Beta Testing Pricing Documentation About Contact
Join Beta Testing

Inspection Tools Reference

Complete reference for FormulaML inspection tools including parameter extraction and decision boundary visualization.

Table of Contents

Inspection Tools Reference

Functions for inspecting, analyzing, and visualizing machine learning models.

ML.INSPECT Namespace

ML.INSPECT.GET_PARAMS()

Extracts parameters from a model or transformer.

Syntax:

=ML.INSPECT.GET_PARAMS(obj, transpose)

Parameters:

  • obj (Object, Required): Model, transformer, or pipeline object
  • transpose (Boolean, Optional): Return wide format (default: FALSE)
    • FALSE: Returns long format (Parameter | Value columns)
    • TRUE: Returns wide format (each parameter as a column)

Returns: DataFrame with model parameters

Use Case: Inspect model configuration, verify hyperparameters

Example:

# Get parameters from model
Cell A1: =ML.INSPECT.GET_PARAMS(trained_model)
Result:
# Parameter | Value
# C         | 1.0
# kernel    | rbf
# gamma     | scale

# Get parameters in wide format
Cell B1: =ML.INSPECT.GET_PARAMS(trained_model, TRUE)
Result:
# C   | kernel | gamma
# 1.0 | rbf    | scale

ML.INSPECT.DECISION_BOUNDARY()

Extracts decision boundary contour points for visualization.

Syntax:

=ML.INSPECT.DECISION_BOUNDARY(model, X, response_method, mesh_step, class_pair, feature_indices, other_features_values, margin)

Parameters:

  • model (Object, Required): Fitted classifier
  • X (Object, Required): Training data (for boundary range)
  • response_method (String, Optional): Method to use (default: “predict”)
    • “predict”: Use predictions
    • “decision_function”: Use decision function values
  • mesh_step (Number, Optional): Grid resolution (default: 0.05)
    • Smaller = smoother boundary, more points
  • class_pair (Array, Optional): Class pair for multiclass (default: [0,1])
    • For multiclass with decision_function
  • feature_indices (Tuple, Optional): Which features to use (default: (0,1))
    • Tuple of two feature indices for boundary
  • other_features_values (Dict, Optional): Values for other features
    • Dict mapping feature index to value
    • Default: Uses mean values
  • margin (Number, Optional): Boundary margin (default: 0.1)

Returns: DataFrame with boundary coordinates

Use Case: Visualize decision boundaries, understand model behavior

Example:

# Basic decision boundary (first two features)
Cell A1: =ML.INSPECT.DECISION_BOUNDARY(trained_svm, X_train)
Result:
# feature_0 | feature_1
# 2.34      | 3.45
# 2.35      | 3.46
# ...

# Boundary for specific features
Cell B1: =ML.INSPECT.DECISION_BOUNDARY(trained_model, X_train, "predict", 0.02, {0,1}, {2,3})

# Use with DataFrame (preserves column names)
Cell C1: =ML.INSPECT.DECISION_BOUNDARY(trained_model, df_train, "predict", 0.05, {0,1}, {0,1})
Result:
# sepal_length | sepal_width
# 5.1          | 3.5
# 5.2          | 3.6
# ...

Common Patterns

Inspect Model Configuration

# Train model
Cell A1: =ML.CLASSIFICATION.SVM(10, "rbf", 3, "scale", 0.0)
Cell B1: =ML.FIT(A1, X_train, y_train)

# Get all parameters
Cell C1: =ML.INSPECT.GET_PARAMS(B1)

# Verify configuration
# Check that C=10, kernel=rbf, etc.

Inspect Pipeline Components

# Create pipeline
Cell A1: =ML.PREPROCESSING.STANDARD_SCALER()
Cell A2: =ML.CLASSIFICATION.SVM()
Cell B1: =ML.PIPELINE(A1, A2)

# Fit pipeline
Cell C1: =ML.FIT(B1, X_train, y_train)

# Get pipeline parameters (shows all steps)
Cell D1: =ML.INSPECT.GET_PARAMS(C1)
Result:
# Model               | Parameter | Value
# standardscaler      | with_mean | True
# standardscaler      | with_std  | True
# svc                 | C         | 1.0
# svc                 | kernel    | rbf

Visualize 2D Decision Boundary

# Train binary classifier
Cell A1: =ML.DATASETS.IRIS()
Cell B1: =ML.DATA.SELECT_COLUMNS(A1, {0,1})  # First 2 features
Cell C1: =ML.DATA.SELECT_COLUMNS(A1, 4)      # Target

Cell D1: =ML.CLASSIFICATION.SVM(1.0, "rbf")
Cell E1: =ML.FIT(D1, B1, C1)

# Extract decision boundary
Cell F1: =ML.INSPECT.DECISION_BOUNDARY(E1, B1, "predict", 0.05, {0,1}, {0,1})

# Create visualization:
# 1. Create scatter plot of B1 data points (color by C1 classes)
# 2. Add line chart of F1 boundary points
# Result: Boundary overlaid on data

Multi-Class Decision Boundaries

# Train multi-class classifier
Cell A1: =ML.DATASETS.IRIS()  # 3 classes
Cell B1: =ML.DATA.SELECT_COLUMNS(A1, {0,1})
Cell C1: =ML.DATA.SELECT_COLUMNS(A1, 4)

Cell D1: =ML.CLASSIFICATION.SVM(1.0, "rbf")
Cell E1: =ML.FIT(D1, B1, C1)

# Get boundaries for each class pair
Cell F1: =ML.INSPECT.DECISION_BOUNDARY(E1, B1, "decision_function", 0.05, {0,1}, {0,1})
Cell F2: =ML.INSPECT.DECISION_BOUNDARY(E1, B1, "decision_function", 0.05, {0,2}, {0,1})
Cell F3: =ML.INSPECT.DECISION_BOUNDARY(E1, B1, "decision_function", 0.05, {1,2}, {0,1})

# Plot all three boundaries

High-Dimensional Boundary (2D Slice)

# Train on 4D data
Cell A1: =ML.DATASETS.IRIS()
Cell B1: =ML.DATA.SELECT_COLUMNS(A1, {0,1,2,3})  # 4 features
Cell C1: =ML.DATA.SELECT_COLUMNS(A1, 4)

Cell D1: =ML.CLASSIFICATION.SVM()
Cell E1: =ML.FIT(D1, B1, C1)

# Visualize boundary for features 0 and 1
# Fix features 2 and 3 at specific values
Cell F1: =ML.INSPECT.DECISION_BOUNDARY(E1, B1, "predict", 0.05, {0,1}, {0,1}, {2:5.5, 3:2.0})

# This shows boundary in 2D slice where feature2=5.5, feature3=2.0

Compare Parameter Settings

# Create multiple models with different params
Cell A1: =ML.CLASSIFICATION.SVM(0.1, "rbf")
Cell A2: =ML.CLASSIFICATION.SVM(1.0, "rbf")
Cell A3: =ML.CLASSIFICATION.SVM(10.0, "rbf")

# Train all
Cell B1: =ML.FIT(A1, X_train, y_train)
Cell B2: =ML.FIT(A2, X_train, y_train)
Cell B3: =ML.FIT(A3, X_train, y_train)

# Get params in wide format for comparison
Cell C1: =ML.INSPECT.GET_PARAMS(B1, TRUE)
Cell C2: =ML.INSPECT.GET_PARAMS(B2, TRUE)
Cell C3: =ML.INSPECT.GET_PARAMS(B3, TRUE)

# Easy to compare C values: 0.1, 1.0, 10.0

Inspect Grid Search Results

# Run grid search
Cell A1: =ML.CLASSIFICATION.SVM()
# ... parameter grid setup ...
Cell B1: =ML.EVAL.GRID_SEARCH(A1, param_grid, "accuracy", 5, TRUE)
Cell C1: =ML.FIT(B1, X_train, y_train)

# Get best model parameters
Cell D1: =ML.EVAL.BEST_PARAMS(C1)

# Get parameters of best model (includes fitted values)
Cell E1: =ML.INSPECT.GET_PARAMS(C1)

Boundary Smoothness Comparison

# Train model
Cell A1: =ML.CLASSIFICATION.SVM(1.0, "rbf")
Cell B1: =ML.FIT(A1, X_train, y_train)

# Coarse boundary (fast)
Cell C1: =ML.INSPECT.DECISION_BOUNDARY(B1, X_train, "predict", 0.1)

# Fine boundary (smooth, slower)
Cell C2: =ML.INSPECT.DECISION_BOUNDARY(B1, X_train, "predict", 0.02)

# Compare point count
Cell D1: =ROWS(C1#)  # Fewer points
Cell D2: =ROWS(C2#)  # More points

Tips and Best Practices

  1. GET_PARAMS Usage

    • Check model configuration before training
    • Verify parameters after grid search
    • Document model settings
    • Debug unexpected behavior

  2. Pipeline Inspection

    • Shows all steps with step__ prefix
    • Verify preprocessing parameters
    • Check model configuration in pipeline
    • Useful for debugging complex workflows

  3. Decision Boundary Visualization

    • 2D only: Can only visualize in 2 dimensions
    • Mesh step: 0.05 good default, 0.02 for smooth
    • Response method:
      • “predict”: Clearer boundaries
      • “decision_function”: Shows confidence regions

  4. Handling High Dimensions

    • Select two most important features
    • Use PCA to reduce to 2D first
    • Fix other features at meaningful values
    • Consider multiple 2D slices

  5. Visualization Tips

    1. Get boundary points with DECISION_BOUNDARY
2. Create scatter plot of training data
3. Color points by class
4. Add boundary as line/contour
5. Adjust mesh_step for smoothness

  6. Performance Considerations

    • Smaller mesh_step = more points, slower
    • Larger data range = more computation
    • Consider sampling large datasets
    • Cache boundary calculations

  7. Common Use Cases

    Model Understanding: GET_PARAMS → verify config
Debugging: GET_PARAMS → check all parameters
Visualization: DECISION_BOUNDARY → plot boundaries
Comparison: Multiple boundaries → understand differences

  8. Boundary Interpretation

    • Linear SVM: Straight boundary
    • RBF SVM: Curved boundary
    • Polynomial: Polynomial curve
    • Complex models: More intricate boundaries
Back to Documentation