SVM Definition (.svm)
An SVM model definition file (.svm) is generated for each SVM ensemble, named as:
<data file name>_I<number of descriptors>.svm
The SVM file contains a record of the SVM parameters for each model in an SVM ensemble. There is one SVM file for each number of descriptors modeled, and each SVM file describes the ensemble of the best models built for each fold of the training pool.
Common information
# ADMET Modeler(TM) <version> SVM <number of members> <number of inputs>
#================================================
type SVM
ProcessType <process_type>
NData <number of all training points>
NInputs <number of descriptors>
N_class <number of classes>
AllTrainingData
<descriptor1> <descriptor2> … <descriptorn>
…
InputNames <descriptorname1> <descriptorname2> … <descriptornamen>
InputShift <descriptorshift1> <descriptorshift2> … <descriptorshiftn>
InputScale <descriptorshift1> <descriptorshift2> … <descriptorshiftn>
OutputShift <outputshift>
OutputScale <outputscale>
SVOutputCategories
<index1> <index2>…
<label1> <label2> …
#================================================
Member model
C * <c_value>*
Gamma * <gamma_value>*
Nu * <nu_value>*
TotalSV * <total number of support vectors>*
NSV * <nsv1> <nsv2>…*
SVIndex * <index1> <index2> … <index>*
Alpha * <value1> <value2> … <valuem>*
Rho * <value1> …*
TrainingStats * <value1> <value2> … <value8>*
VerifyStats * <value1> <value2> … <value8>*
TestStats * <value1> <value2> … <value8>*
Merit * <value>*
#--------------------------------------------------------------------
…
The “member model” section is repeated for every member model.
A sample SVM file follows. Ellipses (...) show where lines have been omitted.
# ADMET Modeler(TM) 2.2.5 SVM 32 16
# ================================================================================
type SVM
ProcessType 2
NData 139
NInputs 16
N_class 2
AllTrainingData
0.319063 0.584373 -0.077506 -0.417692 -1.000000
0.381083 0.575733 -0.01992 0.426238 -0.305645
0.470742 0.602817 0.006429 0.563063 -1.000000
0.463321 0.971684 -0.068765 0.271627 -1.000000
…
InputNames SssCH2 SsssCH SdssC SaasC SaaN
InputShift -0.969109 0.573648 -0.068765 -0.404238 -1.
InputScale 0.182836 0.168351 0.435127 0.204621 0.153246
OutputShift 0.
OutputScale 1.
SVOutputCategories
0 1
P N
# ======================================================================
C 0.
Gamma 0.000794
Nu 0.069409
TotalSV 10
NSV 6 4
SVIndex 1 3 11 50 64 75 6 29 130 135
Alpha
6199.697005 3282.504644 1548.03416 1064.334727 835.59332
Rho 0.812494
TrainingStats 0. 0. 0. 0. 0. 0. 0. 0.127907
VerifyStats 0. 0. 0. 0. 0. 0. 0. 0.122807
TestStats 0. 0. 0. 0. 0. 0. 0. 0.112676
Merit 0.127907
# --------------------------------------------------------------------------------
…