read.liquidSVM: Read and Write Solution from and to File

In liquidSVM: A Fast and Versatile SVM Package

Description

Reads or writes the solution from or to a file. The format of the solutions is the same as used in the command line version of liquidSVM. In addition also configuration data is written and by default also the training data. This can be interchanged also with the other bindings.

Usage

read.liquidSVM(filename, ...)

write.liquidSVM(model, filename)

serialize.liquidSVM(model, writeData = TRUE)

unserialize.liquidSVM(obj, ...)

Arguments

filename	the filename to read from/save to. Can be relative to the working directory.
...	passed to init.liquidSVM
model	the model
writeData	whether the training data should be serialized in the stream
obj	the data to unserialize

Details

The command line version of liquidSVM saves solutions after select in files of the name data.sol or data.fsol and uses those in the test-phase. read.liquidSVM and write.liquidSVM read and write the same format at the specified path. If you give a filename using extension .fsol the training data is written to the file and read from it. On the other hand, if you use the .sol format, you need to be able to reproduce the same data again once you read the solution. readSolution creates a new svm object.

Note

This is not tested thoroughly so use in production is at your own risk. Furthermore the serialize/unserialize hooks write temporary files.

See Also

init.liquidSVM, write.liquidSVM

Examples

## Not run: 
banana <- liquidData('banana-bc')
modelOrig <- mcSVM(Y~., banana$train)
write.liquidSVM(modelOrig, "banana-bc.fsol")
write.liquidSVM(modelOrig, "banana-bc.sol")
clean(modelOrig) # delete the SVM object

# now we read it back from the file
modelRead <- read.liquidSVM("banana-bc.fsol")
# No need to train/select the data!
errors(test(modelRead, banana$test))

# to read the model where no data was saved we have to make sure, we get the same training data:
banana <- liquidData('banana-bc')
# then we can read it
modelDataExternal <- read.liquidSVM("banana-bc.sol", Y~., banana$train)
result <- test(modelDataExternal, banana$test)

# to serialize an object use:
banana <- liquidData('banana-bc')
modelOrig <- mcSVM(Y~., banana$train)
# we serialize it into a raw vector
obj <- serialize.liquidSVM(modelOrig)
clean(modelOrig) # delete the SVM object

# now we unserialize it from that raw vector
modelUnserialized <- unserialize.liquidSVM(obj)
errors(test(modelUnserialized, banana$test))

## End(Not run)

liquidSVM documentation built on Sept. 15, 2019, 1:02 a.m.