Skip to content

Latest commit

 

History

History
498 lines (462 loc) · 24 KB

README.org

File metadata and controls

498 lines (462 loc) · 24 KB
Raw

KaSVM

The multilevel support vector machine framework KaSVM.

Overview

Our framework is build on the multilevel support vector machine approach of https://github.com/esadr/mlsvm and introduces several new coarsening strategies.

While regular SVM perform the entire training in one - time consuming - optimization step, multilevel SVMs first build a hierarchy of problems decreasing in size that resemble the original problem and then train an SVM model for each hierarchy level benefiting from the solved models of previous levels.

Our framework utilizes the near-linear time label propagation algorithm for the construction of the problem hierarchy and we are experimenting with other coarsening strategies.

Paper

Faster Support Vector Machines [ALENEX19] - Sebastian Schlag, Matthias Schmitt, Christian Schulz

Compilation

To compile install the following dependencies through your package manager:

Once you installed these packages you need to compile some additional dependencies.

ThunderSVM

cd extern/thundersvm
git submodule init eigen && git submodule update
mkdir build && cd build
cmake -DUSE_CUDA=OFF -DUSE_EIGEN=ON ..
make -j 4

BayesOpt

The following (Ubuntu/Debian) packages are needed for BayesOpt libboost-dev cmake g++

then

cd extern/bayesopt
mkdir build && cd build
cmake ..
make -j 4

you can test if the compile step worked with

./bin/bo_branin

KaSVM

Compile KaSVM with with

scons program=prepare variant=optimized_output -j 4
scons program=kasvm variant=optimized_output -j 4

Usage

Preprocessing

Before using the framework the data needs to be preprocessed. The prepare utility can read CSV and LibSVM files. It normalized the data and seperates the classes.

./optimized_output/prepare examples/twonorm.csv

output of prepare

read csv time 0.0533679
rows: 7400 cols: 20
normalization time 0.00080204
splitting time 0.00121903
nodes - min 3703 maj 3697
finished writing features to examples/twonorm_min_data in 0.0366721
finished writing features to examples/twonorm_maj_data in 0.035835

Classifier

After the data has been preprocessed run the kasvm program. As path argument use the path of the original data file without the extension.

Parameters

Please note that to reproduce our experiments usually no special configuration is needed as the default values are the values we report in Section 5 of our paper.

./optimized_output/kasvm --help
Usage: ./optimized_output/kasvm [-b] [--help] FILE [--seed=<int>] [-e <int>] [-k <int>] [-s <double>] [--validation=TYPE] [--validation_percent=<double>] [--validation_seperate] [-n <int>] [--stop_rule=VARIANT] [--fix_num_vert_stop=<int>] [--matching=TYPE] [--cluster_upperbound=<int>] [--label_propagation_iterations=<int>] [--diameter_upperbound=<double>] [--beta=<double>] [--refinement=TYPE] [-C <double>] [-g <double>] [--num_skip_ms=<int>] [--no_inherit_ud] [--export_graph] [--output_filename=<string>] [--export_model=<string>] [--timeout=<int>] [-c <int>]
  --help                                   Print help.
  FILE                                     Path to graph file to partition.
  --seed=<int>                             Seed to use for the PRNG.
  -e, --num_experiments=<int>              Number of experiments i.e. full kfold runs (default 1)
  -k, --kfold_iterations=<int>             Number of kfold iterations (Default: 5)
  -s, --sample=<double>                    Percentage of data that is use. Usefull if very slow on large datasets (Default: 1.0 aka use all data)
  --validation=TYPE                        Type of validation. One of {kfold, kfold_import, once, train_test_split} (Default: kfold)
  --validation_percent=<double>            Percentage of data that is use for validation (Default: 0.1)
  --validation_seperate                    Should the validation data be also used for training (Default: 'no' for kasvm  'yes' for single_level - this flag invertse the choice)
  -n, --num_nn=<int>                       Number of nearest neighbors to consider when building the graphs. (Default: 10)
  -b, --bidirectional                      Make the nearest neighbor graph bidirectional
  --stop_rule=VARIANT                      Stop rule to use. One of {simple-fix}. Default: simple-fix
  --fix_num_vert_stop=<int>                Number of vertices to fix stop coarsening at.
  --matching=TYPE                          Type of matchings to use during coarsening. One of {random, gpa, randomgpa, lp_clustering, simple_clustering, low_diameter}.
  --cluster_upperbound=<int>               Set a size-constraint on the size of a cluster. Default: none
  --label_propagation_iterations=<int>     Set the number of label propgation iterations. Default: 10.
  --diameter_upperbound=<double>           Set a size-constraint on the size of a low diameter cluster. Default: 20
  --beta=<double>                          value of the beta parameter when using low diameter clustering. (Default: 0.4)
  --refinement=TYPE                        Type of refinement. One of {ud, bayes, fix} (Default: ud)
  -C <double>                              value of the C parameter when using fix refinement. (use logarithmic scale)
  -g <double>                              value of the gamma parameter when using fix refinement. (use logarithmic scale)
  --num_skip_ms=<int>                      Size of the problem on which no model selection is skipped and only the best parameters of the previous level are used (Default: 10000)
  --no_inherit_ud                          Don't inherit the first UD sweep and do only the second UD sweep in the refinement.
  --export_graph                           Export the graph at every level (this exits after one multilevel cycle).
  --output_filename=<string>               Specify the name of the output file (that contains the partition).
  --export_model=<string>                  Specify the path of the output model (it contains the trained SVM model for later usage) ( a number and ".model" will be appended to the path).
  --timeout=<int>                          Timeout in seconds after the timeout (for a single kfold) run is readched the program is aborted (Default: 0)
  -c, --n_cores=<int>                      How many cores are used (Default: 0 aka. every core)

Example

./optimized_output/kasvm examples/twonorm

This will run the LPSVM variant on the twonorm dataset which is included in the repo.

In our paper we report BEST_AC_TEST and ~BEST_GM_TEST as the “regular” result and INIT_AC_TEST and INIT_GM_TEST as the “fast” result.

Example output:

file: examples/twonorm
num_experiments: 1
kfold_iterations: 5
sample_percent: 1
validation_type: 0
validation_percent: 0.1
validation_seperate: 0
bidirectional: 0
stop rule: 0
fix_num_vert_stop: 500
matching type: 3
cluster_upperbound: 2147483647
upper_bound_partition: 2147483647
label_iterations: 10
node_ordering: 1
diameter_upperbound: 20
beta: 0.4
refinement_type: 0
num_skip_ms: 10000
inherit_ud: 1
timeout: 0
cores: 0
seed: 0
 \/\/\/\/\/\/\/\/\/ EXPERIMENT 0 \/\/\/\/\/\/\/
io time: 0.134555
full graph - min: 3703 maj: 3697 features: 20
------------- K-FOLD ITERATION 0 -------------
fold time: 0.111028
graph - min: 2963 maj: 2958
val - min: 296 maj: 295
test - min: 740 maj: 739
calc new weights took 0.000131845
no of coarser vertices 90 and no of edges 5530
calc new weights took 0.000130177
no of coarser vertices 86 and no of edges 5198
coarsening time: 0.0519309
coarse nodes - min: 90 maj: 86
1st sweep with initial logC=0 logGamma=0
log C=  5.00	log gamma=  0.00	time=0.01  	ACC=0.501	Gmean=0.000	SVs=112 (59,53)
log C= -5.00	log gamma= -2.50	time=0.00  	ACC=0.883	Gmean=0.876	SVs=135 (68,67)
log C= 10.00	log gamma=  7.50	time=0.01  	ACC=0.501	Gmean=0.000	SVs=176 (90,86)
log C= -2.50	log gamma=  5.00	time=0.00  	ACC=0.501	Gmean=0.000	SVs=176 (90,86)
log C=  0.00	log gamma= -7.50	time=0.00  	ACC=0.970	Gmean=0.969	SVs=46 (22,24)
log C= 15.00	log gamma=  2.50	time=0.01  	ACC=0.501	Gmean=0.000	SVs=176 (90,86)
log C= 12.50	log gamma= -5.00	time=0.00  	ACC=0.959	Gmean=0.959	SVs=15 (7,8)
log C=  7.50	log gamma=-10.00	time=0.00  	ACC=0.949	Gmean=0.949	SVs=12 (4,8)
log C=  2.50	log gamma= 10.00	time=0.01  	ACC=0.501	Gmean=0.000	SVs=176 (90,86)
2nd sweep with logC=12.5 logGamma=-5
log C: 12.5 log g: -5 AC:0.959 SN:0.973 SP:0.946 PPV:0.947 NPV:0.972 F1:0.960 GM:0.959 SV_min:7 SV_maj:8 TP:288 TN:279 FP:16 FN:8
log C= 10.00	log gamma= -5.00	time=0.00  	ACC=0.959	Gmean=0.959	SVs=15 (7,8)
log C= 12.50	log gamma=  2.50	time=0.01  	ACC=0.501	Gmean=0.000	SVs=176 (90,86)
log C= 14.99	log gamma= -7.50	time=0.00  	ACC=0.946	Gmean=0.945	SVs=13 (4,9)
log C= 15.00	log gamma=  0.00	time=0.00  	ACC=0.501	Gmean=0.000	SVs=112 (59,53)
BEST (15,-7.5)
log C: 14.994 log g: -7.5 AC:0.946 SN:0.980 SP:0.912 PPV:0.918 NPV:0.978 F1:0.948 GM:0.945 SV_min:4 SV_maj:9 TP:290 TN:269 FP:26 FN:6
init train time: 0.103
inital validation on testing:
log C: 14.994 log g: -7.5 AC:0.969 SN:0.977 SP:0.961 PPV:0.961 NPV:0.977 F1:0.969 GM:0.969 SV_min:4 SV_maj:9 TP:723 TN:710 FP:29 FN:17
init test time: 0.0026
UD refinement at level 1
minority uncoarsed
uncoarsened nodes 2963 SV 4 resulting new_data 23
majority uncoarsed
uncoarsened nodes 2958 SV 9 resulting new_data 117
current level nodes min 23 maj 117
2nd sweep with logC=15 logGamma=-7.5
log C= 12.49	log gamma= -7.50	time=0.00  	ACC=0.948	Gmean=0.947	SVs=34 (14,20)
log C= 14.99	log gamma=  0.00	time=0.00  	ACC=0.501	Gmean=0.058	SVs=140 (23,117)
log C= 15.02	log gamma=-10.00	time=0.00  	ACC=0.956	Gmean=0.956	SVs=23 (11,12)
log C= 14.99	log gamma= -2.50	time=0.00  	ACC=0.508	Gmean=0.130	SVs=138 (23,115)
log C= 15.02	log gamma= -5.00	time=0.00  	ACC=0.932	Gmean=0.931	SVs=50 (15,35)
log C= 14.99	log gamma= -7.50	time=0.00  	ACC=0.948	Gmean=0.947	SVs=34 (14,20)
BEST (15,-10)
log C: 15.019 log g: -10 AC:0.956 SN:0.936 SP:0.976 PPV:0.975 NPV:0.938 F1:0.955 GM:0.956 SV_min:11 SV_maj:12 TP:277 TN:288 FP:7 FN:19
refinement at level 0 took 0.0337
refinement time 0.0338
best validation on testing data:
test time 0.00313
log C: 15.019 log g: -10 AC:0.938 SN:0.896 SP:0.980 PPV:0.978 NPV:0.904 F1:0.935 GM:0.937 SV_min:11 SV_maj:12 TP:663 TN:724 FP:15 FN:77
iteration time: 0.3027
Exporting model to ./svm0.model
------------- K-FOLD ITERATION 1 -------------
fold time: 0.1114
graph - min: 2963 maj: 2958
val - min: 296 maj: 295
test - min: 740 maj: 739
calc new weights took 0.0001
no of coarser vertices 86 and no of edges 4774
calc new weights took 0.0001
no of coarser vertices 95 and no of edges 6086
coarsening time: 0.0519
coarse nodes - min: 86 maj: 95
1st sweep with initial logC=0 logGamma=0
log C=  5.00	log gamma=  0.00	time=0.00  	ACC=0.503	Gmean=0.058	SVs=120 (60,60)
log C= -5.00	log gamma= -2.50	time=0.00  	ACC=0.981	Gmean=0.981	SVs=142 (71,71)
log C= 10.00	log gamma=  7.50	time=0.01  	ACC=0.499	Gmean=0.000	SVs=181 (86,95)
log C= -2.50	log gamma=  5.00	time=0.00  	ACC=0.499	Gmean=0.000	SVs=181 (86,95)
log C=  0.00	log gamma= -7.50	time=0.00  	ACC=0.978	Gmean=0.978	SVs=46 (22,24)
log C= 15.00	log gamma=  2.50	time=0.01  	ACC=0.499	Gmean=0.000	SVs=181 (86,95)
log C= 12.50	log gamma= -5.00	time=0.00  	ACC=0.975	Gmean=0.975	SVs=17 (5,12)
log C=  7.50	log gamma=-10.00	time=0.00  	ACC=0.963	Gmean=0.962	SVs=10 (2,8)
log C=  2.50	log gamma= 10.00	time=0.01  	ACC=0.499	Gmean=0.000	SVs=181 (86,95)
2nd sweep with logC=7.5 logGamma=-10
log C: 7.5 log g: -10 AC:0.963 SN:0.990 SP:0.936 PPV:0.939 NPV:0.989 F1:0.964 GM:0.962 SV_min:2 SV_maj:8 TP:293 TN:276 FP:19 FN:3
log C=  5.00	log gamma=-10.00	time=0.00  	ACC=0.976	Gmean=0.976	SVs=15 (7,8)
log C=  7.50	log gamma= -2.50	time=0.00  	ACC=0.905	Gmean=0.902	SVs=48 (22,26)
log C= 12.50	log gamma= -9.29	time=0.00  	ACC=0.963	Gmean=0.962	SVs=10 (2,8)
log C= 15.00	log gamma= -5.00	time=0.00  	ACC=0.975	Gmean=0.975	SVs=17 (5,12)
BEST (12.5,-9.29)
log C: 12.5 log g: -9.2908 AC:0.963 SN:0.990 SP:0.936 PPV:0.939 NPV:0.989 F1:0.964 GM:0.962 SV_min:2 SV_maj:8 TP:293 TN:276 FP:19 FN:3
init train time: 0.0955
inital validation on testing:
log C: 12.5 log g: -9.2908 AC:0.957 SN:0.986 SP:0.927 PPV:0.931 NPV:0.986 F1:0.958 GM:0.956 SV_min:2 SV_maj:8 TP:730 TN:685 FP:54 FN:10
init test time: 0.00253
UD refinement at level 1
minority uncoarsed
uncoarsened nodes 2963 SV 2 resulting new_data 13
majority uncoarsed
uncoarsened nodes 2958 SV 8 resulting new_data 85
current level nodes min 13 maj 85
2nd sweep with logC=12.5 logGamma=-9.29
log C= 10.00	log gamma= -9.29	time=0.00  	ACC=0.949	Gmean=0.949	SVs=21 (9,12)
log C= 12.50	log gamma= -1.79	time=0.00  	ACC=0.501	Gmean=0.058	SVs=98 (13,85)
log C= 14.99	log gamma=-13.87	time=0.00  	ACC=0.948	Gmean=0.948	SVs=18 (9,9)
log C= 14.98	log gamma= -4.29	time=0.00  	ACC=0.848	Gmean=0.837	SVs=56 (13,43)
log C= 15.00	log gamma= -6.79	time=0.00  	ACC=0.942	Gmean=0.942	SVs=32 (11,21)
log C= 12.50	log gamma= -9.29	time=0.00  	ACC=0.949	Gmean=0.949	SVs=21 (9,12)
BEST (15,-13.9)
log C: 14.986 log g: -13.875 AC:0.948 SN:0.939 SP:0.956 PPV:0.955 NPV:0.940 F1:0.947 GM:0.948 SV_min:9 SV_maj:9 TP:278 TN:282 FP:13 FN:18
refinement at level 0 took 0.0211
refinement time 0.0212
best validation on testing data:
test time 0.0026
log C: 12.5 log g: -9.2908 AC:0.957 SN:0.986 SP:0.927 PPV:0.931 NPV:0.986 F1:0.958 GM:0.956 SV_min:2 SV_maj:8 TP:730 TN:685 FP:54 FN:10
iteration time: 0.2828
Exporting model to ./svm1.model
------------- K-FOLD ITERATION 2 -------------
fold time: 0.1130
graph - min: 2963 maj: 2958
val - min: 296 maj: 295
test - min: 740 maj: 739
calc new weights took 0.0002
no of coarser vertices 87 and no of edges 5602
calc new weights took 0.0001
no of coarser vertices 87 and no of edges 5510
coarsening time: 0.0524
coarse nodes - min: 87 maj: 87
1st sweep with initial logC=0 logGamma=0
log C=  5.00	log gamma=  0.00	time=0.00  	ACC=0.499	Gmean=0.000	SVs=109 (53,56)
log C= -5.00	log gamma= -2.50	time=0.00  	ACC=0.909	Gmean=0.905	SVs=133 (67,66)
log C= 10.00	log gamma=  7.50	time=0.00  	ACC=0.501	Gmean=0.000	SVs=174 (87,87)
log C= -2.50	log gamma=  5.00	time=0.00  	ACC=0.501	Gmean=0.000	SVs=174 (87,87)
log C=  0.00	log gamma= -7.50	time=0.00  	ACC=0.978	Gmean=0.978	SVs=44 (21,23)
log C= 15.00	log gamma=  2.50	time=0.00  	ACC=0.501	Gmean=0.000	SVs=174 (87,87)
log C= 12.50	log gamma= -5.00	time=0.00  	ACC=0.971	Gmean=0.971	SVs=13 (7,6)
log C=  7.50	log gamma=-10.00	time=0.00  	ACC=0.968	Gmean=0.968	SVs=9 (4,5)
log C=  2.50	log gamma= 10.00	time=0.00  	ACC=0.499	Gmean=0.000	SVs=174 (87,87)
2nd sweep with logC=7.5 logGamma=-10
log C: 7.5 log g: -10 AC:0.968 SN:0.980 SP:0.956 PPV:0.957 NPV:0.979 F1:0.968 GM:0.968 SV_min:4 SV_maj:5 TP:290 TN:282 FP:13 FN:6
log C=  5.00	log gamma=-10.00	time=0.00  	ACC=0.971	Gmean=0.971	SVs=14 (7,7)
log C=  7.50	log gamma= -2.50	time=0.00  	ACC=0.975	Gmean=0.975	SVs=42 (19,23)
log C= 12.50	log gamma= -9.15	time=0.00  	ACC=0.968	Gmean=0.968	SVs=9 (4,5)
log C= 15.00	log gamma= -5.00	time=0.00  	ACC=0.971	Gmean=0.971	SVs=13 (7,6)
BEST (12.5,-9.15)
log C: 12.5 log g: -9.153 AC:0.968 SN:0.980 SP:0.956 PPV:0.957 NPV:0.979 F1:0.968 GM:0.968 SV_min:4 SV_maj:5 TP:290 TN:282 FP:13 FN:6
init train time: 0.0815
inital validation on testing:
log C: 12.5 log g: -9.153 AC:0.966 SN:0.965 SP:0.968 PPV:0.967 NPV:0.965 F1:0.966 GM:0.966 SV_min:4 SV_maj:5 TP:714 TN:715 FP:24 FN:26
init test time: 0.00231
UD refinement at level 1
minority uncoarsed
uncoarsened nodes 2963 SV 4 resulting new_data 29
majority uncoarsed
uncoarsened nodes 2958 SV 5 resulting new_data 31
current level nodes min 29 maj 31
2nd sweep with logC=12.5 logGamma=-9.15
log C= 10.00	log gamma= -9.15	time=0.00  	ACC=0.934	Gmean=0.934	SVs=19 (11,8)
log C= 12.50	log gamma= -1.65	time=0.00  	ACC=0.567	Gmean=0.368	SVs=60 (29,31)
log C= 15.02	log gamma=-15.20	time=0.00  	ACC=0.936	Gmean=0.936	SVs=15 (9,6)
log C= 15.00	log gamma= -4.15	time=0.00  	ACC=0.942	Gmean=0.942	SVs=52 (26,26)
log C= 15.00	log gamma= -6.65	time=0.00  	ACC=0.931	Gmean=0.930	SVs=24 (12,12)
log C= 12.50	log gamma= -9.15	time=0.00  	ACC=0.934	Gmean=0.934	SVs=19 (11,8)
BEST (15,-15.2)
log C: 15.019 log g: -15.2 AC:0.936 SN:0.929 SP:0.942 PPV:0.942 NPV:0.930 F1:0.935 GM:0.936 SV_min:9 SV_maj:6 TP:275 TN:278 FP:17 FN:21
refinement at level 0 took 0.0137
refinement time 0.0138
best validation on testing data:
test time 0.00233
log C: 12.5 log g: -9.153 AC:0.966 SN:0.965 SP:0.968 PPV:0.967 NPV:0.965 F1:0.966 GM:0.966 SV_min:4 SV_maj:5 TP:714 TN:715 FP:24 FN:26
iteration time: 0.2631
Exporting model to ./svm2.model
------------- K-FOLD ITERATION 3 -------------
fold time: 0.1130
graph - min: 2963 maj: 2958
val - min: 296 maj: 295
test - min: 740 maj: 739
calc new weights took 0.0001
no of coarser vertices 79 and no of edges 4394
calc new weights took 0.0001
no of coarser vertices 83 and no of edges 4904
coarsening time: 0.0518
coarse nodes - min: 79 maj: 83
1st sweep with initial logC=0 logGamma=0
log C=  5.00	log gamma=  0.00	time=0.00  	ACC=0.514	Gmean=0.174	SVs=100 (49,51)
log C= -5.00	log gamma= -2.50	time=0.00  	ACC=0.964	Gmean=0.964	SVs=130 (65,65)
log C= 10.00	log gamma=  7.50	time=0.00  	ACC=0.501	Gmean=0.058	SVs=162 (79,83)
log C= -2.50	log gamma=  5.00	time=0.00  	ACC=0.499	Gmean=0.000	SVs=162 (79,83)
log C=  0.00	log gamma= -7.50	time=0.00  	ACC=0.973	Gmean=0.973	SVs=44 (22,22)
log C= 15.00	log gamma=  2.50	time=0.00  	ACC=0.501	Gmean=0.058	SVs=161 (79,82)
log C= 12.50	log gamma= -5.00	time=0.00  	ACC=0.966	Gmean=0.966	SVs=20 (10,10)
log C=  7.50	log gamma=-10.00	time=0.00  	ACC=0.971	Gmean=0.971	SVs=13 (8,5)
log C=  2.50	log gamma= 10.00	time=0.00  	ACC=0.501	Gmean=0.058	SVs=162 (79,83)
2nd sweep with logC=7.5 logGamma=-10
log C: 7.5 log g: -10 AC:0.971 SN:0.976 SP:0.966 PPV:0.967 NPV:0.976 F1:0.971 GM:0.971 SV_min:8 SV_maj:5 TP:289 TN:285 FP:10 FN:7
log C=  5.00	log gamma=-10.00	time=0.00  	ACC=0.970	Gmean=0.970	SVs=17 (9,8)
log C=  7.50	log gamma= -2.50	time=0.00  	ACC=0.949	Gmean=0.949	SVs=40 (20,20)
log C= 12.50	log gamma=-11.54	time=0.00  	ACC=0.971	Gmean=0.971	SVs=13 (8,5)
log C= 15.00	log gamma= -5.00	time=0.00  	ACC=0.966	Gmean=0.966	SVs=20 (10,10)
BEST (12.5,-11.5)
log C: 12.5 log g: -11.537 AC:0.971 SN:0.976 SP:0.966 PPV:0.967 NPV:0.976 F1:0.971 GM:0.971 SV_min:8 SV_maj:5 TP:289 TN:285 FP:10 FN:7
init train time: 0.0807
inital validation on testing:
log C: 12.5 log g: -11.537 AC:0.974 SN:0.969 SP:0.978 PPV:0.978 NPV:0.969 F1:0.974 GM:0.974 SV_min:8 SV_maj:5 TP:717 TN:723 FP:16 FN:23
init test time: 0.00264
UD refinement at level 1
minority uncoarsed
uncoarsened nodes 2963 SV 8 resulting new_data 54
majority uncoarsed
uncoarsened nodes 2958 SV 5 resulting new_data 58
current level nodes min 54 maj 58
2nd sweep with logC=12.5 logGamma=-11.5
log C= 10.00	log gamma= -9.14	time=0.00  	ACC=0.937	Gmean=0.937	SVs=23 (6,17)
log C= 12.50	log gamma= -4.04	time=0.00  	ACC=0.944	Gmean=0.944	SVs=79 (35,44)
log C= 15.00	log gamma= -9.35	time=0.00  	ACC=0.937	Gmean=0.937	SVs=22 (6,16)
log C= 14.98	log gamma= -6.54	time=0.00  	ACC=0.936	Gmean=0.936	SVs=32 (13,19)
log C= 15.00	log gamma= -9.04	time=0.00  	ACC=0.939	Gmean=0.939	SVs=23 (6,17)
log C= 12.50	log gamma=-11.54	time=0.00  	ACC=0.941	Gmean=0.941	SVs=20 (6,14)
BEST (12.5,-11.5)
log C: 12.5 log g: -11.537 AC:0.941 SN:0.959 SP:0.922 PPV:0.925 NPV:0.958 F1:0.942 GM:0.941 SV_min:6 SV_maj:14 TP:284 TN:272 FP:23 FN:12
refinement at level 0 took 0.0207
refinement time 0.0208
best validation on testing data:
test time 0.00261
log C: 12.5 log g: -11.537 AC:0.974 SN:0.969 SP:0.978 PPV:0.978 NPV:0.969 F1:0.974 GM:0.974 SV_min:8 SV_maj:5 TP:717 TN:723 FP:16 FN:23
iteration time: 0.2691
Exporting model to ./svm3.model
------------- K-FOLD ITERATION 4 -------------
fold time: 0.1152
graph - min: 2963 maj: 2958
val - min: 296 maj: 295
test - min: 740 maj: 739
calc new weights took 0.0001
no of coarser vertices 88 and no of edges 5118
calc new weights took 0.0001
no of coarser vertices 85 and no of edges 5520
coarsening time: 0.0520
coarse nodes - min: 88 maj: 85
1st sweep with initial logC=0 logGamma=0
log C=  5.00	log gamma=  0.00	time=0.00  	ACC=0.501	Gmean=0.000	SVs=115 (59,56)
log C= -5.00	log gamma= -2.50	time=0.00  	ACC=0.662	Gmean=0.567	SVs=135 (68,67)
log C= 10.00	log gamma=  7.50	time=0.00  	ACC=0.501	Gmean=0.000	SVs=173 (88,85)
log C= -2.50	log gamma=  5.00	time=0.00  	ACC=0.501	Gmean=0.000	SVs=173 (88,85)
log C=  0.00	log gamma= -7.50	time=0.00  	ACC=0.985	Gmean=0.985	SVs=46 (22,24)
log C= 15.00	log gamma=  2.50	time=0.01  	ACC=0.501	Gmean=0.000	SVs=173 (88,85)
log C= 12.50	log gamma= -5.00	time=0.00  	ACC=0.980	Gmean=0.980	SVs=20 (10,10)
log C=  7.50	log gamma=-10.00	time=0.00  	ACC=0.983	Gmean=0.983	SVs=10 (7,3)
log C=  2.50	log gamma= 10.00	time=0.00  	ACC=0.501	Gmean=0.000	SVs=173 (88,85)
2nd sweep with logC=7.5 logGamma=-10
log C: 7.5 log g: -10 AC:0.983 SN:0.980 SP:0.986 PPV:0.986 NPV:0.980 F1:0.983 GM:0.983 SV_min:7 SV_maj:3 TP:290 TN:291 FP:4 FN:6
log C=  5.00	log gamma=-10.00	time=0.00  	ACC=0.980	Gmean=0.980	SVs=16 (10,6)
log C=  7.50	log gamma= -2.50	time=0.00  	ACC=0.621	Gmean=0.491	SVs=44 (25,19)
log C= 12.50	log gamma= -9.74	time=0.02  	ACC=0.983	Gmean=0.983	SVs=10 (7,3)
log C= 15.00	log gamma= -5.00	time=0.01  	ACC=0.980	Gmean=0.980	SVs=20 (10,10)
BEST (12.5,-9.74)
log C: 12.5 log g: -9.7435 AC:0.983 SN:0.980 SP:0.986 PPV:0.986 NPV:0.980 F1:0.983 GM:0.983 SV_min:7 SV_maj:3 TP:290 TN:291 FP:4 FN:6
init train time: 0.104
inital validation on testing:
log C: 12.5 log g: -9.7435 AC:0.983 SN:0.982 SP:0.984 PPV:0.984 NPV:0.982 F1:0.983 GM:0.983 SV_min:7 SV_maj:3 TP:727 TN:727 FP:12 FN:13
init test time: 0.0028
UD refinement at level 1
minority uncoarsed
uncoarsened nodes 2963 SV 7 resulting new_data 58
majority uncoarsed
uncoarsened nodes 2958 SV 3 resulting new_data 67
current level nodes min 58 maj 67
2nd sweep with logC=12.5 logGamma=-9.74
log C= 10.00	log gamma= -9.74	time=0.00  	ACC=0.956	Gmean=0.956	SVs=21 (10,11)
log C= 12.50	log gamma= -2.24	time=0.00  	ACC=0.819	Gmean=0.801	SVs=125 (58,67)
log C= 15.00	log gamma=-11.82	time=0.00  	ACC=0.956	Gmean=0.956	SVs=20 (10,10)
log C= 14.99	log gamma= -4.74	time=0.00  	ACC=0.961	Gmean=0.961	SVs=53 (29,24)
log C= 15.00	log gamma= -7.24	time=0.00  	ACC=0.964	Gmean=0.964	SVs=30 (15,15)
log C= 12.50	log gamma= -9.74	time=0.00  	ACC=0.956	Gmean=0.956	SVs=21 (10,11)
BEST (15,-11.8)
log C: 14.997 log g: -11.82 AC:0.956 SN:0.932 SP:0.980 PPV:0.979 NPV:0.935 F1:0.955 GM:0.956 SV_min:10 SV_maj:10 TP:276 TN:289 FP:6 FN:20
refinement at level 0 took 0.0273
refinement time 0.0274
best validation on testing data:
test time 0.00249
log C: 12.5 log g: -9.7435 AC:0.983 SN:0.982 SP:0.984 PPV:0.984 NPV:0.982 F1:0.983 GM:0.983 SV_min:7 SV_maj:3 TP:727 TN:727 FP:12 FN:13
iteration time: 0.3008
Exporting model to ./svm4.model
-------------- K-FOLD DONE -------------- 
KFOLD_TIME	0.1127
COARSE_TIME	0.0520
COARSE_MIN	86.0000
COARSE_MAJ	87.2000
HIERARCHY_MIN_SIZE	1.0000
HIERARCHY_MAJ_SIZE	1.0000
	INIT_TRAIN_TIME	0.0928
INIT_AC  	0.9662
INIT_GM  	0.9659
INIT_AC_TEST	0.9697
INIT_GM_TEST	0.9696
LEVEL0_AC	0.9472
LEVEL0_GM	0.9471
	REFINEMENT_TIME	0.0234
[BEST_INDEX]
fold 0: 1
fold 1: 0
fold 2: 0
fold 3: 0
fold 4: 0
BEST_AC	0.9682
BEST_SN	0.9723
BEST_SP	0.9641
BEST_GM	0.9680
BEST_F1	0.9683
	TEST_TIME	0.0026
BEST_AC_TEST	0.9635
BEST_SN_TEST	0.9597
BEST_SP_TEST	0.9673
BEST_GM_TEST	0.9632
BEST_F1_TEST	0.9632
TIME	0.2837

Licences