operations Package

operations Package

Large parameterized processing task, divided into smaller parts with different parameters

Each operation consists of a set of processes. Processes need to be independent so that potentially they can be executed concurrently on different machines. For instance, an operation might consist of comparing the performance of a NodeChain on an EEG dataset for different parameter values of the NodeChain nodes (based on 10 runs of ten-fold crossvalidation). Then, a process would be to do one run of ten-fold crossvalidation of the NodeChain on the test dataset for one specific parameter setting. An operation can take several input datasets that need to contain some meta information.

Naming Conventions

The mapping between operations and the type parameter in their specification file has been automatized. For each operation there is a separate module with the same name as the type. Furthermore, the type should not use any capital letters and be written with underscores, if several subnames are required as in node_chain. The corresponding class name of the operation should be the default transformation form under_score to CamelCase. Here it is important to mention that abbreviations are not written with capital letters.

Documentation Conventions

The user documentation of each operation has to be at the docstring of the module. It is structured using sections and subsections.

It is important to define the sections Specification file Parameters and Exemplary Call. An exhaustive example can be found at: concatenate.

Programming your own Operation

Each operation module consists of an operation, named as mentioned above, and a process. The operation defines the mapping between the specification and the elemental tasks, called processes.

Each operation inherits from Operation and each process from Process. The __init__ method should always call its corresponding base part. Processes should be generated, by using and defining a classmethod called: _createProcesses. Each operation has to define the classmethod create and the method consolidate. A process itself basically just needs to define its __call__ and __init__ method. The latter is needed to define the parameter mappings, since there is currently no standard definition of parameters of a process.

Modules

analysis	Create one plot for each possible parameter combination from a PerformanceResultSummary
base	Interface for the numerous operations and processes
comp_analysis	Creates various comparing plots on several levels for a PerformanceResultSummary
concatenate	Concatenate datasets of time series data
generic	Use a generic external python script in a process
merge	Define train and test data for One versus Rest or Rest versus One in cross validation fashion
mmlf	Execute MMLF experiments
node_chain	Interface to node_chain using the BenchmarkNodeChain
shuffle	Take combinations of datasets in the summary for training and test each
statistic	Calculate a two-tailed paired t-test on a result collection for a certain parameter
weka_classification	Classification using the WEKA experimenter
weka_filter	Use Weka’s Filter that transform one arff file into another.