search

CogComp / saul

Saul : Declarative Learning-Based Programming
Other
64 stars 18 forks source link

formatted output for JointTraining #254

Open danyaljj opened 8 years ago

danyaljj commented 8 years ago

Currently this is how the output of JointTraining is being shown: 

[info] Loc
[info] tp: 2290.0 fp: 1015.0 tn: 62177.0 fn: 67.0 
[info]  accuracy    0.9834932645806953 
[info]  precision   0.6928895612708018 
[info]  recall      0.9715740347899873 
[info]  f1          0.8089014482515012 
[info] Loc
[info] tp: 2298.0 fp: 1015.0 tn: 62177.0 fn: 59.0 
[info]  accuracy    0.9836153106836107 
[info]  precision   0.693631150015092 
[info]  recall      0.9749681798896903 
[info]  f1          0.8105820105820106 
[info] Work_For
[info] tp: 281.0 fp: 54.0 tn: 743.0 fn: 1.0 
[info]  accuracy    0.9490268767377201 
[info]  precision   0.8388059701492537 
[info]  recall      0.9964539007092199 
[info]  f1          0.9108589951377634 
[info] Work_For
[info] tp: 282.0 fp: 1.0 tn: 796.0 fn: 0.0 
[info]  accuracy    0.9990732159406858 
[info]  precision   0.9964664310954063 
[info]  recall      1.0 
[info]  f1          0.9982300884955753 
[info] Live_In
[info] tp: 244.0 fp: 147.0 tn: 688.0 fn: 0.0 
[info]  accuracy    0.8637627432808156 
[info]  precision   0.6240409207161125 
[info]  recall      1.0 
[info]  f1          0.768503937007874 
[info] Live_In
[info] tp: 243.0 fp: 18.0 tn: 817.0 fn: 1.0 
[info]  accuracy    0.9823911028730306 
[info]  precision   0.9310344827586207 
[info]  recall      0.9959016393442623 
[info]  f1          0.9623762376237623

would be much cleaner if we use the save formatted output of LBJava; something line the following: 

[info] ==============================================
[info] WorkFor Classifier Evaluation
[info] 
[info]  Label   Precision Recall   F1   LCount PCount
[info] ----------------------------------------------
[info] false       98.253 81.967 89.374   1647   1374
[info] true        55.935 94.015 70.140    401    674
[info] ----------------------------------------------
[info] Accuracy    84.326   -      -      -      2048
[info] ==============================================
[info] LivesIn Classifier Evaluation
[info] 
[info]  Label   Precision Recall   F1   LCount PCount
[info] ----------------------------------------------
[info] false       85.581 96.005 90.494   1527   1713
[info] true        81.791 52.591 64.019    521    335
[info] ----------------------------------------------
[info] Accuracy    84.961   -      -      -      2048
[info] ==============================================
kordjamshidi commented 8 years ago

In fact I think we need to address a bigger issue of integrating all versions of test that we have and lbjava has. For the moment we do not use this evaluation you print above. We made some efforts with @christos-c to add some new versions of test in SRL PR and we should continue on that.