Creates a density model of a n-dimensional vector space, based on a kNN graph. The kNN graph is created by connecting each row with k rows closest (in terms of Euclidean distance) to it in the feature space (k is the specified number of neighbors). This way each row has at least k edges in the kNN graph, however there are two cases in which a row may have more than k edges:

• It is among the k nearest neighbors of a row that is not among its own nearest neighbors.
• There are multiple rows that would be the kth nearest neighbor because they have the same distance to the row in question.

Each edge in the kNN graph is weighted using a Gaussian kernel over the distance of the connected rows with standard deviation Sigma. The density of a specific row is calculated as the mean weight of all its edge weights. For more details see the RALF Paper by Ebert et al.
If the node fails to execute due to memory problems, this is usually because the number of neighbors is set too high.

Options

Column Selection

The columns that make up the vector space.

Number of Neighbors

The number of neighbors which are considered.

Sigma

The Sigma for the Gaussian distance weighting function.

Missing Value Handling

Missing values can't be used to build the model therefore two strategies exist to cope with missing values. The node can either fail if it encounters a missing value in one of the used columns or it can ignore the row in which the missing value occurred. Note that an ignored row is not part of the model and consequently the Density Scorer node will treat it as an unknown row.

Views

This node has no views

Workflows

• 01_Guided_Labeling_for_Document_ClassificationKNIME Hub
• 01_Guided_Labeling_for_Document_Classification-FIXED-HEADERKNIME Forum
• 02_Active_Learning_for_Document_ClassificationKNIME Hub

Links

• No links available

Developers