Feature Analysis Controls
Feature Analysis Controls builds a square matrix relating the numeric columns of a bag to each other. You choose the matrix type, an optional scaling transform, and which features to include; the matrix is computed on the backend and drawn as a heatmap in the companion Feature Analysis View panel.
Where to find it
View → Quantitative → Feature Analysis Controls. Open Feature Analysis View next to it, or apply the Quantitative Analysis or Coding layout preset, which opens both.
Preconditions
- A bag must be selected. The panel reads the most recently selected bag and labels the feature list Selection: <bag name>.
- That bag needs more than 3 points. With none, the panel refuses with “Please select at least 4 points to perform analysis.”
- The space needs more than 2 numeric columns. The backend rejects the request otherwise.
When the panel opens with a bag of more than three points already selected, it runs once automatically. Every recomputation after that is manual.
Controls
Matrix Type
- Correlation: Pearson correlation between each pair of numeric columns.
- Spearman Correlation: rank correlation, which catches monotone but non-linear relationships.
- Covariance: unnormalized, so its scale depends on the units of the columns.
Scaling
Applied to the numeric columns before the matrix is computed. None is the default; the rest are Standardization (Z-score), Min-Max Scaling, Robust Scaling, Max-Abs Scaling, Quantile Transformation, Power Transformation, Log Transformation, Geometric Scaling, Unit Vector Scaling, Box-Cox Transformation, Yeo-Johnson Transformation, L1 Normalization, L2 Normalization, and Regression-based Scaling.
Choosing Regression-based Scaling reveals a Target Col dropdown; pick the column the others are regressed against.
Feature list
Below the dropdowns is the list of numeric columns returned by the last run:
- The toggle on each row includes or excludes that feature from the heatmap.
- Active features can be dragged into any order, and the heatmap follows that order.
- Search… filters the list by substring.
- Regex Group… groups the list by the captured regex match, which is how you keep families of columns such as
pssm_A,pssm_Btogether. - Sort by val takes a column name and orders features by that column’s values.
- [Select All] and [Clear All] apply to whatever the search currently shows, not to the whole list.
- A red (!) badge marks a column that contained NaNs. Those gaps were filled in on the backend before the matrix was computed, and a note under the list says so. Read those rows and columns with more caution than the rest.
Run Analysis
Run Analysis 🚀 recomputes the matrix with the current Matrix Type, Scaling, and Target Col. The button reads “Computing…” while the request is in flight.
Toggling and reordering features redraws the existing heatmap instantly. Changing Matrix Type or Scaling does not: press Run Analysis 🚀 again.
How to use it
- Select a bag of at least four points.
- Choose a Matrix Type. Start with Correlation; switch to Spearman when the relationships look monotone but bent.
- Choose a Scaling if the columns have wildly different units, then press Run Analysis 🚀.
- Trim the feature list. Press [Clear All], then toggle on the dozen columns you care about. The heatmap becomes readable well before the full list does.
- Read the heatmap in Feature Analysis View, and click a cell to send that feature pair to Plot Controls as X and Y.
Limits
- Numeric columns only. Categorical columns take no part in the matrix.
- A failed request replaces the whole panel with an Analysis Error card and a Try Again button.
- Nothing is saved. There are no named feature presets; the active list lives for the session and resets with the panel.
- No descriptive statistics are computed here. For per-feature mean, min, max, and standard deviation use the statistics cards in Plot Controls; for group comparison and outliers use Meaningful Features.
- There is no agent button in this panel.