The following node is available in the Open Source KNIME predictive analytics and data mining platform version 2.7.1. Discover over 1000 other nodes, as well as enterprise functionality at
http://knime.com.
MDS Projection
This node maps data of a high dimensional space onto a lower (usually
2 or 3) dimensional space with respect to a set of fixed data points.
Therefore modified Sammons mapping is applied, which iteratively
decreases the difference of the distances of high and low dimensional
data. When adjusting the position a low dimensional data point by default
not its neighbors (or all other data points) are taken into account but a
specified set of fixed data points which are not modified. Additionally
the data points (and not only the fixed points) can be taken into account
when adjusting its positions, therefore the setting "Project only"
has to be unchecked. If the setting is checked the data points will be
mapped only with respect to the fixed data, which we call a projection.
The algorithm converges like a common mds algorithm due to a
decreasing learning rate.
Dialog Options
- Number of rows to use
-
Specifies the number of rows to apply the MDS on.
- Output dimension
-
Specifies the dimension of the mapped output data.
- Epochs
-
Specifies the number of epochs to train.
- Learn rate
-
Specifies the learning rate to use. The learning rate
is decreased automatically over the trained epochs.
- Random seed
-
Specifies the random seed to use, which allows to reproduce a mapping
even if the initialization is done randomly.
- Distance metric
-
The distance metric to use Euclidean or Manhattan.
The Euclidean distance metric is used by default.
- Project only
-
If checked the input data is mapped only with respect to the specified
fixed data points (see tab 'Fixed data'). If unchecked, the other
(not fixed) data points are taken into account too, when adjusting
the position of each single data point.
- Input data
-
Specifies the columns to use by the mapping.
- Fixed data
-
Specifies the columns to use as fixed data. Be aware that the chosen
columns represent the lower dimensional data which is used to project
the input data at.
The number of columns to choose has to be equal to the value set for
output dimension.
Ports
Input Ports
0 |
Data table containing the fixed data points.
|
1 |
Data table containing the data to map.
|
Output Ports
0 |
The input data and the mapped data.
|
This node is contained in KNIME Base Nodes
provided by KNIME GmbH, Konstanz, Germany.