IDP4+_training vs nala_training (in CV)
abojchevski opened this issue · comments
Aleksandar Bojchevski commented
nala_training_51
fold 0
tp:135 fp:60 fn:55 fpo:34 fno:36 P:0.6923 R:0.7105 F:0.7013 0 exact
tp:30 fp:60 fn:65 fpo:45 fno:43 P:0.3333 R:0.3158 F:0.3243 1 exact
tp:13 fp:9 fn:18 fpo:7 fno:9 P:0.5909 R:0.4194 F:0.4906 2 exact
tp:178 fp:129 fn:138 fpo:86 fno:88 P:0.5798 R:0.5633 F:0.5714 TOTAL exact
tp:135 fp:60 fn:55 fpo:34 fno:36 P:0.8874 R:0.9152 F:0.9011 0 overlapping
tp:30 fp:60 fn:65 fpo:45 fno:43 P:0.8872 R:0.8429 F:0.8645 1 overlapping
tp:13 fp:9 fn:18 fpo:7 fno:9 P:0.9355 R:0.7632 F:0.8406 2 overlapping
tp:178 fp:129 fn:138 fpo:86 fno:88 P:0.8911 R:0.8756 F:0.8833 TOTAL overlapping
fold 1
tp:96 fp:30 fn:30 fpo:13 fno:13 P:0.7619 R:0.7619 F:0.7619 0 exact
tp:30 fp:84 fn:82 fpo:72 fno:60 P:0.2632 R:0.2679 F:0.2655 1 exact
tp:16 fp:19 fn:20 fpo:12 fno:12 P:0.4571 R:0.4444 F:0.4507 2 exact
tp:142 fp:133 fn:132 fpo:97 fno:85 P:0.5164 R:0.5182 F:0.5173 TOTAL exact
tp:96 fp:30 fn:30 fpo:13 fno:13 P:0.8777 R:0.8777 F:0.8777 0 overlapping
tp:30 fp:84 fn:82 fpo:72 fno:60 P:0.9310 R:0.8804 F:0.9050 1 overlapping
tp:16 fp:19 fn:20 fpo:12 fno:12 P:0.8511 R:0.8333 F:0.8421 2 overlapping
tp:142 fp:133 fn:132 fpo:97 fno:85 P:0.9000 R:0.8733 F:0.8865 TOTAL overlapping
fold 2
tp:125 fp:32 fn:38 fpo:15 fno:12 P:0.7962 R:0.7669 F:0.7812 0 exact
tp:25 fp:50 fn:77 fpo:42 fno:40 P:0.3333 R:0.2451 F:0.2825 1 exact
tp:7 fp:11 fn:17 fpo:9 fno:9 P:0.3889 R:0.2917 F:0.3333 2 exact
tp:157 fp:93 fn:132 fpo:66 fno:61 P:0.6280 R:0.5433 F:0.5826 TOTAL exact
tp:125 fp:32 fn:38 fpo:15 fno:12 P:0.8994 R:0.8539 F:0.8761 0 overlapping
tp:25 fp:50 fn:77 fpo:42 fno:40 P:0.9304 R:0.7431 F:0.8263 1 overlapping
tp:7 fp:11 fn:17 fpo:9 fno:9 P:0.9259 R:0.7576 F:0.8333 2 overlapping
tp:157 fp:93 fn:132 fpo:66 fno:61 P:0.9132 R:0.8000 F:0.8529 TOTAL overlapping
fold 3
tp:234 fp:46 fn:55 fpo:26 fno:26 P:0.8357 R:0.8097 F:0.8225 0 exact
tp:46 fp:76 fn:91 fpo:67 fno:59 P:0.3770 R:0.3358 F:0.3552 1 exact
tp:7 fp:12 fn:15 fpo:6 fno:7 P:0.3684 R:0.3182 F:0.3415 2 exact
tp:287 fp:134 fn:161 fpo:99 fno:92 P:0.6817 R:0.6406 F:0.6605 TOTAL exact
tp:234 fp:46 fn:55 fpo:26 fno:26 P:0.9346 R:0.9079 F:0.9211 0 overlapping
tp:46 fp:76 fn:91 fpo:67 fno:59 P:0.9503 R:0.8431 F:0.8935 1 overlapping
tp:7 fp:12 fn:15 fpo:6 fno:7 P:0.7692 R:0.7143 F:0.7407 2 overlapping
tp:287 fp:134 fn:161 fpo:99 fno:92 P:0.9318 R:0.8739 F:0.9019 TOTAL overlapping
fold 4
tp:183 fp:41 fn:60 fpo:25 fno:27 P:0.8170 R:0.7531 F:0.7837 0 exact
tp:53 fp:85 fn:95 fpo:75 fno:65 P:0.3841 R:0.3581 F:0.3706 1 exact
tp:19 fp:22 fn:31 fpo:16 fno:20 P:0.4634 R:0.3800 F:0.4176 2 exact
tp:255 fp:148 fn:186 fpo:116 fno:112 P:0.6328 R:0.5782 F:0.6043 TOTAL exact
tp:183 fp:41 fn:60 fpo:25 fno:27 P:0.9363 R:0.8769 F:0.9056 0 overlapping
tp:53 fp:85 fn:95 fpo:75 fno:65 P:0.9507 R:0.8655 F:0.9061 1 overlapping
tp:19 fp:22 fn:31 fpo:16 fno:20 P:0.9016 R:0.8333 F:0.8661 2 overlapping
tp:255 fp:148 fn:186 fpo:116 fno:112 P:0.9379 R:0.8671 F:0.9011 TOTAL overlapping
Aleksandar Bojchevski commented
IDP4+ training
fold 0
tp:592 fp:148 fn:110 fpo:91 fno:61 P:0.8000 R:0.8433 F:0.8211 0 exact
tp:27 fp:89 fn:121 fpo:60 fno:56 P:0.2328 R:0.1824 F:0.2045 1 exact
tp:13 fp:21 fn:36 fpo:10 fno:11 P:0.3824 R:0.2653 F:0.3133 2 exact
tp:632 fp:258 fn:267 fpo:161 fno:128 P:0.7101 R:0.7030 F:0.7065 TOTAL exact
tp:592 fp:148 fn:110 fpo:91 fno:61 P:0.9288 R:0.9382 F:0.9335 0 overlapping
tp:27 fp:89 fn:121 fpo:60 fno:56 P:0.8314 R:0.6875 F:0.7526 1 overlapping
tp:13 fp:21 fn:36 fpo:10 fno:11 P:0.7556 R:0.5763 F:0.6538 2 overlapping
tp:632 fp:258 fn:267 fpo:161 fno:128 P:0.9047 R:0.8689 F:0.8864 TOTAL overlapping
fold 1
tp:615 fp:112 fn:166 fpo:73 fno:70 P:0.8459 R:0.7875 F:0.8156 0 exact
tp:35 fp:82 fn:105 fpo:59 fno:55 P:0.2991 R:0.2500 F:0.2724 1 exact
tp:12 fp:18 fn:26 fpo:9 fno:11 P:0.4000 R:0.3158 F:0.3529 2 exact
tp:662 fp:212 fn:297 fpo:141 fno:136 P:0.7574 R:0.6903 F:0.7223 TOTAL exact
tp:615 fp:112 fn:166 fpo:73 fno:70 P:0.9511 R:0.8876 F:0.9182 0 overlapping
tp:35 fp:82 fn:105 fpo:59 fno:55 P:0.8663 R:0.7487 F:0.8032 1 overlapping
tp:12 fp:18 fn:26 fpo:9 fno:11 P:0.7805 R:0.6809 F:0.7273 2 overlapping
tp:662 fp:212 fn:297 fpo:141 fno:136 P:0.9297 R:0.8536 F:0.8900 TOTAL overlapping
fold 2
tp:48 fp:93 fn:105 fpo:63 fno:57 P:0.3404 R:0.3137 F:0.3265 1 exact
tp:18 fp:33 fn:27 fpo:16 fno:17 P:0.3529 R:0.4000 F:0.3750 2 exact
tp:782 fp:237 fn:220 fpo:130 fno:130 P:0.7674 R:0.7804 F:0.7739 TOTAL exact
tp:716 fp:111 fn:88 fpo:51 fno:56 P:0.9320 R:0.9626 F:0.9471 0 overlapping
tp:48 fp:93 fn:105 fpo:63 fno:57 P:0.8485 R:0.7778 F:0.8116 1 overlapping
tp:18 fp:33 fn:27 fpo:16 fno:17 P:0.7500 R:0.8361 F:0.7907 2 overlapping
tp:782 fp:237 fn:220 fpo:130 fno:130 P:0.9069 R:0.9205 F:0.9136 TOTAL overlapping
fold 3
tp:1073 fp:307 fn:162 fpo:93 fno:95 P:0.7775 R:0.8688 F:0.8207 0 exact
tp:51 fp:113 fn:118 fpo:70 fno:66 P:0.3110 R:0.3018 F:0.3063 1 exact
tp:14 fp:37 fn:25 fpo:16 fno:15 P:0.2745 R:0.3590 F:0.3111 2 exact
tp:1138 fp:457 fn:305 fpo:179 fno:176 P:0.7135 R:0.7886 F:0.7492 TOTAL exact
tp:1073 fp:307 fn:162 fpo:93 fno:95 P:0.8549 R:0.9495 F:0.8998 0 overlapping
tp:51 fp:113 fn:118 fpo:70 fno:66 P:0.8130 R:0.7824 F:0.7974 1 overlapping
tp:14 fp:37 fn:25 fpo:16 fno:15 P:0.6818 R:0.8182 F:0.7438 2 overlapping
tp:1138 fp:457 fn:305 fpo:179 fno:176 P:0.8430 R:0.9205 F:0.8800 TOTAL overlapping
fold 4
tp:465 fp:65 fn:163 fpo:40 fno:43 P:0.8774 R:0.7404 F:0.8031 0 exact
tp:48 fp:74 fn:85 fpo:59 fno:52 P:0.3934 R:0.3609 F:0.3765 1 exact
tp:13 fp:21 fn:33 fpo:13 fno:15 P:0.3824 R:0.2826 F:0.3250 2 exact
tp:526 fp:160 fn:281 fpo:112 fno:110 P:0.7668 R:0.6518 F:0.7046 TOTAL exact
tp:465 fp:65 fn:163 fpo:40 fno:43 P:0.9564 R:0.8204 F:0.8832 0 overlapping
tp:48 fp:74 fn:85 fpo:59 fno:52 P:0.9138 R:0.8281 F:0.8689 1 overlapping
tp:13 fp:21 fn:33 fpo:13 fno:15 P:0.8367 R:0.6949 F:0.7593 2 overlapping
tp:526 fp:160 fn:281 fpo:112 fno:110 P:0.9397 R:0.8139 F:0.8723 TOTAL overlapping
Dr. Juan Miguel Cejuela commented
I repeated the experiment by deleting classes 1,2. IDP4+ won: F:0.8934 vs F:0.9097
# wins P:0.9372 R:0.8535 F:0.8934 -- 10 421693.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus nala_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner parts --ps_random "0.0"
# 9 421692.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus nala_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_random "0.0"
# 8 421691.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus nala_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_ST --ps_random "0.0"
# 7 421690.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus nala_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_NL --ps_random "0.0"
# 6 421689.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus nala_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_ST --ps_NL --ps_random "0.0"
# wins P:0.9060 R:0.9135 F:0.9097 -- 5 421686.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus IDP4+_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner parts --ps_random "0.0"
# 4 421685.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus IDP4+_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_random "0.0"
# 3 421684.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus IDP4+_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_ST --ps_random "0.0"
# 2 421683.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus IDP4+_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_NL --ps_random "0.0"
# 1 421682.1-5:1 time /mnt/home/cejuela/anaconda3/latest/bin/python /mnt/home/cejuela/nala/nala/scripts/train.py --training_corpus IDP4+_training --delete_subclasses "1,2" --cv_n 5 --cv_fold $cv_fold --pruner sentences --ps_ST --ps_NL --ps_random "0.0"
Note, run without window features nor we for greater speed