FeatureElimination.plot_report

FeatureElimination.plot_report()

Genera un grafico delle performance (train e validation score) rispetto al numero di feature.

Utilizza il DataFrame dell’attributo report generato dal metodo make_report() per creare un grafico a linee che mostra l’andamento dello score di training e di validazione man mano che le feature vengono eliminate. L’asse x rappresenta il numero di feature, l’asse y lo score.

Parameters:

None – (metodo basato sul report risultante dal metodo make_report()).

Esempio:

>>> import pandas as pd
>>> from sklearn.linear_model import LogisticRegression
>>> from cefeste.elimination import FeatureElimination
>>> # Dati di esempio
>>> data = pd.DataFrame({
... 'feature1': [1,2,3,4,5,6,7,8,9,2,3,1,3,6,43,2,4,6,3,2,6,3,2,6,3,2,5],
... 'feature2': [5,4,3,2,1,6,3,7,3,5,8,4,2,9,75,4,5,7,5,2,5,8,6,3,5,7,8],
... 'feature3': [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
... 'feature4': [8,8,8,9,5,3,4,5,6,2,6,8,4,4,4,6,5,7,8,5,9,3,4,7,5,78,8],
... 'feature5': [2,12,4,14,2,3,1,2,3,14,1,1,14,12,15,16,2,12,13,12,1,15,17,2,1,15,1],
... 'target':   [0,1,0,1,0,0,0,0,0,1,1,1,1,1,1,1,0,1,1,1,0,1,1,0,1,1,1]
... })
>>> # Parametri
>>> target_col = 'target'
>>> model = LogisticRegression()
>>> grid = {'C': [0.1, 1, 10]}
>>> # Inizializzazione
>>> fe = FeatureElimination(
...    db=data,
...    target_col=target_col,
...    model=model,
...    grid=grid,
...    min_n_feat_step=1
... )
>>> # Generazione del report
>>> fe.make_report()

	n_feat	train_score	valid_score	n_feat_to_remove	feat_used	feat_to_remove	feat_select	best_estimator
0	5	0.964706	0.891667	1	[feature3, feature5, feature1, feature2, feature4]	[feature3]	[feature5, feature1, feature2, feature4]	LogisticRegression(C=0.1)
1	4	0.964706	0.891667	1	[feature4, feature5, feature2, feature1]	[feature4]	[feature5, feature2, feature1]	LogisticRegression(C=0.1)
2	3	0.982353	0.925000	1	[feature2, feature5, feature1]	[feature2]	[feature5, feature1]	LogisticRegression(C=1)
3	2	0.817647	0.750000	1	[feature1, feature5]	[feature1]	[feature5]	LogisticRegression(C=0.1)
4	1	0.788235	0.791667	0	[feature5]	[]	[feature5]	LogisticRegression(C=0.1)

>>> fe.plot_report()