Model Parameter Reference

This guide provides comprehensive parameter references for all supported models in GlassAlpha.

!!! info "Quick Links" - Choosing a model? → Start with Model Selection Guide - Need configuration help? → See Configuration Guide - Getting started? → Try the Quick Start Guide first

Understanding Parameter Validation

GlassAlpha passes model parameters directly to the underlying ML libraries (scikit-learn, XGBoost, LightGBM). These libraries have different behaviors:

• scikit-learn: Ignores unknown parameters (no error, may log warning)
• XGBoost: Warns about unknown parameters in logs
• LightGBM: Warns about unknown parameters in logs

Check Your Logs

If you see warnings in your audit output about unknown parameters, double-check your parameter names against this reference.

---

LogisticRegression

Logistic Regression is the baseline model for binary and multiclass classification. It's interpretable, fast, and works well for linearly separable data.

Common Parameters

Parameter	Type	Default	Valid Range	Description
random_state	int	None	0-2³¹	Random seed for reproducibility
max_iter	int	100	1-10000	Maximum iterations for solver
C	float	1.0	0.001-100.0	Inverse regularization strength (smaller = more regularization)
penalty	str	'l2'	'l1', 'l2', 'elasticnet', 'none'	Regularization type
solver	str	'lbfgs'	'lbfgs', 'liblinear', 'saga', 'newton-cg'	Optimization algorithm
tol	float	0.0001	1e-6 - 0.01	Tolerance for stopping criteria
fit_intercept	bool	True	True, False	Whether to calculate intercept
class_weight	str/dict	None	'balanced', dict	Weights for imbalanced classes

Usage Example

model:
  type: logistic_regression
  params:
    random_state: 42
    max_iter: 1000
    C: 1.0
    penalty: "l2"
    solver: "lbfgs"

Common Mistakes

❌ Wrong:

model:
  type: logistic_regression
  params:
    max_iter: -1 # ❌ Negative value causes error
    regul: 1.0 # ❌ Typo - should be 'C'
    iterations: 1000 # ❌ Wrong name - should be 'max_iter'
    C: 0 # ❌ Zero is invalid (must be > 0)

✅ Correct:

model:
  type: logistic_regression
  params:
    random_state: 42
    max_iter: 1000
    C: 1.0
    penalty: "l2"

Solver Compatibility

Solver	l1	l2	elasticnet	none	Multiclass
lbfgs	❌	✅	❌	✅	✅
liblinear	✅	✅	❌	❌	❌ (binary only)
saga	✅	✅	✅	✅	✅
newton-cg	❌	✅	❌	✅	✅

Full Reference

See scikit-learn LogisticRegression

---

XGBoost

XGBoost is a powerful gradient boosting library that works well for structured/tabular data.

Common Parameters

Parameter	Type	Default	Valid Range	Description
random_state	int	0	0-2³¹	Random seed
n_estimators	int	100	10-1000	Number of boosting trees
max_depth	int	6	2-15	Maximum tree depth
learning_rate	float	0.3	0.01-0.3	Step size shrinkage (eta)
subsample	float	1.0	0.5-1.0	Fraction of samples for each tree
colsample_bytree	float	1.0	0.5-1.0	Fraction of features for each tree
min_child_weight	int	1	1-10	Minimum sum of instance weight in child
gamma	float	0	0-10	Minimum loss reduction for split
reg_alpha	float	0	0-10	L1 regularization
reg_lambda	float	1	0-10	L2 regularization

Usage Example

model:
  type: xgboost
  params:
    random_state: 42
    n_estimators: 100
    max_depth: 6
    learning_rate: 0.1
    subsample: 0.8
    colsample_bytree: 0.8

Performance Tuning

Fast training (less accuracy):

params:
  n_estimators: 50
  max_depth: 3
  learning_rate: 0.3

Better accuracy (slower):

params:
  n_estimators: 500
  max_depth: 8
  learning_rate: 0.01
  subsample: 0.8
  colsample_bytree: 0.8

Prevent overfitting:

params:
  max_depth: 3 # Shallower trees
  min_child_weight: 5 # Require more samples per leaf
  gamma: 1 # Higher split threshold
  subsample: 0.7 # Use less data per tree
  reg_alpha: 1 # L1 regularization
  reg_lambda: 1 # L2 regularization

Common Mistakes

❌ Wrong:

model:
  type: xgboost
  params:
    n_trees: 100 # ❌ Wrong name - should be 'n_estimators'
    depth: 6 # ❌ Wrong name - should be 'max_depth'
    learning_rate: 5.0 # ❌ Too high (typically 0.01-0.3)
    max_depth: -1 # ❌ Negative depth invalid for XGBoost

✅ Correct:

model:
  type: xgboost
  params:
    n_estimators: 100
    max_depth: 6
    learning_rate: 0.1

Full Reference

See XGBoost Parameters

---

LightGBM

LightGBM is a fast gradient boosting framework that uses tree-based learning algorithms.

Common Parameters

Parameter	Type	Default	Valid Range	Description
random_state	int	None	0-2³¹	Random seed
n_estimators	int	100	10-1000	Number of boosting trees
max_depth	int	-1	-1, 2-15	Maximum tree depth (-1 = no limit)
learning_rate	float	0.1	0.01-0.3	Boosting learning rate
num_leaves	int	31	2-131072	Maximum leaves in one tree
subsample	float	1.0	0.5-1.0	Fraction of data for training
colsample_bytree	float	1.0	0.5-1.0	Fraction of features for each tree
min_child_samples	int	20	1-100	Minimum samples in one leaf
reg_alpha	float	0.0	0-10	L1 regularization
reg_lambda	float	0.0	0-10	L2 regularization

Usage Example

model:
  type: lightgbm
  params:
    random_state: 42
    n_estimators: 100
    num_leaves: 31
    learning_rate: 0.1
    subsample: 0.8
    colsample_bytree: 0.8

Performance Tuning

Fast training:

params:
  n_estimators: 50
  num_leaves: 15
  learning_rate: 0.1

Better accuracy:

params:
  n_estimators: 500
  num_leaves: 63
  learning_rate: 0.01
  subsample: 0.8
  min_child_samples: 10

Prevent overfitting:

params:
  num_leaves: 15 # Fewer leaves
  min_child_samples: 50 # More samples per leaf
  subsample: 0.7 # Use less data
  reg_alpha: 1 # L1 regularization
  reg_lambda: 1 # L2 regularization

Important Notes

max_depth = -1

In LightGBM, max_depth: -1 means no limit on tree depth. This is different from XGBoost where negative values are invalid.

num_leaves vs max_depth

LightGBM grows trees leaf-wise (best-first) rather than level-wise (depth-first). You typically control tree complexity via num_leaves rather than max_depth.

Common Mistakes

❌ Wrong:

model:
  type: lightgbm
  params:
    n_trees: 100 # ❌ Wrong name - should be 'n_estimators'
    max_leaves: 31 # ❌ Wrong name - should be 'num_leaves'
    learning_rate: -0.1 # ❌ Negative learning rate

✅ Correct:

model:
  type: lightgbm
  params:
    n_estimators: 100
    num_leaves: 31
    learning_rate: 0.1

Full Reference

See LightGBM Parameters

---

Parameter Validation Tips

Check Your Config

Use glassalpha validate to check your configuration:

glassalpha validate --config my_config.yaml

Enable Verbose Logging

See parameter warnings in your logs:

logging:
  level: INFO # Or DEBUG for more detail

Common Parameter Mistakes

1. Typos in parameter names - Double-check spelling
2. Using wrong library's parameter names - Each library has different conventions
3. Invalid value ranges - Check the valid range column in tables above
4. Negative values where not allowed - Most counts/sizes must be positive
5. Wrong types - Use int for counts, float for rates

Getting Help

If you're unsure about a parameter:

1. Check this reference guide
2. Run glassalpha validate --config your_config.yaml
3. Check official library documentation
4. See our FAQ for common issues

---

See Also

• Model Selection Guide - Choosing the right model
• Configuration Guide - Full config reference
• Troubleshooting - Common issues and solutions