Skip to content

Getting started

This page walks you from "I have no formulaML add-in installed" to "I trained my first model and used it to predict new values" — entirely in Excel cells.

1. Install the add-in

Visit formulaml.com and follow the install button. Excel's add-in pane will load the formulaML task pane on the right side of your spreadsheet. After installation, the ML.* namespace becomes available in any cell.

To confirm the add-in is connected, type =ML. in a cell. Excel's formula autocomplete should suggest ML.CLASSIFICATION, ML.REGRESSION, ML.CLUSTERING, and the other top-level namespaces.

2. Load a sample dataset

formulaML ships with built-in datasets you can call directly. Pick any empty cell and enter:

=ML.DATASETS.IRIS()

The cell now displays an object handle — a small icon next to a label like Dataset. The full dataset lives on the formulaML server; the cell holds a pointer to it.

3. Split the data into features and target

Most models expect two ranges: a matrix of features (X) and a vector of labels (y). Use the dataset helpers to extract them. Suppose the dataset handle from step 2 is in A1:

=ML.DATA.GET_X(A1)
=ML.DATA.GET_Y(A1)

Spill the results into adjacent ranges, e.g. B1:E150 for X and F1:F150 for y.

4. Train your first model

Use a logistic regression classifier — fast, interpretable, and a sensible baseline:

=ML.CLASSIFICATION.LOGISTIC()
=ML.FIT(H1, B1:E150, F1:F150)

ML.CLASSIFICATION.LOGISTIC() returns an unfitted model handle in H1. ML.FIT(...) consumes that handle and the data, and returns a fitted model handle.

5. Predict on new data

Pass the fitted-model handle plus a new feature range to ML.PREDICT:

=ML.PREDICT(H2, B151:E160)

The result spills as a vector of predicted labels.

What's next

  • Learn the FIT/PREDICT pattern used by every model in formulaML.
  • Browse the Reference to see every available function and its arguments.
  • Replace the logistic regression with ML.CLASSIFICATION.RANDOM_FOREST_CLF or ML.REGRESSION.RIDGE and re-run the FIT/PREDICT chain — every model follows the same workflow.