Predictions

The Prediction workspace evaluates up to three Pareto formulas on ordered custom input points. It is intended for comparison, scenario exploration, and export, not for changing saved search results.

Select formulas

Choose one to three formulas from the Pareto front. Prediction keeps its own ordered selection, independent of the active row in Search and the selection in Solution Diagnostics.

Each selected formula can be edited temporarily. Equarith validates the text in the background and continues to use the last valid expression while a draft is incomplete or invalid. Reset restores the published formula.

Temporary edits:

• do not modify the Pareto solution;
• are not written back into the project;
• do not change the original objective values; and
• should be exported separately if you intend to keep them.

Create custom points

Each point supplies every required input as either:

• a fixed finite value; or
• a range with start, end, and non-zero step.

A point containing ranges expands to their Cartesian product. For example, three values for x and four for y produce twelve prediction rows. The full workspace is limited to 100,000 expanded rows so an accidental tiny step cannot exhaust memory.

Use the supplied actions to fill a point with dataset minimum, mean, or maximum values. These are convenient reference scenarios, but the mean of each column taken independently may describe a combination that never occurred in the dataset.

The blank draft row at the end lets you add another ordered point. Predictions recalculate automatically after valid changes.

Read the output

The result table and chart compare selected formulas. Choose an input for the chart x-axis. Each computed value carries a status so a missing input, invalid function domain, or non-finite result is not confused with an ordinary number.

You can copy selected cells and export prediction rows as CSV. This export contains the current custom grid after range expansion, with each formula’s text in its headers and a per-result status column. By contrast, exporting predictions from a selected Search solution evaluates that formula on every row of the active dataset and includes the inputs, target, and prediction without a status column; a non-finite prediction is an empty cell. Review the headers and input ordering before passing either file to another system, and inspect the custom export’s formula and status columns when using it.

Reference function

The chart can display a temporary reference f(x). In this expression, x means the input selected as the chart axis, and implicit multiplication is accepted by the reference parser.

The reference:

• appears only on the chart;
• is not included in the result table or prediction CSV;
• is included when exporting the chart image or plotted chart data; and
• is not saved in the project.

Use it for a theoretical curve, threshold, or baseline, not as a hidden fourth prediction model.

History-function limitation

Custom points have no unambiguous preceding-row chronology. Formulas containing delay, moving average, decay, difference, EMA, rolling statistics, or rolling slope therefore cannot be evaluated in the custom-point workspace.

Evaluate such formulas against an ordered dataset where Equarith has original row positions and the required warm-up history.

Extrapolation

Symbolic formulas can produce plausible-looking but extreme values outside their training domain. Before using a scenario:

• compare every input with its training range and units;
• inspect strict logarithm, root, division, power, and gamma domains;
• look for discontinuities from conditions, rounding, min, max, or clamp;
• remember that safe functions remain finite by changing behavior near a boundary; and
• test small perturbations around the intended point.

Prediction is numerical evaluation, not an uncertainty guarantee. Equarith v1.0.3 does not attach causal meaning or calibrated prediction intervals to a formula.

For deployment code, read Exporting data and formulas.