Run cluster inference on new data using a pre-trained Gaussian Mixture Model (GMM). For each input sample, the component produces two outputs: a cluster assignment (the most probable cluster index) and a probability distribution (the likelihood of belonging to each cluster).
Limits
Supported computing engines: MaxCompute, Flink, or DLC.
Configure the component in Designer
Input columns
| Parameter | Description |
|---|---|
| Vector column name | Column containing the input feature vectors. |
| Reserved columns | Columns carried through to the prediction output unchanged. |
Output columns
| Parameter | Description |
|---|---|
| Prediction result column name | Cluster index assigned to each sample (hard assignment). |
| Prediction detail column name | Per-cluster probabilities for each sample. The vector length equals the number of clusters. |
Execution settings (Execution tuning tab)
| Parameter | Description |
|---|---|
| Number of threads for the component | Number of threads used for prediction. Default: 1. |
| Number of workers | Number of parallel workers. Valid range: 1–9999. Set together with Memory per worker (MB). See Estimate resource usage. |
| Memory per worker (MB) | Memory allocated per worker. Valid range: 1024–65536 MB. See Estimate resource usage. |
Estimate resource usage
Use the following guidelines to size your nodes before running the component.
Memory per node
Allocate approximately 30 times the model size in memory per node.
Example: For a 1 GB model, allocate 30 GB per node.
Number of nodes
Adding nodes improves performance initially, but communication overhead eventually degrades speed. Stop adding nodes when performance decreases.
该文章对您有帮助吗？