Calculator Settings
Configure data parsing, density estimation, ordering, scaling, and chart presentation.
Formula Used
Kernel density estimation calculates each smooth ridgeline curve. Every sample contributes weight around nearby horizontal positions. The selected bandwidth controls the contribution width.
f̂(x) = (1 / nh) Σ K((x − xᵢ) / h)
Here, n represents observations within one group. The value h represents the smoothing bandwidth. The function K represents the selected kernel shape.
h = 0.9 × min(s, IQR / 1.34) × n−1/5
Automatic mode applies a robust Silverman bandwidth estimate. Manual mode applies one shared bandwidth across groups. Shared smoothing supports more controlled distribution comparisons.
How to Use This Calculator
Enter one named dataset on each line. Place a colon before its numeric observations. Provide at least two observations for every group.
Choose a kernel and preferred bandwidth method. Adjust scaling, overlap, spacing, markers, and visual labels. Submit the form to generate the result above.
Inspect peaks, tails, centers, and relative spread. Compare displayed statistics with the visual density shapes. Download the chart through the Plotly toolbar.
Understanding Ridgeline Distribution Analysis
Why Ridgeline Plots Matter
Ridgeline plots stack density curves across related machine learning groups. Each ridge reveals distribution shape while preserving shared horizontal scales. This view supports fast comparisons across models, classes, folds, or epochs.
Traditional histograms can hide subtle differences between neighboring distributions. Ridgelines expose shifts, multiple peaks, skewness, and heavy tails clearly. Their layered structure also uses vertical space more efficiently.
Machine learning teams often compare residuals across validation folds. They also inspect confidence scores across classes or demographic groups. Ridgelines make those comparisons compact, visual, and immediately understandable.
Density Estimation Choices
Kernel density estimation creates every smooth curve shown here. Each observation contributes a small weighted shape around its value. Summed contributions form an estimated probability density across the range.
Bandwidth controls how strongly nearby observations blend together. Small bandwidths reveal local detail but may create noisy curves. Large bandwidths produce smoother shapes while hiding narrow structures.
Gaussian kernels create soft curves with unlimited mathematical support. Epanechnikov kernels emphasize nearby values within a fixed boundary. Triangular and uniform kernels provide simpler bounded alternatives.
Building a Useful Plot
Start by entering one named group on each line. Separate the group name and numbers using a colon. You may also choose paired rows for spreadsheet exports.
Select automatic bandwidth when exploring unfamiliar data. Choose manual bandwidth when comparing charts under identical smoothing. Keep the same kernel for fair visual comparisons.
Adjust overlap until each ridge remains distinct and readable. Higher overlap emphasizes shape differences across adjacent groups. Lower overlap improves labels and reduces visual crowding.
Interpreting Model Patterns
A shifted peak suggests changing central behavior between groups. Wider ridges indicate greater variability across recorded observations. Separate peaks may reveal hidden subgroups or mixed processes.
Long tails deserve attention because rare values can matter greatly. Unequal tails may signal changing errors, risk, or calibration. Isolated bumps can also indicate data leakage or collection problems.
Ridgeline plots support exploration, not final statistical proof. Confirm important patterns with tests, intervals, and domain knowledge. Exported charts can strengthen reports and model review discussions.
Scaling and Reliable Comparison
Normalization changes vertical interpretation without changing horizontal positions. Peak scaling gives every group equal maximum visual height. Global scaling preserves relative density differences across all displayed groups.
Summary statistics provide context beside the graphical shapes. Means describe centers, while medians resist extreme observations better. Quartiles and standard deviations explain spread from different perspectives.
Review sample sizes before trusting small ridges or isolated peaks. Compare several bandwidths before reporting any surprising distribution feature. Careful settings produce clearer, fairer, and more reliable comparisons.
Frequently Asked Questions
What does a ridgeline plot show?
It shows several smoothed distributions on one shared horizontal scale. Each ridge represents one named data group. Overlapping shapes reveal shifts, spread, peaks, and tails.
Which bandwidth setting should beginners choose?
Automatic bandwidth offers a practical starting point. It adapts smoothing to every group’s spread and size. Manual values help controlled comparisons across repeated reports.
How does bandwidth change the result?
Smaller bandwidths create detailed and sometimes noisy curves. Larger bandwidths create smoother and broader curves. Compare several settings before interpreting minor peaks.
Which density kernel should I select?
Gaussian kernels work well for most exploratory tasks. Bounded kernels reduce influence outside nearby observations. Keep kernels consistent when comparing separate charts.
What is independent peak scaling?
Each ridge receives the same maximum displayed height. This emphasizes shape instead of absolute density magnitude. Sample sizes still appear within the summary table.
When should global scaling be used?
Global scaling preserves density height differences across groups. It helps compare concentration under shared smoothing conditions. Sparse distributions may appear shorter than concentrated ones.
Can I compare unequal group sizes?
Yes, but review counts before drawing conclusions. Count scaling emphasizes groups containing more observations. Peak scaling focuses more strongly on distribution shape.
What do multiple peaks usually indicate?
Multiple peaks may indicate distinct hidden subgroups. They can also reflect sampling noise or narrow bandwidths. Validate important peaks using domain evidence and larger samples.
Can the chart be saved?
Yes, the interactive toolbar provides image export. Hover tools also reveal values and density estimates. Save settings separately whenever reproducible reporting is strictly required.