Oracle® Data Mining Concepts 12c Release 1 (12.1) E1769215 


PDF · Mobi · ePub 
This chapter describes Orthogonal Partitioning Clustering (OCluster), an Oracleproprietary clustering algorithm. See Chapter 7 for information about clustering.
This chapter contains these topics:
Reference:
Campos, M.M., Milenova, B.L., "Clustering Large Databases with Numeric and Nominal Values Using Orthogonal Projections", Oracle Data Mining Technologies, Oracle Corporation.
OCluster is a fast, scalable gridbased clustering algorithm wellsuited for mining large, highdimensional data sets. The algorithm can produce high quality clusters without relying on userdefined parameters.
The objective of OCluster is to identify areas of high density in the data and separate the dense areas into clusters. It uses axisparallel unidimensional (orthogonal) data projections to identify the areas of density. The algorithm looks for splitting points that result in distinct clusters that do not overlap and are balanced in size.
OCluster operates recursively by creating a binary tree hierarchy. The number of leaf clusters is determined automatically. The algorithm can be configured to limit the maximum number of clusters.
Partitioning strategy refers to the process of discovering areas of density in the attribute histograms. The process differs for numerical and categorical data. When both are present in the data, the algorithm performs the searches separately and then compares the results.
In choosing a partition, the algorithm balances two objectives: finding well separated clusters, and creating clusters that are balanced in size. The following paragraphs detail how partitions for numerical and categorical attributes are identified.
To find the best valid cutting plane, OCluster searches the attribute histograms for bins of low density (valleys) between bins of high density (peaks). OCluster attempts to find a pair of peaks with a valley between them where the difference between the peak and valley histogram counts is statistically significant.
A sensitivity level parameter specifies the lowest density that may be considered a peak. Sensitivity is an optional parameter for numeric data. It may be used to filter the splitting point candidates.
Categorical values do not have an intrinsic order associated with them. Therefore it is impossible to apply the notion of histogram peaks and valleys that is used to partition numerical values.
Instead the counts of individual values form a histogram. Bins with large counts are interpreted as regions with high density. The clustering objective is to separate these highdensity areas and effectively decrease the entropy (randomness) of the data.
OCluster identifies the histogram with highest entropy along the individual projections. Entropy is measured as the number of bins above sensitivity level. OCluster places the two largest bins into separate partitions, thereby creating a splitting predicate. The remainder of the bins are assigned randomly to the two resulting partitions.
The OCluster algorithm operates on a data buffer of a limited size. It uses an active sampling mechanism to handle data sets that do not fit into memory.
After processing an initial random sample, OCluster identifies cases that are of no further interest. Such cases belong to frozen partitions where further splitting is highly unlikely. These cases are replaced with examples from ambiguous regions where further information (additional cases) is needed to find good splitting planes and continue partitioning. A partition is considered ambiguous if a valid split can only be found at a lower confidence level.
Cases associated with frozen partitions are marked for deletion from the buffer. They are replaced with cases belonging to ambiguous partitions. The histograms of the ambiguous partitions are updated and splitting points are reevaluated.
The OCluster algorithm evaluates possible splitting points for all projections in a partition, selects the best one, and splits the data into two new partitions. The algorithm proceeds by searching for good cutting planes inside the newly created partitions. Thus OCluster creates a binary tree structure that divides the input space into rectangular regions with no overlaps or gaps.
The main processing stages are:
Load the buffer. Assign all cases from the initial buffer to a single active root partition.
Compute histograms along the orthogonal unidimensional projections for each active partition.
Find the best splitting points for active partitions.
Flag ambiguous and frozen partitions.
When a valid separator exists, split the active partition into two new active partitions and start over at step 2.
Reload the buffer after all recursive partitioning on the current buffer is completed. Continue loading the buffer until either the buffer is filled again, or the end of the data set is reached, or until the number of cases is equal to the data buffer size.
Note:
OCluster requires at most one pass through the dataThe clusters discovered by OCluster are used to generate a Bayesian probability model that can be used to score new data. The generated probability model is a mixture model where the mixture components are represented by a product of independent normal distributions for numerical attributes and multinomial distributions for categorical attributes.
The OCluster algorithm supports two buildtime settings. Both settings have default values. There is no reason to override the defaults unless you want to influence the behavior of the algorithm in some specific way.
You can configure OCluster by specifying any of the following:
Buffer size — Size of the processing buffer. (See Active Sampling.)
Sensitivity factor — A fraction that specifies the peak density required for separating a new cluster. (See Partitioning Strategy.)
See Also:
"Algorithm Settings: OCluster" in Oracle Database PL/SQL Packages and Types ReferenceAutomatic Data Preparation bins numerical attributes for OCluster. It uses a specialized form of equiwidth binning that computes the number of bins per attribute automatically. Numerical columns with all nulls or a single value are removed. OCluster handles missing values naturally as missing at random.
Note:
OCluster does not support nested columns, sparse data, or unstructured text.See Also:
Chapter 3, "Preparing the Data" in Oracle Data Mining User's Guide
Chapter 4, "Transforming the Data" in Oracle Data Mining User's Guide
Keep the following in mind if you choose to prepare the data for OCluster:
OCluster does not necessarily use all the input data when it builds a model. It reads the data in batches (the default batch size is 50000). It will only read another batch if it believes, based on statistical tests, that there may still exist clusters that it has not yet uncovered.
Binary attributes should be declared as categorical.
Automatic equiwidth binning is highly recommended. The bin identifiers are expected to be positive consecutive integers starting at 1. See Oracle Database PL/SQL Packages and Types Reference for an example.
The presence of outliers can significantly impact clustering algorithms. Use a clipping transformation before binning or normalizing. Outliers with equiwidth binning can prevent OCluster from detecting clusters. As a result, the whole population appears to fall within a single cluster.