The Gun-Point Data

Description

This is the Gun-Point data from The UCR Time Series Database.

Usage

data(GunPoint)

Format

GunPoint is a list with one training time series dataset and one test time series dataset provided as separate matrices. There are 50 cases (rows) for training dataset with 150 variables (columns). Similarly there are 150 cases for test dataset with 150 variables. Variables are representing the observations over time. In other words, they are ordered so that a row is a univariate time series. Originally, this is a classification problem where there are two classes. Therefore, list stores the class information for both training and test time series. This information is stored in arrays of length 50 and 150 for training and test time series respectively (so each time series is associated with a class).

Description by Chotirat Ann Ratanamahatana and Eamonn Keogh in their publication “Everything you know about Dynamic Time Warping is Wrong“ is as follows:

“...This dataset comes from the video surveillance domain. The dataset has two classes, each containing 100 instances. All instances were created using one female actor and one male actor in a single session. The two classes are: Gun-Draw: The actors have their hands by their sides. They draw a replicate gun from a hip-mounted holster, point it at a target for approximately one second, then return the gun to the holster, and their hands to their sides. Point: The actors have their gun by their sides. They point with their index fingers to a target for approximately one second, and then return their hands to their sides. For both classes, we tracked the centroid of the actor's right hands in both X- and Y-axes, which appear to be highly correlated; therefore, in this experiment, we only consider the X-axis for simplicity...“

Author(s)

Mustafa Gokce Baydogan

Source

The original data is at http://www.cs.ucr.edu/~eamonn/time_series_data/.

References

Ratanamahatana, C. A. and Keogh. E. (2004). Everything you know about Dynamic Time Warping is Wrong. In proceedings of SIAM International Conference on Data Mining (SDM05), pp.506-510 Newport Beach, CA, April 21-23

See Also

learnPattern, computeSimilarity

Examples

data(GunPoint)
set.seed(71)

## Learn patterns on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries)
print(ensemble)

## Find the similarity between test and training series based on the learned model
similarity=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries)

## Find the index of 1 nearest neighbor (1NN) training series for each test series
NearestNeighbor=apply(similarity,1,which.min)

## Predicted class for each test series
predicted=GunPoint$trainclass[NearestNeighbor]
print(predicted)

Show the NEWS file

Description

Show the NEWS file of the LPStimeSeries package.

Usage

LPSNews()

Value

None.

Compute similarity between time series based on learned patterns

Description

Compute similarity between time series. Raw time series can be provided together with learnPattern object so that the representation for the time series are generated internally and similarity is computed based on these representations. The other option is to provide the representations (instead of raw time series) and to compute the similarity without a need for learnPattern object.

Usage

computeSimilarity(object=NULL,testseries=NULL,refseries=NULL,
   maxdepth=NULL,which.tree=NULL,sim.type=0, terminal=TRUE,
   testrepresentation,refrepresentation,
   nthreads=1, normalize=FALSE)

Arguments

Value

A similarity matrix of size “the number of test series“ by “the number of reference series“ is returned. Similarity between test series and reference series is defined as the number of mismatching patterns based on the representation generated by the trees. See LPS paper for details.

Note

Similarity matrix can also be computed over representations if it is generated using predict.learnPattern. This will probably take longer time compared to computing the similarity directly using the ensemble. However, if you are using LPS for retrieval purposes, bounding schemes (such as early abondon) can be used (requires further implementation) with the learned representations.

Author(s)

Mustafa Gokce Baydogan

References

Baydogan, M. G. and Runger, G. (2016), “Time Series Representation and Similarity Based on Local Autopatterns“, Data Mining and Knowledge Discovery, 30(2), 476-509. doi:10.1007/s10618-015-0425-y.

See Also

learnPattern, predict.learnPattern

Examples

data(GunPoint)
set.seed(71)
## Learn patterns on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries)

## Find the similarity between test and training series
sim=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries)

## Find similarity using representations, 
## First generate representations
trainRep=predict(ensemble, GunPoint$trainseries, nodes=TRUE)
testRep=predict(ensemble, GunPoint$testseries, nodes=TRUE)

## Then compute the similarity (city-block distance), 
## takes longer but we keep the representation
sim2=computeSimilarity(testrepresentation=testRep,refrepresentation=trainRep)

## Find the similarity based on first 100 trees
sim=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries,which.tree=c(1:100))

Find and Visualize the Most Class-Discriminative Patterns

Description

Identifies which terminal nodes (learned autopatterns) from a learnPattern ensemble are most discriminative for distinguishing between classes. Patterns are ranked by a chi-squared-like statistic that measures how much the per-class mean frequency in each terminal node deviates from the overall mean. The top patterns can be plotted with series colored by class label.

Usage

discriminativePatterns(object, x, classes, n=5, plot=TRUE,
   orient=c(2,2), palette=NULL)

Arguments

Value

A list (returned invisibly) with the following components:

Note

The discriminability score for terminal node j is computed as

\sum_c n_c (\bar{f}_{cj} - \bar{f}_j)^2 / \bar{f}_j

where n_c is the number of series in class c, \bar{f}_{cj} is the mean frequency of class c in node j, and \bar{f}_j is the overall mean frequency. Nodes where one class lands much more (or less) often than others receive high scores.

When plot=TRUE, the predictor (triangle) and target (circle) segments are overlaid for all series, colored by class. At most prod(orient) patterns are plotted per page.

Author(s)

Mustafa Gokce Baydogan

References

Baydogan, M. G. and Runger, G. (2016), “Time Series Representation and Similarity Based on Local Autopatterns”, Data Mining and Knowledge Discovery, 30(2), 476-509. doi:10.1007/s10618-015-0425-y.

See Also

learnPattern, visualizePattern, predict.learnPattern, plotMDS

Examples

data(GunPoint)
set.seed(71)
ensemble <- learnPattern(GunPoint$trainseries)

## Find the 4 most discriminative patterns on the training set
dp <- discriminativePatterns(ensemble, GunPoint$trainseries,
                             GunPoint$trainclass, n=4)

## Inspect the top pattern scores
dp$scores[dp$top]

Extract a single tree from the ensemble.

Description

This function extracts the structure of a tree from a learnPattern object.

Usage

getTreeInfo(object, which.tree=1)

Arguments

Value

is a list with the following components:

The seven columns of the tree structure matrix are:

Note

For numerical predictors, data with values of the variable less than or equal to the splitting point go to the left daughter node.

Author(s)

Mustafa Gokce Baydogan

See Also

learnPattern

Examples

data(GunPoint)
set.seed(71)

## Learn patterns on GunPoint training series with 50 trees
ensemble=learnPattern(GunPoint$trainseries,ntree=50)
getTreeInfo(ensemble, 3)

Learn Local Auto-Patterns for Time Series Representation and Similarity

Description

learnPattern implements ensemble of regression trees (based on Breiman and Cutler's original Fortran code) to learn local auto-patterns for time series representation. Ensemble of regression trees are used to learn an autoregressive model. A local time-varying autoregressive behavior is learned by the ensemble.

Usage

## Default S3 method:
learnPattern(x,
   segment.factor=c(0.05,0.95),
   random.seg=TRUE, target.diff=TRUE, segment.diff=TRUE, 
   random.split=0,
   ntree=200,
   mtry=1,
   replace=FALSE,
   sampsize=if (replace) ceiling(0.632*nrow(x)) else nrow(x),
   maxdepth=6,
   nodesize=5,
   do.trace=FALSE,
   keep.forest=TRUE,
   oob.pred=FALSE,
   keep.errors=FALSE,
   keep.inbag=FALSE,
   nthreads=1,
   wrap=TRUE, ...)
## S3 method for class 'learnPattern'
print(x, ...)

Arguments

Value

An object of class learnPattern, which is a list with the following components:

Note

OOB predictions may have missing values (i.e. NA) if time series is not left out-of-bag during computations. Even, it is left out-of-bag, there is a potential of some observations (i.e. time frames) not being selected as the target. In such cases, there will no OOB predictions.

Author(s)

Mustafa Gokce Baydogan baydoganmustafa@gmail.com, based on original Fortran code by Leo Breiman and Adele Cutler, R port by Andy Liaw and Matthew Wiener.

References

Baydogan, M. G. and Runger, G. (2016), “Time Series Representation and Similarity Based on Local Autopatterns“, Data Mining and Knowledge Discovery, 30(2), 476-509. doi:10.1007/s10618-015-0425-y.

Breiman, L. (2001), Random Forests, Machine Learning 45(1), 5-32.

See Also

predict.learnPattern, computeSimilarity, tunelearnPattern

Examples

data(GunPoint)
set.seed(71)

## Learn patterns on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries)
print(ensemble)

## Find the similarity between test and training series based on the learned model
similarity=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries)

## Find the index of 1 nearest neighbor (1NN) training series for each test series
NearestNeighbor=apply(similarity,1,which.min)

## Predicted class for each test series
predicted=GunPoint$trainclass[NearestNeighbor]

## Compute the percentage of accurate predictions
accuracy=sum(predicted==GunPoint$testclass)/nrow(GunPoint$testseries)
print(100*accuracy)

## Learn patterns randomly on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries, random.split=1)

## Find the similarity between test and training series and classify test series
similarity=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries)
NearestNeighbor=apply(similarity,1,which.min)
predicted=GunPoint$trainclass[NearestNeighbor]
accuracy=sum(predicted==GunPoint$testclass)/nrow(GunPoint$testseries)
print(100*accuracy)

## Learn patterns by training each tree on a random subsample
## and classify test time series
ensemble=learnPattern(GunPoint$trainseries,replace=TRUE)
similarity=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries)
NearestNeighbor=apply(similarity,1,which.min)
predicted=GunPoint$trainclass[NearestNeighbor]
print(predicted)

## Learn patterns and do predictions on OOB time series
ensemble=learnPattern(GunPoint$trainseries,replace=TRUE,target.diff=FALSE,oob.pred=TRUE)
## Plot first series and its OOB approximation
plot(GunPoint$trainseries[1,],xlab='Time',ylab='Observation',
	type='l',lty=1,lwd=2)
points(c(1:ncol(GunPoint$trainseries)),ensemble$oobpredictions[1,],
	type='l',col=2,lty=2,lwd=2)
legend('topleft',c('Original series','Approximation'),
	col=c(1,2),lty=c(1,2),lwd=2)

Plot method for learnPattern objects

Description

Plot the MSE of a learnPattern object over trees based on out-of-bag predictions

Usage

## S3 method for class 'learnPattern'
plot(x, type="l", main=deparse(substitute(x)), ...)

Arguments

Value

Invisibly, MSE of the learnPattern object.

Note

This function does not work for learnPattern if oob.predict=FALSE during training.

Author(s)

Mustafa Gokce Baydogan

See Also

learnPattern

Examples

data(GunPoint)
ensemble=learnPattern(GunPoint$trainseries,oob.pred=TRUE,replace=TRUE)
plot(ensemble)

Multi-dimensional Scaling Plot of Learned Pattern Similarity

Description

Plot the scaling coordinates of the Learned Pattern Similarity.

Usage

plotMDS(object, newdata, classinfo, k=2, palette=NULL, pch=20, ...) 

Arguments

Value

The output of cmdscale on scaled Learned Pattern similarity is returned invisibly.

Note

If k > 2, pairs is used to produce the scatterplot matrix of the coordinates.

The entries of the similarity matrix is divided by the maximum possible similarity which is 2*sum(object$nobs)

Author(s)

Mustafa Gokce Baydogan

See Also

learnPattern

Examples

set.seed(1)
data(GunPoint)
## Learn patterns on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries)
plotMDS(ensemble, GunPoint$trainseries,GunPoint$trainclass)

## Using different symbols for the classes:
plotMDS(ensemble, GunPoint$trainseries,GunPoint$trainclass, 
         palette=rep(1, 2), pch=as.numeric(GunPoint$trainclass))
         
## Learn patterns on GunPoint training series with random splits
ensemble=learnPattern(GunPoint$trainseries,random.split=1)
plotMDS(ensemble, GunPoint$trainseries,GunPoint$trainclass,main='Random Splits')

predict method for learnPattern objects

Description

Representation generation for test data using learnPattern.

Usage

## S3 method for class 'learnPattern'
predict(object, newdata, which.tree=NULL,
   nodes=TRUE, maxdepth=NULL, normalize=FALSE, ...)

Arguments

Value

Returns the learned pattern representation for the time series in the dataset if nodes is set TRUE. Basically, it is the count of observed patterns at each terminal node. Otherwise predicted values for each time series in newdata are returned.

Author(s)

Mustafa Gokce Baydogan

References

Baydogan, M. G. and Runger, G. (2016), “Time Series Representation and Similarity Based on Local Autopatterns“, Data Mining and Knowledge Discovery, 30(2), 476-509. doi:10.1007/s10618-015-0425-y.

Breiman, L. (2001), Random Forests, Machine Learning 45(1), 5-32.

See Also

learnPattern

Examples

data(GunPoint)
set.seed(71)
## Learn patterns on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries)

## Find representations
trainRep=predict(ensemble, GunPoint$trainseries, nodes=TRUE)
testRep=predict(ensemble, GunPoint$testseries, nodes=TRUE)

## Check size of the representation for training data
print(dim(trainRep))

## Learn patterns on GunPoint training series (target cannot be difference series)
ensemble=learnPattern(GunPoint$trainseries,target.diff=FALSE)

## Predict observations for test time series
predicted=predict(ensemble,GunPoint$testseries,nodes=FALSE)

## Plot an example test time series 
plot(GunPoint$testseries[5,],type='l',lty=1,xlab='Time',ylab='Observation',lwd=2)
points(c(1:ncol(GunPoint$testseries)),predicted$predictions[5,],type='l',col=2,lty=2,lwd=2)
legend('topleft',c('Original series','Approximation'),col=c(1,2),lty=c(1,2),lwd=2)

Tune Parameters of LPS for Time Series Classification

Description

tunelearnPattern implements parameter selection for LPS in time series classification problems. LPS requires the setting of segment length (if segment length is not random) and depth parameter. Given training time series and alternative parameter settings, the best set of parameters that minimizes the cross-validation error rate is returned.

Usage

tunelearnPattern(x, y, unlabeledx=NULL, nfolds=5,
   segmentlevels=c(0.25,0.5,0.75), random.split=0,
   mindepth=4, maxdepth=8, depthstep=2,
   ntreeTry=25, target.diff=TRUE, segment.diff=TRUE,
   nthreads=1, ...)

Arguments

Value

A list with the following components:

Author(s)

Mustafa Gokce Baydogan baydoganmustafa@gmail.com, based on original Fortran code by Leo Breiman and Adele Cutler, R port by Andy Liaw and Matthew Wiener.

References

Baydogan, M. G. and Runger, G. (2016), “Time Series Representation and Similarity Based on Local Autopatterns“, Data Mining and Knowledge Discovery, 30(2), 476-509. doi:10.1007/s10618-015-0425-y.

Breiman, L. (2001), Random Forests, Machine Learning 45(1), 5-32.

See Also

learnPattern, computeSimilarity

Examples

data(GunPoint)
set.seed(71)

## Tune segment length level and depth on GunPoint training series
tuned=tunelearnPattern(GunPoint$trainseries,GunPoint$trainclass)
print(tuned$best.error)
print(tuned$best.seg)
print(tuned$best.depth)

## Use tuned parameters to learn patterns
ensemble=learnPattern(GunPoint$trainseries,segment.factor=tuned$best.seg,
					  maxdepth=tuned$best.depth)

## Find the similarity between test and training series based on the learned model
similarity=computeSimilarity(ensemble,GunPoint$testseries,GunPoint$trainseries)

## Find the index of 1 nearest neighbor (1NN) training series for each test series
NearestNeighbor=apply(similarity,1,which.min)

## Predicted class for each test series
predicted=GunPoint$trainclass[NearestNeighbor]

## Compute the percentage of accurate predictions
accuracy=sum(predicted==GunPoint$testclass)/nrow(GunPoint$testseries)
print(100*accuracy)

Plot of the patterns learned by the ensemble of the regression trees

Description

visualizePattern visualizes the patterns implied by the terminal nodes of the trees from learnPattern object.

Usage

visualizePattern(object, x, which.terminal, orient=c(2,2))

Arguments

Value

A list with the following components are returned invisibly.

Note

Patterns are visualized for the time series for which the frequency of the observations in the pattern is the largest. If more than one plot is requested through the setting of orient, the patterns are plotted for the time series based on the descending order of the frequency.

Currently, patterns are visualized based on the first predictor segment (sampled at the root node). This visualization can be done based on the predictor segment sampled at each level of the tree.

predictor and target are of size x where the patterns are numerical values and the rest of the entries are NAs.

Author(s)

Mustafa Gokce Baydogan

See Also

learnPattern,predict.learnPattern

Examples

set.seed(71)
data(GunPoint)
## Learn patterns on GunPoint training series with default parameters
ensemble=learnPattern(GunPoint$trainseries)

## Find representations
trainRep=predict(ensemble, GunPoint$trainseries, nodes=TRUE)

## Find the average frequency over the terminal nodes
avgFreq=apply(trainRep,2,mean)

## Find the terminal node that has the maximum average and visualize
termid=which.max(avgFreq)
visualizePattern(ensemble,GunPoint$trainseries,termid,c(2,1))