Feature-Based Plots¶

Beyond evaluating model predictions, it’s often crucial to understand the features themselves. The commands on this page provide powerful, feature-centric visualizations. They help you explore how features interact to influence a target and compare feature importance profiles across different models or datasets [1].

List of Commands¶
Command	Description
plot-feature-fingerprint	Creates a radar chart comparing feature importance profiles.
plot-feature-interaction	Creates a polar heatmap to visualize feature interactions.

Common Conventions¶

The tools on this page read a tabular data file (e.g., data.csv) specified as a positional argument or via -i/--input. To save a plot, simply add the --savefig out.png flag. For detailed help on any command, run it with the -h flag.

plot-feature-interaction¶

How do two features jointly affect an outcome? This command helps you answer that by creating a polar heatmap. One feature is mapped to the angle, a second is mapped to the radius, and the color of each cell shows the aggregated value of a third, target column. It’s especially powerful for visualizing interactions with cyclical features like the hour of the day or month of the year [2].

The general usage for this command is:

k-diagram plot-feature-interaction INPUT
  --theta-col CYCLIC_FEATURE
  --r-col OTHER_FEATURE
  --color-col TARGET_VARIABLE
  [--statistic mean]
  [--theta-period 24]
  [--theta-bins 24] [--r-bins 10]

For example, to see how solar panel output is affected by the interaction of the hour of the day and the amount of cloud cover:

k-diagram plot-feature-interaction data/solar.csv \
  --theta-col hour \
  --r-col cloud_cover \
  --color-col panel_output \
  --theta-period 24 \
  --theta-bins 24 \
  --r-bins 8 \
  --statistic mean \
  --cmap inferno \
  --title "Solar Output by Hour and Cloud Cover" \
  --savefig solar_interaction.png

plot-feature-fingerprint¶

This command creates a radar (or “spider”) chart to visualize and compare the feature importance profiles of different models or groups. Each polygon on the chart represents a “layer” (like a model), and each axis represents a feature. This “fingerprint” makes it easy to see which features are most important for each model and how these profiles differ [3][4].

The command is highly flexible, allowing you to shape the input data in different ways:

k-diagram plot-feature-fingerprint INPUT
  --cols f1,f2,f3,...
  [--labels L1 L2 ... | --labels-col NAME_COL]
  [--features F1 F2 ...]
  [--transpose]
  [--normalize / --no-normalize]
  [--fill / --no-fill]

Example 1: Standard Orientation By default, each row in your data is treated as a layer (a model). Here, we get the layer names from the “layer” column.

k-diagram plot-feature-fingerprint data/importances.csv \
  --cols feature_1,feature_2,feature_3,feature_4,feature_5 \
  --labels-col layer_name \
  --title "Model Importance Fingerprints" \
  --cmap tab10 \
  --savefig fingerprint_layers.png

Example 2: Explicit Labels You can also provide labels for both the layers (models) and the features (axes) directly on the command line.

k-diagram plot-feature-fingerprint data/importances.csv \
  --cols f1,f2,f3,f4,f5,f6 \
  --labels "Model A" "Model B" "Model C" \
  --features "Temp" "Wind" "Pressure" "Humidity" "Solar" "Time" \
  --normalize \
  --fill \
  --savefig fingerprint_with_labels.png

Example 3: Transposed Data If your data is arranged with features in rows and models in columns, just add the --transpose flag.

k-diagram plot-feature-fingerprint data/transposed_importances.csv \
  --cols Model_A,Model_B,Model_C \
  --labels-col feature_name \
  --transpose \
  --cmap Set3 \
  --title "Transposed Fingerprint" \
  --savefig fingerprint_transposed.png

Troubleshooting & Tips¶

Readability: For fingerprints with many features, the axis labels can get crowded. Consider using shorter feature names or generating a larger figure with --figsize.
Color Choice: When preparing figures for publication, use a colorblind-friendly palette like --cmap tab10 or --cmap viridis.
Need more help? Run any command with the -h or --help flag to see its full list of options and their descriptions.
See Also: After examining feature importance with a fingerprint, you might use plot-feature-interaction to dive deeper into how the top two features interact.

References