Utility Functions

Based on: Utilities for developer (opens in a new tab)

Assert

• validation.assert_all_finite: Throw an error if array contains NaNs or Infs.
• _testing.assert_allclose: quasi equality of arrays, using a tol parameter

Formater / converter

• validation.as_float_array: convert input to array of float
• validation.check_array:
  • check that input is a 2D array. Sparse matrix and other dimensions can be optionally allowed
  • call assert_all_finite
• validation.check_X_y:
  • check that X and y have consistent lengths
  • call check_array on X, column_or_1d on y (use multi_output=True for multilabel y)
• validation.indexable: check that input arrays have consistent length and can be sliced or indexed (useful for cross-validation)
• do not use np.asanyarray or np.atleast_2d since it lets np.matrix through (different API from np.array)

Random state

• validation.check_random_state: randomness must be handle with np.random.RandomState only

Estimators

• validation.check_is_fitted: check that estimator has been fitted before calling predict or transform method
• validation.has_fit_parameter: check that the fit method has a given parameter
• all_estimators: return a list of all scikit-learn estimators
• multiclass.type_of_target: return the type of the target y among continuous, continuous-multioutput, binary, multiclass, multiclass-multioutput, multilabel-indicator or unknown
• multiclass.unique_labels: Extract an ordered array of unique labels, "multiclass-multioutput" input type not allowed.

Warnings & Exceptions

• utils.deprecated: decorator to mark a function or class as deprecated
• exceptions.ConvergenceWarning: Custom warning to catch convergence problems (used in Lasso)