………………………………………………………………………………………………………..
    Task training semi-markov voted-perceptron: 79.71% (279/350 sequences) in 551.03 sec
    Task training semi-markov voted-perceptron: 80.00% (280/350 sequences) in 552.23 sec

    Epoch 3: sequenceErr=35 transitionErrors=654/90433

    Task training semi-markov voted-perceptron: 80.29% (281/350 sequences) in 553.52 sec
    …………………………………………………………………………………………………………………………
    ………………………………………………………………………………………………………………………..
    Task training semi-markov voted-perceptron: 99.71% (349/350 sequences) in 695.49 sec
    Task training semi-markov voted-perceptron: 100.00% (350/350 sequences) in 696.51 sec

    Epoch 4: sequenceErr=35 transitionErrors=608/90433

    | Task tagging with segmenter: 1 document(s) in 7.82 sec
    | Task tagging with segmenter: 3 document(s) in 10.88 sec
    …………………………………………………………………………………………………………………………
    ………………………………………………………………………………………………………………………..
    | Task tagging with segmenter: 30 document(s) in 116.77 sec
    | Task tagging with segmenter: 31 document(s) in 117.84 sec

    Test partition 1:
    TokenPrecision: 1.0000 TokenRecall: 0.0960 F: 0.1752
    SpanPrecision:  1.0000 SpanRecall:  0.0845 F: 0.1558
    Task train/test experiment: 100.00% (1/1 folds) in 982.71 sec
    Overall performance:
    TokenPrecision: 1.0000 TokenRecall: 0.0960 F: 0.1752
    SpanPrecision:  1.0000 SpanRecall:  0.0845 F: 0.1558

    Total time for task: 982.735 sec

    Prakash

     
  • Frank Lin
    Frank Lin
    2011-10-11

    Everything before "Test partition 1" are logs/debug outputs by minorthird and the specific learning algorithm - in this case VotedPerceptron. You can learn more about voted perceptron by using google scholar.

    Test partition shows you the evaluation of the performance of the extractor on the particular test partition. You can read about the different between Token- and Span- Precision and Recall on a previous thread:

    https://sourceforge.net/projects/minorthird/forums/forum/358215/topic/2826472

    The overall performance is the same as the test partition 1 because you have only 1 test partition.