Variable Relationship Charts

This shows examples of CorrPlot, Graph, TSNEPlot, and LocalGaussianCorrelationPlot. CorrPlot displays correlation matrices with hierarchical clustering dendrograms. Graph visualizes correlations as a network. TSNEPlot performs t-SNE dimensionality reduction in the browser. LocalGaussianCorrelationPlot shows how correlation varies across the joint distribution of two variables.

Stock Market Correlation Analysis - Multiple Scenarios

A Correlation Plot with Dendrogram shows relationships between variables using hierarchical clustering. The dendrogram (top) groups similar variables based on their correlation patterns, with clustering performed dynamically in your browser. Note that it will only appear if i) you select order by dendrogram and ii) you select all of the variables. The correlation matrix (bottom) uses two different correlation measures: Pearson correlations (top-right triangle, marked with 'P:') measure linear relationships, while Spearman correlations (bottom-left triangle, marked with 'S:') measure monotonic relationships and are robust to outliers. You can change the clustering linkage method (Ward, Average, Single, Complete) to see different groupings. Variables are automatically reordered by the clustering to reveal correlation blocks. See here for CorrPlot examples



Data: stock_corr_data.parquet




Graph

This Graph allows you to visualise the relationships between variables. The dataformat to make this is a DataFrame with columns: node1, node2, strength, scenario, correlation_method, sector. See here for more Graph examples


(Apply variable selection & reorganize layout)
(Changing layout recalculates)
Tip: Deselected variables become translucent. Click "Recalculate Graph" to remove them and reorganize. Switching scenarios updates edges but keeps node positions.

Data: stock_corr_data.parquet




Stock Similarity (t-SNE)

TSNEPlot performs t-SNE dimensionality reduction entirely in the browser. This example shows 10 stocks positioned by similarity across 6 statistical features. Key Features: Interactive feature selection (Available/Selected lists let you choose which variables define similarity), Rescaling options (None, Z-score, Z-score capped, Quantile), Configurable early exaggeration (iterations and factor), Three step modes: 'Step (small)' for single non-exaggerated steps, 'Exaggerated Step' for large movements, 'Run to Convergence' for automatic optimization. You can drag nodes to manually reposition them and resume iteration. Color by sector to see if similar stocks cluster together. See here for TSNEPlot examples

Feature Selection for Distance Calculation:
(Changes which variables determine similarity)
(5-100, typically 15-50)
(10-1000, typically 100-500)
Iteration: 0
Movement (last iteration): 0.000
Status: Ready
Tip: Drag nodes to manually reposition. "Run to Convergence" uses early exaggeration for the configured number of iterations; "Step (small)" never uses exaggeration; "Exaggerated Step" always uses it.

Data: tsne_stock_data.parquet




Local Gaussian Correlation Between Stock Returns

LocalGaussianCorrelationPlot visualizes how correlation between two variables changes across their joint distribution. See here for LocalGaussianCorrelationPlot examples

Filters

Plot Attributes

Display Mode

Bootstrap: Not computed

Axes

Bandwidth

auto (0 = auto)
About Local Gaussian Correlation

Local Gaussian Correlation shows how the correlation between two variables varies across their joint distribution. The heatmap displays the local correlation at each (x, y) point, computed using a Gaussian kernel to weight nearby observations.

The green line on the right shows the average correlation at each Y value (integrated over X), weighted by the kernel density. The red line at the bottom shows the average correlation at each X value (integrated over Y).

Blue = positive, Red = negative, White = zero.

Bootstrap t-statistic: Resamples the data 200 times to estimate the standard error of the local correlation. The t-statistic (correlation / SE) indicates statistical significance. Values beyond ±1.96 are significant at p < 0.05.

1.0 (0.25 - 2.5)

Data: lgc_stock_data.parquet


JSPlots.jl This page was created using JSPlots.jl v0.6.3.