Grid Prediction - Prediction Engine

<div class="section-label">02 Grid Prediction</div> <div class="section-title">The primary form of RBP analysis.</div> Evaluates combinations of predictive attributes across relevance and similarity thresholds, then forms a fit-weighted composite prediction for circumstances theta. Combinations that fit this task more reliably pull harder on the forecast. The result also includes variable-level **Impact on Fit** and **Impact on Prediction**. Recommended default path for analysis. Takes the same inputs as the other prediction calls: outcomes y, attributes X, and circumstances theta. ## Signature ```text predict_grid( y, # outcomes, length N X, # attributes, N × K theta, # circumstances, length K options? # GridOptions; omit → defaults ) → results ``` **Call:** `predict_grid` **Options:** `GridOptions` (PredictOptions base + Grid-specific options) **Returns:** [[Results/Overview|Results]] (composite; typically T = 1) ## Arguments | Name | Shape | Description | | --- | --- | --- | | y | N | Observed outcomes | | X | N × K | Training attributes aligned with y | | theta | K | Circumstances for the prediction task | | options | (object) | GridOptions; optional, omit for defaults | N = number of observations (rows), K = number of independent variables (columns). ## Options A grid prediction call is ready with defaults: threshold [0, 0.2, 0.5, 0.8] and censor_type both. Pass a [[Settings/Options#GridOptions|GridOptions]] object when you want to change those, or any of the Grid-specific settings below. GridOptions extends PredictOptions except missing_moments (Grid chooses complete vs pairwise per combination). Retain keys, inherited fields, and the full parameter list live on that page. Allowed value names: [[Settings/Allowed Values|Allowed values]]. When X has holes, **Impact on Fit** and **Impact on Prediction** for incomplete columns are measured against an uninformative version of that column by default, so missingness does not make those attributes look artificially weak. Observed cells are replaced with noise at that column's own scale; missing rows stay missing. Turn that off with adjust_impact_for_missing. See [[Results/Grid Insights|Grid Insights]]. | Name | Type / values | Default | Description | | --- | --- | --- | --- | | max_iter | integer > 0 | 1000 | Cap on combination search iterations | | k | integer > 0 | 1 | Combination size / sampling parameter | | seed | unsigned integer | 42 | RNG seed for combination sampling and the missing-column noise used in Impact | | attribute_combi | matrix Q × K | generated | Fixed combination matrix; omit to sample | | adjust_impact_for_missing | on / off | on | Incomplete-column Impact vs uninformative noise at that column's scale (same missing rows). Off is the old include-versus-exclude baseline | | retain_grid_objects | omit \| comma-separated keys | omit (lean) | Which per-cell objects to keep | | inner_parallel | auto \| off | auto | Parallel over combinations | ## Results A successful Grid call gives you one composite forecast for this task, with fit alongside it, and variable-level **Impact on Fit** and **Impact on Prediction**. Those insights come with the usual result; you do not need to keep per-cell tables to see them. | You get | Look here | | --- | --- | | Forecast and fit | yhat, fit, adjusted_fit, … See [[Results/Core\|Core]] | | Variable impact | **Impact on Fit**, **Impact on Prediction**. See [[Results/Grid Insights\|Grid Insights]] | | Solo distribution / composite ξ | When you retain ysolo_distribution. See [[Results/Solo Distribution\|Solo Distribution]] | | Cell-level tables | When you set retain_grid_objects. See [[Results/Grid Cells\|Grid Cells]] | The full layout of a result object: [[Results/Overview|Results]]. ## Related - [[Functions/Overview|Functions]] - [[Predict]] · [[MaxFit]] - [[Settings/Options#GridOptions|GridOptions]] · [[Results/Overview|Results]]