Biplot VIP Exclusive
🔀 What is this?
Normally you need to switch back and forth between two charts:
- Score Plot —— see where the samples are (rows)
- Loading Plot —— see which way the variables point (columns)
Biplot overlays the two on the same chart. So you can directly answer one question:
"Which variables pulled this cluster of samples on the left over there?"
🧐 How to read?
A Biplot has two kinds of elements:
| Element | Represents | Reading |
|---|---|---|
| Scatter points | Samples (rows) | Position indicates the sample's score in the latent space |
| Arrows / vectors | Variables (columns) | Direction indicates the variable's loading direction on that component |
Three core rules
A sample lies in the direction of a variable's arrow → it means the sample has a relatively high value on this variable
Two variable arrows point in the same direction and are close together → the two are positively correlated (they carry similar information)
Two variable arrows point in opposite directions → the two are negatively correlated (one rises as the other falls)
💡 Vector length also matters: the longer the arrow, the stronger the variable's explanatory power on these two components; a very short arrow means the variable "has little presence" in this chart.
🛠️ How to use?
Configuration items
| Configuration item | Description |
|---|---|
| X-axis component | Choose the principal component for the X-axis (e.g. t1) |
| Y-axis component | Choose the principal component for the Y-axis (e.g. t2) |
Click Generate Chart to render.
Typical uses
| Scenario | How to read |
|---|---|
| Explain a cluster | Find a group of samples clustered together, see which arrow they face → that is their common feature |
| Distinguish two classes of samples | See which variables each of the two classes gathers toward → these variables are the drivers of the classification |
| Find redundant variables | Variables whose arrows almost overlap → highly collinear, consider keeping only one of them |
| Look for "opposing" indicators | Pairs of variables with opposite arrow directions → a trade-off relationship exists (improving one often sacrifices the other) |
Working with the lasso
Biplot also supports lasso selection:
- Circle a group of samples → use the arrow directions to infer their commonality
- Circle a red group and a blue group → can trigger Comparison Analysis
🎯 Division of labor with other charts
| Chart | What to look at | When to use |
|---|---|---|
| t1-t2 Score Plot | Only sample distribution | Find anomalies, view clusters; the chart is cleaner |
| Loading Plot | Only variable directions | Analyze the structural relationships between variables |
| Biplot | Samples and variables together | When you need to explain "why this group of samples clusters together" |
💡 Biplot has a higher reading threshold than the previous two——when there are many points or variables the chart becomes very crowded. It is recommended to use them together: first use t1-t2 to find anomalies, then use Biplot to explain the structure.
⚠️ Notes
- Only supervised models are supported (PLS / PLS-DA / OPLS / OPLS-DA)——PCA has no Y, so Biplot is not provided
- Vectors are only meaningful on components with "relatively high explanatory power": when you choose a component with too low an explained rate, the arrow directions are not stable and the conclusions are unreliable
- Whether the data is standardized significantly affects vector length; judge in combination with the actual business context when interpreting