|
DETAILS
|
SimpleKNN
|
SimpleKNN
MODEL
|
| |
| |
K-nearest neighbor classifier. |
| |
| |
+ |
k
|
UINT
|
1
|
'3'
|
LOCK
|
k neighbours
|
| |
+ |
dist
|
INT
|
1
|
'0'
|
LOCK
|
distance measure function ( 0 = Eucidian )
|
| |
|
SimpleFusion
|
SimpleFusion
FUSION
|
| |
| |
Maximum / sum / product decision fusion. |
| |
| |
+ |
method
|
INT
|
1
|
'0'
|
LOCK
|
fusion method ( 0=MAXIMUM, 1=SUM, 2=PRODUCT )
|
| |
|
SVM
|
SVM
MODEL
|
| |
| |
Support vector machine classifier |
| |
| |
+ |
svm
|
INT
|
1
|
'0'
|
LOCK
|
SVM type ( C-SVC=0, nu-SVC=1)
|
| |
+ |
kernel
|
INT
|
1
|
'0'
|
LOCK
|
Kernel type ( 0=linear: u'*v, 1=polynomial: (gamma*u'*v + coef0)^degree), 2=radial basis function: exp(-gamma*|u-v|^2), 3=sigmoid: tanh(gamma*u'*v + coef0)
|
| |
+ |
degree
|
INT
|
1
|
'1'
|
LOCK
|
degree
|
| |
+ |
gamma
|
DOUBLE
|
1
|
'0.01000'
|
LOCK
|
gamma
|
| |
+ |
coef0
|
DOUBLE
|
1
|
'0.00000'
|
LOCK
|
coef0
|
| |
+ |
nu
|
DOUBLE
|
1
|
'0.50000'
|
LOCK
|
nu in nu-SVC
|
| |
+ |
C
|
DOUBLE
|
1
|
'1.00000'
|
LOCK
|
cost in C-SVC
|
| |
+ |
eps
|
DOUBLE
|
1
|
'0.10000'
|
LOCK
|
set tolerance of termination criterion
|
| |
+ |
p
|
DOUBLE
|
1
|
'0.10000'
|
LOCK
|
epsilon in loss function of epsilon-SVR
|
| |
+ |
shrink
|
INT
|
1
|
'1'
|
LOCK
|
whether to use the shrinking heuristics (0=false,1=true)
|
| |
+ |
srand
|
UINT
|
1
|
'0'
|
LOCK
|
if >0 use fixed seed to initialize random number generator, otherwise timestamp will be used
|
| |
+ |
balance
|
INT
|
1
|
'0'
|
LOCK
|
balance #samples per class (0=off, 1=remove surplus, 2=create missing)
|
| |
+ |
multicore
|
BOOL
|
1
|
'false'
|
LOCK
|
use all available CPU cores
|
| |
|
Relief
|
Relief
SELECTION
|
| |
| |
Relief feature selection. |
| |
| |
+ |
norm
|
BOOL
|
1
|
'false'
|
LOCK
|
normalize scores in interval [0..1]
|
| |
+ |
mem
|
BOOL
|
1
|
'false'
|
LOCK
|
saves memory but less efficient
|
| |
|
Rank
|
Rank
SELECTION
|
| |
| |
Rank feature selection. |
| |
| |
+ |
kfold
|
UINT
|
1
|
'2'
|
LOCK
|
#folds used during evaluation
|
| |
+ |
loo
|
BOOL
|
1
|
'false'
|
LOCK
|
use leave-one-out instead of kfold
|
| |
+ |
louo
|
BOOL
|
1
|
'false'
|
LOCK
|
use leave-one-user-out instead of kfold
|
| |
|
RandomFusion
|
RandomFusion
FUSION
|
| |
| |
Randomly picks one channel for classification, just for testing purpose. |
| |
| |
|
PCA
|
PCA
OBJECT
|
| |
| |
Principal components analysis. |
| |
| |
+ |
percentage
|
FLOAT
|
1
|
'0.00000'
|
LOCK
|
if > 0 keeps this percentage from the total variance
|
| |
|
OnlineNaiveBayes
|
OnlineNaiveBayes
MODEL
|
| |
| |
Online Naive Bayes classifier. |
| |
| |
+ |
log
|
BOOL
|
1
|
'true'
|
LOCK
|
use log normal distribution
|
| |
+ |
prior
|
BOOL
|
1
|
'true'
|
LOCK
|
use prior probability
|
| |
|
OnlineClassifier
|
OnlineClassifier
CONSUMER
|
| |
| |
Applies classifier to a stream and outputs result as an event. |
| |
| |
+ |
address
|
CHAR
|
1024
|
''
|
LOCK
|
event address (event@sender)
|
| |
+ |
baseModel
|
CHAR
|
1024
|
''
|
LOCK
|
path and name of warm start model (by default: file model.model in application directory)
|
| |
+ |
actualModel
|
CHAR
|
1024
|
'actual.model'
|
LOCK
|
name of the newest lerned model (by default: file actual.model in application directory; must be in the application directory
|
| |
+ |
modelOut
|
CHAR
|
1024
|
''
|
LOCK
|
path where incremented model will be saved (excluding filename)
|
| |
+ |
sampleOut
|
CHAR
|
1024
|
''
|
LOCK
|
path where sampleLists will be saved
|
| |
+ |
interval
|
BOOL
|
1
|
'false'
|
LOCK
|
switch for confidence interval
|
| |
+ |
intervalMin
|
FLOAT
|
1
|
'0.00000'
|
LOCK
|
Min confidence of interval (if interval = true)
|
| |
+ |
intervalMax
|
FLOAT
|
1
|
'1.00000'
|
LOCK
|
Max confidence of interval (if interval = true)
|
| |
+ |
confidence
|
FLOAT
|
1
|
'0.50000'
|
LOCK
|
Set minimal value confidence must reach to be interesting (if interval = false)
|
| |
+ |
console
|
BOOL
|
1
|
'false'
|
LOCK
|
output classification to console
|
| |
+ |
useDecisionClass
|
BOOL
|
1
|
'false'
|
LOCK
|
Switch for using decision class only to forward if false each class will be forwarded
|
| |
+ |
decisionClass
|
CHAR
|
1024
|
''
|
LOCK
|
Set name of class for which classifier should react (if useDecisionClass = true
|
| |
+ |
sname
|
CHAR
|
1024
|
''
|
LOCK
|
name of sender (if sent to event board) [deprecated, see address]
|
| |
+ |
ename
|
CHAR
|
1024
|
''
|
LOCK
|
name of event (if sent to event board) [deprecated, see address]
|
| |
+ |
valueName
|
CHAR
|
1024
|
''
|
LOCK
|
Name of value to react on
|
| |
+ |
oldSamplesUsage
|
BOOL
|
1
|
'false'
|
LOCK
|
if true: use old samples, which are not labeled by user for oldSamplesClass; if false: discard older samples not labled by user
|
| |
+ |
oldSamplesClass
|
CHAR
|
1024
|
''
|
LOCK
|
Class for which old samples should be used for training
|
| |
+ |
annoOut
|
CHAR
|
1024
|
''
|
LOCK
|
location and name of annotation file
|
| |
+ |
training
|
BOOL
|
1
|
'true'
|
LOCK
|
Enable (true) or disable (false) online training of model
|
| |
|
NaiveBayes
|
NaiveBayes
MODEL
|
| |
| |
Naive bayes classifier. |
| |
| |
+ |
log
|
BOOL
|
1
|
'true'
|
LOCK
|
user log normal distribution
|
| |
+ |
prior
|
BOOL
|
1
|
'false'
|
LOCK
|
use prior probability
|
| |
|
LDA
|
LDA
MODEL
|
| |
| |
Linear discriminant analysis classifier. |
| |
| |
+ |
norm
|
BOOL
|
1
|
'true'
|
LOCK
|
normalize probabilities
|
| |
+ |
scale
|
BOOL
|
1
|
'true'
|
LOCK
|
scale feature values
|
| |
|
KNearestNeighbors
|
KNearestNeighbors
MODEL
|
| |
| |
K-nearest neighbors classifier. |
| |
| |
+ |
k
|
UINT
|
1
|
'3'
|
LOCK
|
k neighbours
|
| |
+ |
distsum
|
BOOL
|
1
|
'false'
|
LOCK
|
instead of counting neighbors use average distance
|
| |
|
KMeans
|
KMeans
MODEL
|
| |
| |
K-Means++ implementation. |
| |
| |
+ |
k
|
UINT
|
1
|
'5'
|
LOCK
|
number of clusters
|
| |
+ |
iter
|
INT
|
1
|
'1'
|
LOCK
|
number of times to independently run k-means with different starting clusters
|
| |
+ |
pp
|
BOOL
|
1
|
'true'
|
LOCK
|
use kmeans++ instead of kmeans
|
| |
+ |
smote
|
BOOL
|
1
|
'false'
|
LOCK
|
use smote to over sample under represented classes
|
| |
+ |
norm
|
BOOL
|
1
|
'false'
|
LOCK
|
apply normalization in interval [-1,1]
|
| |
+ |
random_seed
|
BOOL
|
1
|
'true'
|
LOCK
|
apply random seed to center selection
|
| |
+ |
seed
|
INT
|
1
|
'1'
|
LOCK
|
apply seed when not using randomized center selection
|
| |
|
HierarchicalModel
|
HierarchicalModel
MODEL
|
| |
| |
A hierarchical classifier. |
| |
| |
|
FrameFusion
|
FrameFusion
MODEL
|
| |
| |
Within sample fusion. |
| |
| |
+ |
method
|
INT
|
1
|
'0'
|
LOCK
|
fusion method ( 0 = Mean )
|
| |
+ |
context
|
UINT
|
1
|
'0'
|
LOCK
|
number of context frames
|
| |
|
FloatingSearch
|
FloatingSearch
SELECTION
|
| |
| |
Floating Feature Selection (SFS, SBS, Plus-L-Minus-R, SFFS) |
| |
| |
+ |
kfold
|
UINT
|
1
|
'2'
|
LOCK
|
#folds used during evaluation
|
| |
+ |
loo
|
BOOL
|
1
|
'false'
|
LOCK
|
use leave-one-out instead of kfold
|
| |
+ |
louo
|
BOOL
|
1
|
'false'
|
LOCK
|
use leave-one-user-out instead of kfold
|
| |
+ |
split
|
FLOAT
|
1
|
'0.00000'
|
LOCK
|
use the first split% for training and the rest for testing instead of kfold (split = ]0..1[, selected if > 0)
|
| |
+ |
nfirst
|
UINT
|
1
|
'0'
|
LOCK
|
terminate after n first (0 for all)
|
| |
+ |
method
|
INT
|
1
|
'0'
|
LOCK
|
method (0=SFS, 1=SBS, 2=PLUS-L-MINUS-R, 3=SFFS)
|
| |
+ |
l
|
UINT
|
1
|
'1'
|
LOCK
|
plus l
|
| |
+ |
r
|
UINT
|
1
|
'0'
|
LOCK
|
minus r
|
| |
+ |
eval
|
INT
|
1
|
'0'
|
LOCK
|
evaluation method (0=CLASSWISE, 1=ACCURACY)
|
| |
+ |
nthread
|
INT
|
1
|
'0'
|
LOCK
|
distribute work on n threads
|
| |
|
FloatingCFS
|
FloatingCFS
SELECTION
|
| |
| |
Floating Correlation based Feature Selection (SFS, SBS, Plus-L-Minus-R, SFFS) |
| |
| |
+ |
nfirst
|
UINT
|
1
|
'0'
|
LOCK
|
terminate after n first (0 for all)
|
| |
+ |
method
|
INT
|
1
|
'0'
|
LOCK
|
method (0=SFS, 1=SBS, 2=PLUS-L-MINUS-R, 3=SFFS)
|
| |
+ |
l
|
UINT
|
1
|
'1'
|
LOCK
|
plus l
|
| |
+ |
r
|
UINT
|
1
|
'0'
|
LOCK
|
minus r
|
| |
|
Fisher
|
Fisher
OBJECT
|
| |
| |
Fisher projection based on linear discriminant analysis. |
| |
| |
+ |
n
|
UINT
|
1
|
'0'
|
LOCK
|
if > 0 keeps only the first n dimensions
|
| |
|
Dollar$1
|
Dollar$1
MODEL
|
| |
| |
Dollar$1 classifier. |
| |
| |
+ |
indx
|
INT
|
1
|
'0'
|
LOCK
|
dimension of x coordinate (0)
|
| |
+ |
indy
|
INT
|
1
|
'1'
|
LOCK
|
dimension of y coordinate (1)
|
| |
+ |
norm
|
BOOL
|
1
|
'false'
|
LOCK
|
recognition probabilities to sum up to 1
|
| |
|
DecisionSmoother
|
DecisionSmoother
OBJECT
|
| |
| |
Returns smoothed decisions at a regular update rate |
| |
| |
+ |
address
|
CHAR
|
1024
|
''
|
LOCK
|
event address (event@sender)
|
| |
+ |
sname
|
CHAR
|
1024
|
'decision'
|
LOCK
|
name of sender (if sent to event board) [deprecated, see address]
|
| |
+ |
ename
|
CHAR
|
1024
|
'smoothed'
|
LOCK
|
name of event (if sent to event board) [deprecated, see address]
|
| |
+ |
update
|
UINT
|
1
|
'0'
|
LOCK
|
update rate in ms (if 0 an event is only sent after a new event was received)
|
| |
+ |
decay
|
FLOAT
|
1
|
'0.00000'
|
FREE
|
factor by which the target decision will be decreased if no new decision arrives
|
| |
+ |
speed
|
FLOAT
|
1
|
'0.00000'
|
FREE
|
factor by which the smoothed decision will move towards the target decision (if <= 0 it will be immediatelly reached)
|
| |
+ |
average
|
BOOL
|
1
|
'false'
|
FREE
|
target decision is replaced with cumulative moving average of previous decisions
|
| |
+ |
window
|
DOUBLE
|
1
|
'0.00000'
|
FREE
|
window size in seconds over which previous decisions will be considered (0 -> infinite)
|
| |
|
Collector
|
Collector
CONSUMER
|
| |
| |
Transforms current stream to a sample and stores it to a sample list. |
| |
| |
+ |
user
|
CHAR
|
1024
|
''
|
LOCK
|
user name
|
| |
+ |
classes
|
CHAR
|
1024
|
''
|
LOCK
|
list of class names (separated by blank)
|
| |
+ |
iter
|
INT
|
1
|
'5'
|
LOCK
|
number of iterations per class
|
| |
+ |
ask
|
BOOL
|
1
|
'false'
|
LOCK
|
always ask before recording next sample
|
| |
|
ClassifierT
|
ClassifierT
FEATURE
|
| |
| |
Applies classifier to a stream and continuously outputs result to a new stream. |
| |
| |
+ |
trainer
|
CHAR
|
1024
|
''
|
LOCK
|
filepath of trainer
|
| |
+ |
flat
|
BOOL
|
1
|
'false'
|
LOCK
|
in case of multiple samples merge to single sample
|
| |
|
Classifier
|
Classifier
CONSUMER
|
| |
| |
Applies classifier to a stream and outputs result as an event. |
| |
| |
+ |
address
|
CHAR
|
1024
|
''
|
LOCK
|
event address (event@sender)
|
| |
+ |
sname
|
CHAR
|
1024
|
''
|
LOCK
|
name of sender (if sent to event board) [deprecated, see address]
|
| |
+ |
ename
|
CHAR
|
1024
|
''
|
LOCK
|
name of event (if sent to event board) [deprecated, see address]
|
| |
+ |
path
|
CHAR
|
1024
|
''
|
LOCK
|
path to trainer 'name:filepath' (if several separate by ;)
|
| |
+ |
pthres
|
FLOAT
|
1
|
'0.00000'
|
LOCK
|
probablity threshold
|
| |
+ |
merge
|
BOOL
|
1
|
'false'
|
LOCK
|
in case of multiple streams merge to single stream
|
| |
+ |
flat
|
BOOL
|
1
|
'false'
|
LOCK
|
in case of multiple samples merge to single sample
|
| |
+ |
console
|
BOOL
|
1
|
'false'
|
LOCK
|
output classification to console
|
| |
+ |
winner
|
BOOL
|
1
|
'false'
|
LOCK
|
send winning class only
|
| |
+ |
select
|
CHAR
|
1024
|
''
|
LOCK
|
foward only specific classes (indices separated by ',') [ignored if winner=true]
|
| |
+ |
trainer
|
CHAR
|
1024
|
''
|
LOCK
|
filepath of trainer [deprecated use 'path']
|
| |
|