Results stage
Results is the last stage of an experiment. It lists every analysis of the experiment, newest first, one card per analysis. Above the cards is the results summary, a table that compares the cross-validated accuracy of all classification analyses.
The stage has, from top to bottom:
- the header Analysis Results & Visualizations with the buttons Delete All and Refresh;
- a progress panel, only while something is still running;
- the Analysis Performance Summary (results summary);
- the analysis cards. With no analyses, a No Analyses Yet message asks you to go back to the Data Analysis stage.
Analysis Results & Visualizations
View and manage your analysis results, create visualizations, and export data.
Analysis cards
Each card has a header and three tabs.
Header, left to right: the analysis name, its status, its method, its Id, the date it was created, and the buttons:
| Button | What it does |
|---|---|
| Predict | Opens the prediction dialog. Only on completed analyses of methods that can predict new samples (every classifier, not PCA). It is disabled, with the tooltip Rerun analysis to enable predictions, when the analysis has no saved model — for example, an analysis from before prediction existed. |
| Trash bin (Delete Analysis) | Deletes the analysis after a confirmation. See Deleting. |
| ⋯ (More Options) | Create Visualization (completed analyses only) and Show Details / Hide Details, which shows the full analysis record as raw JSON under the tabs. |
Summary tab — the main numbers. What it shows depends on the method:
| Method | Summary tab |
|---|---|
| PCA | Variance Explained (first six components) and Details |
| LDA, Logistic, SVM, KNN, PLS-DA, SIMCA | the classification block (below) |
| Random Forest, XGBoost, BBC, BRF | the classification block and Top Feature Importance |
LDA Analysis
completedLDAId: 7c1e4b2a-9d3f-4e61-8a52-0b6f1d2c3e4aCross-Validation
| CV Accuracy: | 0.9143±0.0700 |
| CV Precision: | 0.9350±0.0539 |
| CV AUC: | 0.9583±0.0373 |
| CV AUC Agg: | 0.9556 |
| CV Sensitivity: | 0.8667±0.1633 |
| CV Specificity: | 0.9524±0.1000 |
| CV F1-Score: | 0.9092±0.0743 |
| Folds: | 5 |
Class Metrics (CV Average)
| Precision: | 0.9200 |
| Recall: | 0.9500 |
| F1-Score: | 0.9270 |
| Precision: | 0.9500 |
| Recall: | 0.8667 |
| F1-Score: | 0.8914 |
Model self-test on estimation samples
| Accuracy: | 0.9444 |
| Precision: | 0.9348 |
| R2: | N/A |
| Sensitivity: | 0.8667 |
| Specificity: | 1.0000 |
| F1 Score: | 0.9416 |
Cross-Validation Confusion Matrix (Aggregated)
| P:China | P:India | |
|---|---|---|
| A:China | 20 | 1 |
| A:India | 2 | 13 |
The classification block has four parts. All metrics are defined on the Metrics page.
- Cross-Validation — the metrics of the test folds, averaged over the
folds, with the standard deviation over folds after
±: CV Accuracy, CV Precision, CV F1-Score and Folds (the number of folds that ran). For two classes also CV Sensitivity, CV Specificity, CV AUC and CV AUC Agg (the pooled AUC; not computed by XGBoost, BBC and BRF), and for PLS-DA CV R². For more than two classes, CV Recall replaces sensitivity and specificity. Under the table: the per-fold values (Fold Sensitivity, Fold Specificity, Fold Accuracy Scores); hover a badge to see its fold number. - Class Metrics (CV Average) — precision, recall and F1-score of each class, averaged over the folds.
- Model self-test on estimation samples — the final model, fitted on all rows, tested on the same rows. These numbers are always optimistic; judge the model by the cross-validated ones. See Cross-validated vs training metrics. R² is computed only by PLS-DA and shows N/A for the other methods.
- Cross-Validation Confusion Matrix (Aggregated) — the predictions of all
test folds together. Rows (
A:) are the actual classes, columns (P:) the predicted ones; correct predictions on the diagonal are green.
A yellow note above the block, like Note: Accuracy (0.944) differs from (sensitivity + specificity)/2 (0.933) by 0.011, is not an error. It appears when the classes have different sizes. See Consistency note.
In the Cross-Validation table, CV Precision, CV Recall, CV Sensitivity, CV Specificity and CV F1-Score show N/A when the value is exactly zero, and so do Sensitivity, Specificity and R² of the self-test. N/A can therefore mean "0" as well as "not computed". Open the Details tab to see the real value.
For PCA, the Summary tab shows the share of variance of each principal component and the cumulative share, plus Components, Dimensions (components × rows) and 95% Variance (how many components explain 95 % of the variance).
PCA Analysis
completedPCAId: 2b9d6f41-0c8e-4a37-b5e2-6a1f3c7d9e08Variance Explained
Details
Tree-based methods add Top Feature Importance: the six columns with the
highest importance, from feature_importances_ of the final model. See
Feature importance.
Top Feature Importance
Details tab — the complete results of the analysis as JSON: every
metric, per-fold list, confusion matrix and model output (scores, loadings,
predictions). Field names are listed on the Metrics page.
Config tab — how the analysis was run: Type, Data Range, Label
Column, CV Method (group, stratified or kfold) and CV Folds,
followed, when they apply, by:
- Averaging Of Replicates — shown when replicate averaging was on;
- Feature Selection — method, Features Selected (for example
2 of 4), the names of the selected columns and the five best scores. See Feature selection; - Model Parameters — the main parameters of the fitted model (for example Solver: svd for LDA);
- Feature Importance — the five most important features (tree-based methods).
Below the tabs, a card lists its plots under Visualizations (n) and its predictions under Unknown Sample Predictions (n), both newest first.
Statuses
| Status | Card | Meaning |
|---|---|---|
pending | grey border | Created, waiting for a worker. |
queued | no border colour | Part of a batch (ALL or All Classes Sets), waiting for its turn. |
running | yellow border, Analysis in progress... | Being computed. |
completed | green border, tabs with results | Finished. Only completed analyses have tabs, plots and Predict. |
failed | red border, red error message | Stopped with an error. The message is the reason, for example Need at least 2 classes for supervised learning. |
SVM Analysis
runningSVMId: a41f07c2-5e9b-4d18-9c3a-7f2e8b1d6c50KNN Analysis
failedK-Nearest NeighborsId: e8d35b19-2a7c-4f06-b1e4-93c0d5a8f217Tea - SIMCA
pendingSIMCAId: 5f0a2c8e-7b3d-4e91-a6c5-1d9e4f7b2a38The page refreshes itself while any analysis, plot or prediction is
pending, queued or running. The progress panel shows how many
analyses are done (completed or failed) out of all analyses of the
experiment, and updating every 8s.... With many running jobs, the page
checks less often (every 10 or 15 seconds).
- After 15 minutes the automatic refresh stops, with the message Auto-refresh paused after 15 minutes. Click Refresh to check again and restart it.
- If the server answers Too many requests, the page waits 30 seconds before it refreshes again (Rate limit reached. Updates will resume shortly.).
- If any analysis failed, the panel says Some analyses failed. Check individual analysis cards for details.
A failed analysis does not affect the others. Fix the cause (usually the data range, the label column or the classes left after filtering) and run the analysis again from the analysis dialog.
Actions: plots, prediction, download, delete
Plots
Some plots are created automatically when an analysis completes:
| Analysis | Plots created automatically |
|---|---|
| PCA | Score plot, loading plot |
| LDA, PLS-DA | Score plot, loading plot; for two classes also a permutation plot |
| Logistic, KNN, SIMCA | For two classes, a permutation plot |
The automatic permutation plot uses 200 permutations and the metric AUC (R² for PLS-DA). For SVM and Random Forest it is not created automatically because it is slow; create it yourself.
To add a plot, open ⋯ → Create Visualization, choose the plot type and its options, and confirm. The new plot appears on the card with status Pending, then Running, then Completed or Failed. The plot types and their options are described in the visualizations section.
LDA Analysis - Permutation Plot
A plot card shows the plot name, its type, its status and its date. A permutation plot also shows the p-value, the CV p-value (green if ≤ 0.05, red otherwise) and the test settings. See Permutation test. Click the image or View Full Size to open it in a large window.
Prediction
Predict applies the trained model of the analysis to samples of another experiment. Each run appears on the card under Unknown Sample Predictions. See Prediction on new data.
Download
- Plots: Download on a plot card (or in the full-size window) saves the image file.
- Numbers: Export on the Data stage downloads an Excel workbook with the summary table and the metrics of every completed analysis.
- Raw results: the Details tab shows the whole
resultsobject as text you can copy.
Deleting
| What | How | Confirmation |
|---|---|---|
| One analysis | Trash bin in the card header | Yes: Delete "name"? This cannot be undone. |
| All analyses of one configuration | Delete All in a row of the summary | Yes |
| All analyses of the experiment | Delete All in the stage header | Yes: Delete All Analyses shows how many analyses and plots will be removed |
| One plot | Trash bin on the plot card | No — the plot is deleted at once |
| One prediction | Trash bin on the prediction card | Yes |
Deleting an analysis also deletes its plots, its predictions and its saved model, and removes it from the summary. It cannot be undone. The data table is not changed. If you share the experiment with a group, Delete All skips the analyses you are not allowed to delete and says how many were skipped.