Function Reference

Function Categories

📊 Data Functions

Functions for loading, exploring, and manipulating data.

• ML.DATASETS.* - Built-in datasets (Iris, Diabetes, Digits, OpenML)
• ML.DATA.* - Data manipulation and exploration

📈 Regression Models

Functions for predicting continuous values.

• ML.REGRESSION.LINEAR - Linear Regression
• ML.REGRESSION.RIDGE - Ridge Regression (L2 regularization)
• ML.REGRESSION.LASSO - Lasso Regression (L1 regularization)
• ML.REGRESSION.ELASTIC_NET - Elastic Net (L1 + L2)
• ML.REGRESSION.RANDOM_FOREST_REG - Random Forest Regression

🎯 Classification Models

Functions for categorizing data.

• ML.CLASSIFICATION.LOGISTIC - Logistic Regression
• ML.CLASSIFICATION.SVM - Support Vector Machines
• ML.CLASSIFICATION.RANDOM_FOREST_CLF - Random Forest Classifier

🔍 Clustering Models

Functions for finding groups in data.

• ML.CLUSTERING.KMEANS - K-Means clustering with advanced parameters

⚙️ Preprocessing Functions

Functions for preparing data.

• ML.PREPROCESSING.TRAIN_TEST_SPLIT - Split train/test sets
• ML.PREPROCESSING.STANDARD_SCALER - Standardize features
• ML.PREPROCESSING.MIN_MAX_SCALER - Scale to range [0,1]
• ML.PREPROCESSING.ROBUST_SCALER - Scale robust to outliers
• ML.PREPROCESSING.ONE_HOT_ENCODER - One-hot encode categories
• ML.PREPROCESSING.ORDINAL_ENCODER - Ordinal encode categories

📏 Evaluation Functions

Functions for assessing model performance.

• ML.EVAL.SCORE - R² or accuracy score
• ML.EVAL.CV_SCORE - Cross-validation
• ML.EVAL.GRID_SEARCH - Hyperparameter tuning
• ML.EVAL.BEST_PARAMS - Extract best parameters
• ML.EVAL.BEST_SCORE - Get best CV score
• ML.EVAL.SEARCH_RESULTS - Detailed search results

📐 Evaluation Metrics Reference

Standalone metric functions for regression, classification, and clustering evaluation.

• ML.EVAL.REGRESSION.* — R²/RMSE/MAE and 12 more regression metrics
• ML.EVAL.CLASSIFICATION.* — Accuracy/F1/ROC AUC and more (label-based and score-based)
• ML.EVAL.CLUSTERING.* — Adjusted Rand Score/V-Measure and more

🔧 Core ML Functions

Essential functions for model training and prediction.

• ML.FIT - Train models and transformers
• ML.PREDICT - Make predictions
• ML.TRANSFORM - Transform data
• ML.FIT_TRANSFORM - Fit and transform in one step
• ML.PIPELINE - Create ML workflows
• ML.OBJECT_INFO - Inspect object details

📉 Dimensionality Reduction

Functions for reducing feature dimensions.

• ML.DIM_REDUCTION.PCA - Principal Component Analysis
• ML.DIM_REDUCTION.PCA.RESULTS - PCA detailed results
• ML.DIM_REDUCTION.KERNEL_PCA - Non-linear PCA

🔬 Inspection Tools

Functions for model analysis and visualization.

• ML.INSPECT.GET_PARAMS - Extract model parameters
• ML.INSPECT.DECISION_BOUNDARY - Visualize decision boundaries

🧩 Compose Functions

Functions for advanced column transformations.

• ML.COMPOSE.COLUMN_TRANSFORMER - Apply transformer to columns
• ML.COMPOSE.DATA_TRANSFORMER - Combine transformers
• ML.COMPOSE.COLUMN_SELECTOR - Select columns by pattern/type
• ML.COMPOSE.TRANSFORMERS.DROP - Drop columns
• ML.COMPOSE.TRANSFORMERS.PASSTHROUGH - Pass columns unchanged

🔧 Impute & Feature Selection

Functions for handling missing values and selecting features.

• ML.IMPUTE.SIMPLE_IMPUTER - Impute missing values
• ML.FEATURE_SELECTION.SELECT_PERCENTILE - Select top features

Function Naming Convention

All FormulaML functions follow a consistent naming pattern:

ML.[NAMESPACE].[FUNCTION_NAME](parameters)

Examples:

• ML.DATASETS.IRIS() - Load Iris dataset
• ML.REGRESSION.LINEAR() - Create linear regression model
• ML.EVAL.SCORE() - Evaluate model performance

Free vs Premium Functions

✅ Free Functions

Most core functionality is available in the free tier:

• Basic models (Linear, Logistic, SVM, K-Means)
• Data handling and exploration
• Model training and prediction
• Basic evaluation

Premium Functions

Advanced capabilities require premium subscription:

• Random Forest models
• Cross-validation (ML.EVAL.CV_SCORE)
• Grid search (ML.EVAL.GRID_SEARCH)
• Kernel PCA (ML.DIM_REDUCTION.KERNEL_PCA)
• OpenML datasets (ML.DATASETS.OPENML)

Understanding Object Handles

Many functions return or accept “object handles” - references to complex data structures:

Cell A1: =ML.DATASETS.IRIS()           → Returns: <Dataset>
Cell A2: =ML.REGRESSION.LINEAR()       → Returns: <LinearRegression>
Cell A3: =ML.FIT(A2, features, target) → Returns: <LinearRegression> (with 🧠 brain icon)

These handles allow Excel to manage complex ML objects efficiently.

Common Parameters

Frequently Used Parameters

random_state

• Type: Integer
• Purpose: Ensures reproducible results
• Example: 42 (any integer works)

fit_intercept

• Type: Boolean (TRUE/FALSE)
• Purpose: Whether to calculate the intercept
• Default: TRUE

alpha

• Type: Float
• Purpose: Regularization strength
• Range: > 0 (higher = more regularization)

n_estimators

• Type: Integer
• Purpose: Number of trees in ensemble
• Default: 100

max_iter

• Type: Integer
• Purpose: Maximum iterations
• Default: Varies by algorithm

Return Value Types

Functions return different types of values:

1. Object Handles: Complex objects (models, dataframes)

   • Example: <SVC>

2. Numeric Values: Single numbers

   • Example: 0.95 (accuracy score)

3. Arrays: Multiple values

   • Example: Cross-validation scores

4. DataFrames: Tabular data

   • Example: Sample data, parameters

Quick Function Lookup

By Task

Load Data:

• ML.DATASETS.IRIS() - Classification dataset
• ML.DATASETS.DIABETES() - Regression dataset
• ML.DATASETS.FORCE_2020() - Well log data (petrophysics)
• ML.DATA.CONVERT_TO_DF() - Excel to DataFrame

Explore Data:

• ML.DATA.INFO() - Data structure
• ML.DATA.DESCRIBE() - Statistics
• ML.DATA.SAMPLE() - View rows

Filter / Query Data:

• ML.DATA.QUERY() - Run DuckDB SQL against a DataFrame

Prepare Data:

• ML.DATA.SELECT_COLUMNS() - Choose columns
• ML.PREPROCESSING.TRAIN_TEST_SPLIT() - Split data
• ML.PREPROCESSING.STANDARD_SCALER() - Scale features

Train Models:

• ML.FIT() - Train any model
• ML.PREDICT() - Make predictions
• ML.TRANSFORM() - Transform data

Evaluate:

• ML.EVAL.SCORE() - Basic scoring (model-level)
• ML.EVAL.CV_SCORE() - Cross-validation
• ML.EVAL.GRID_SEARCH() - Hyperparameter tuning
• Evaluation Metrics Reference - Array-level metrics (RMSE, ROC AUC, F1, Adjusted Rand Score, and 30+ more)

Error Messages

Common error messages and their meanings:

“Object handle not found”

• The referenced cell doesn’t contain a valid object
• Solution: Check cell reference is correct

“Invalid parameter value”

• Parameter is outside acceptable range
• Solution: Check parameter constraints

“Dimension mismatch”

• Data shapes don’t match
• Solution: Ensure X and y have same number of rows

“Premium function”

• Function requires premium subscription
• Solution: Upgrade or use free alternative

Best Practices

1. Always use consistent data shapes

   • Features (X) and target (y) must have same number of rows

2. Set random_state for reproducibility

   • Use same seed value across related operations

3. Check data types

   • Ensure numerical data isn’t stored as text

4. Handle missing values

   • Use ML.IMPUTE or clean data before analysis

5. Start with simple models

   • Use as baseline before complex models

Navigation

Browse functions by category:

• Data Functions - Data loading and manipulation
• Regression Models - Continuous prediction
• Classification Models - Category prediction
• Clustering Models - Group discovery
• Preprocessing - Data preparation
• Evaluation - Model assessment
• Evaluation Metrics - Standalone metric functions
• Core Functions - Training and prediction
• Dimensionality Reduction - Feature reduction
• Inspection Tools - Model analysis
• Compose Functions - Advanced pipelines
• Impute & Feature Selection - Data cleaning and feature selection

Or return to:

• Documentation Home
• Getting Started
• Examples