nodes Package

nodes Package

Nodes are elemental signal processing steps

They are arranged in mostly serial node chains. Between the nodes in one chain several signal types can be sent:

There is one BaseNode for all nodes in the package module and so every node only specifies the relevant transformation methods. A basic overview on the overloading of methods can be found in the templates module. All nodes are grouped in extra packages, depending on their main processing category. Bigger differences in the algorithms’ concepts are considered with the module structure or even further subpackages.

The (Node Class) - (Configuration File) - (Node Name) Mapping

A complete list of all nodes and their mapping can be found at: List of all Nodes.

The __init__ of this package imports all existing nodes and maps their names, given by the dict _NODE_MAPPING of each imported file, to their class name. These are the optional names. The standard name of a node is the class name, which has to end with Node. Furthermore you can use the class name without this ending as alternative name. Every node has an exemplary call, where you get to know the basic configuration structure. And there is also a list of all possible usable names at the end of the basic description of the node. The mapping is used in the definition of node chains when writing the yaml configuration files. Especially when comparing different algorithms it is useful to use their short names as parameters to avoid long names in the folder names and the final comparison graphics.

The advantage is, that most import statements are done at the beginning and that the user just needs to know the methods name and not the corresponding class and module name. The disadvantage is, that maybe not used packages are imported or that import errors occur, because certain packages are not installed. The import errors are mainly prevented by throwing exceptions.

../../_images/node.png

See also

Modules

base_node Skeleton for an elemental transformation of the signal
decorators Define parameter distributions for pySPACE nodes.
external Import wrapper to get access to externally defined nodes
scikit_nodes Wrap the algorithms defined in scikit.learn in pySPACE nodes
templates Tell the developer about general coding and documentation approaches for nodes

Subpackages Summary

classification Classification of the incoming signal
data_selection Selection of data
debug Support nodes for debugging of node chains, other nodes, etc.
feature_generation Generate features from a time series (amplitudes or frequency spectrum for example)
feature_selection Select features by learning algorithms or simple name filtering
meta Nodes, wrapping other groups of nodes or node chains
postprocessing Final modification or clean up of FeatureVector and PredictionVector
preprocessing Standard preprocessing of the incoming TimeSeries
regression Regression of the incoming signal
sink Collect incoming signal types for further processing or to store in datasets
source Load a special signal or data type as a stream of samples
spatial_filtering Erase and/or recombine channels of multivariate TimeSeries
splitter Control how data is split into training and testing data
type_manipulation Manipulation of the data data_types
visualization Visualize the single different data samples or averages

Subpackages Tree