Coverage for mlinsights/metrics/scoring_metrics.py: 100%

1""" 

2@file 

3@brief Metrics to compare machine learning. 

4""" 

5import numpy 

6from sklearn.metrics import r2_score 

7 

8_known_functions = { 

9 'exp': numpy.exp, 

10 'log': numpy.log 

11} 

12 

13 

14def comparable_metric(metric_function, y_true, y_pred, 

15 tr="log", inv_tr='exp', **kwargs): 

16 """ 

17 Applies function on either the true target or/and the predictions 

18 before computing r2 score. 

19 

20 :param metric_function: metric to compute 

21 :param y_true: expected targets 

22 :param y_pred: predictions 

23 :param sample_weight: weights 

24 :param multioutput: see :epkg:`sklearn:metrics:r2_score` 

25 :param tr: transformation applied on the target 

26 :param inv_tr: transformation applied on the predictions 

27 :return: results 

28 """ 

29 tr = _known_functions.get(tr, tr) 

30 inv_tr = _known_functions.get(inv_tr, inv_tr) 

31 if tr is not None and not callable(tr): 

32 raise TypeError("Argument tr must be callable.") 

33 if inv_tr is not None and not callable(inv_tr): 

34 raise TypeError("Argument inv_tr must be callable.") 

35 if tr is None and inv_tr is None: 

36 raise ValueError( 

37 "tr and inv_tr cannot be both None at the same time.") 

38 if tr is None: 

39 return metric_function(y_true, inv_tr(y_pred), **kwargs) 

40 if inv_tr is None: 

41 return metric_function(tr(y_true), y_pred, **kwargs) 

42 return metric_function(tr(y_true), inv_tr(y_pred), **kwargs) 

43 

44 

45def r2_score_comparable(y_true, y_pred, *, sample_weight=None, 

46 multioutput='uniform_average', 

47 tr=None, inv_tr=None): 

48 """ 

49 Applies function on either the true target or/and the predictions 

50 before computing r2 score. 

51 

52 :param y_true: expected targets 

53 :param y_pred: predictions 

54 :param sample_weight: weights 

55 :param multioutput: see :epkg:`sklearn:metrics:r2_score` 

56 :param tr: transformation applied on the target 

57 :param inv_tr: transformation applied on the predictions 

58 :return: results 

59 

60 Example: 

61 

62 .. runpython:: 

63 :showcode: 

64 

65 import numpy 

66 from sklearn import datasets 

67 from sklearn.model_selection import train_test_split 

68 from sklearn.linear_model import LinearRegression 

69 from sklearn.metrics import r2_score 

70 from mlinsights.metrics import r2_score_comparable 

71 

72 iris = datasets.load_iris() 

73 X = iris.data[:, :4] 

74 y = iris.target + 1 

75 

76 X_train, X_test, y_train, y_test = train_test_split(X, y) 

77 

78 model1 = LinearRegression().fit(X_train, y_train) 

79 print('r2', r2_score(y_test, model1.predict(X_test))) 

80 print('r2 log', r2_score(numpy.log(y_test), numpy.log(model1.predict(X_test)))) 

81 print('r2 log comparable', r2_score_comparable( 

82 y_test, model1.predict(X_test), tr="log", inv_tr="log")) 

83 

84 model2 = LinearRegression().fit(X_train, numpy.log(y_train)) 

85 print('r2', r2_score(numpy.log(y_test), model2.predict(X_test))) 

86 print('r2 log', r2_score(y_test, numpy.exp(model2.predict(X_test)))) 

87 print('r2 log comparable', r2_score_comparable( 

88 y_test, model2.predict(X_test), inv_tr="exp")) 

89 """ 

90 return comparable_metric(r2_score, y_true, y_pred, 

91 sample_weight=sample_weight, 

92 multioutput=multioutput, 

93 tr=tr, inv_tr=inv_tr)