K-means clustering

Supported in: Batch

K-means clustering is an unsupervised machine learning algorithm. It groups dataset vectors into k clusters. The k value is determined by computing the best silhouette score of the specified range between minimum k and maximum k. Number of k values defines how many k values should be tried within this range, inclusive of the boundaries.

Transform categories: Other

Declared arguments

• Input dataset - Source dataset containing vector column.
  Table
• Maximum k - Maximum number of clusters.
  Literal<Integer>
• Minimum k - Minimum number of clusters.
  Literal<Integer>
• Number of k values - Number of k values to test, over values from minimum k to maximum k. Note that we will train number of k clustering models, so the pipeline execution might be slow if number of k values is set to a high value. The best model is selected based on the silhouette score.
  Literal<Integer>
• Vector column - Column containing the float vectors that will be used for the clustering.
  Column<Array<Float>>

Examples

Example 1: Base case

Argument values:

• Input dataset: ri.foundry.main.dataset.a
• Maximum k: 12
• Minimum k: 3
• Number of k values: 4
• Vector column: feature_column

Input:

Output:

---

Example 2: Null case

Argument values:

• Input dataset: ri.foundry.main.dataset.a
• Maximum k: 12
• Minimum k: 3
• Number of k values: 4
• Vector column: feature_column

Input:

Output:

---

Example 3: Edge case

Argument values:

• Input dataset: ri.foundry.main.dataset.a
• Maximum k: 3
• Minimum k: 3
• Number of k values: 1
• Vector column: feature_column

Input:

Output:

---