Skip to content

Model Prediction (with Copy-Paste)

Once a model is trained, the most valuable thing you can do is use it to predict new data.

The traditional approach is a lot of hassle: organize the data into a file → import → match column names → run the prediction → export again. The StarWay Data Insight platform compresses this whole chain into a single action — Ctrl+V.


🚪 Entry

Top toolbar → Model Prediction (prediction icon)

⚠️ Prerequisite: You must have a model that is activated and fitted (Fit) in the model list on the right. When no usable model exists, this button cannot be clicked.


📋 Interface Layout

When opened, it is a large dialog divided into three sections from top to bottom:

1. Model information bar

Displayed itemDescription
InstanceThe name of the current instance
ModelThe current model name + the model type tag (PCA / PLS / PLS-DA / OPLS / OPLS-DA)
MetricsThe model's core evaluation metrics; hover the mouse to see explanations

What the metrics bar displays depends on the model type:

Model typeDisplayed metrics
PLS / OPLS (regression)R² : … | Q² : …
PLS-DA / OPLS-DA (classification)Acc: … | AUC: … | F1: …
PCAUnsupervised model; no prediction metrics are displayed

💡 Take a look at the metrics before predicting: if Q² is only 0.3, the prediction results themselves are of limited credibility, so don't push your conclusions too far.

2. Data table area

  • The first column is the row number index
  • Then come the X columns (the header shows two rows: the upper row is the variable's English name, the lower row is the Chinese name)
  • Then the original Y columns (if you paste the true values along with the data, they can be used for on-the-spot validation)
  • Finally the prediction columns, with a header such as x3 (Predicted)

3. Prediction comparison chart area

After you click "Predict", each Y variable automatically generates a prediction comparison chart, with evaluation metrics displayed in real time above the chart:

Model typeChart metrics
RegressionRMSE (root mean square error), (coefficient of determination), MAE (mean absolute error)
ClassificationAcc (accuracy), F1 (macro F1 score), Precision (macro precision)

💡 Hover over any metric to see a plain-language explanation. For example, R² = "how well the prediction model fits the real data; the closer to 1, the better".


📥 How to Feed Data In: Copy and Paste

This is the only input method for the prediction panel — there is no need to import a file.

Steps

  1. In Excel, select the data range you want to predict and copy it with Ctrl+C / Cmd+C
  2. Go back to the prediction dialog and click once inside the dialog area (so that the dialog gets focus)
  3. Paste with Ctrl+V / Cmd+V
  4. When you see the "Successfully pasted N rows of data" message, it is done
  5. Click Predict

⚠️ The single most critical rule: the column order must match

The platform reads pasted data by column order, and does not match by column name. Therefore:

Make sure the column order of the pasted data is exactly the same as the order of the X variable list (and the Y variable list).

The figure above gives explicit hints:

  • PCA models: "Please click inside this area, then use Ctrl+V / Cmd+V to paste the X variable data from Excel, making sure the column order matches the X variable list."
  • Other models: "…paste data, making sure the column order matches the X and Y variable lists."

Paste format requirements

  • Simply copy directly from Excel (tab-separated); there is no need to save it as CSV
  • Pasting multiple rows at once is supported, for batch prediction
  • The number of columns should correspond to the number of X (+ Y) variables

💡 Recommended practice: rearrange the new data in Excel into the same column order as the prediction panel's header before copying, which avoids order misalignment.


📤 Output Results: Paste Back to Excel in One Click

After the prediction is complete, click Copy Results:

  • The results are written to the system clipboard, with the message "Results copied to the clipboard; you can paste them directly into Excel"
  • Go back to Excel and press Ctrl+V to paste the predicted values back into your original table

This forms a closed loop: Excel → paste into the platform → predict → paste back into Excel, with no intermediate files produced at any point.

Other actions

ButtonFunction
ClearClears the currently pasted data and prediction results
PredictRuns the prediction on the data in the table

💡 If the pasted data also contains the true Y values, RMSE / R² (or Acc / F1) are calculated and displayed automatically after prediction — effectively an external validation done along the way.


🔄 Reverse Prediction: Working Back from the Target to the Parameters

Besides "predicting Y from X", the platform also provides a reverse calculation capability: given the principal component scores of a target, it back-calculates the corresponding raw variable values.

This capability has no separate button; it is integrated into the Explore Mode of Model Exploration:

  • Drag the red dot on the score plot to the target position
  • The X-Space / Y-Space panel below displays the theoretical values of each variable in real time

💡 This is where the answer to "to reach this quality metric, what should the process parameters be set to" comes from. See Model Exploration and Parameter Optimization for details.


🎯 Typical Use Cases

Use case 1: Quick judgment of incoming materials

Receive the test data of a new batch of raw materials → copy it in X column order → paste it into the prediction panel → predict → copy the results back

Use case 2: Offline data re-validation

Paste in historical data (including the true Y) → predict → look at R² / RMSE → evaluate the model's actual performance on that batch

Use case 3: Multi-target prediction

When a PLS model has multiple Y columns, each Y gets its own comparison chart and its own set of metrics

Use case 4: Reverse optimization

Model Exploration → Explore Mode → drag the red dot to the target quality region → read X-Space → get the recommended process parameters

⚠️ Notes

  • Order-sensitive: a misaligned column order will not raise an error, but it will produce completely wrong prediction results — this is the most common usage accident
  • Do not paste the header row: paste only data rows; column names are not recognized
  • Numeric format: make sure the cells are numbers rather than text (in Excel, left-aligned "numbers" are often text)
  • PCA models can only show structure: an unsupervised model has no Y, so the prediction panel is mainly used to view the projection of samples in the latent space
  • Extrapolation risk: if the value range of the new data clearly exceeds that of the training data, the credibility of the prediction drops — refer to the boundary violations (Violations) hints in Model Exploration

Let data speak, make decisions simpler.