Vérification des versions

#version de Python
import sys
sys.version

'3.11.6 | packaged by conda-forge | (main, Oct  3 2023, 10:29:11) [MSC v.1935 64 bit (AMD64)]'

#version de pycaret
import pycaret
pycaret.__version__

'3.2.0'

Importation des données

#dossier par défaut
import os
os.chdir("C:/Users/ricco/Desktop/demo")

#chargement des données
import pandas
donnees = pandas.read_excel("pima-subset-pycaret.xlsx")
donnees.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 724 entries, 0 to 723
Data columns (total 5 columns):
 #   Column     Non-Null Count  Dtype  
---  ------     --------------  -----  
 0   diastolic  724 non-null    int64  
 1   bodymass   724 non-null    float64
 2   age        724 non-null    int64  
 3   plasma     724 non-null    int64  
 4   diabete    724 non-null    object 
dtypes: float64(1), int64(3), object(1)
memory usage: 28.4+ KB

#distribution des classes
donnees.diabete.value_counts()

negative    475
positive    249
Name: diabete, dtype: int64

Expérimentation - Développement du modèle

Initialisation + choix de l'algo + paramètrage de la session

#création d'une session de travail (expérimentations)
#code en mode "objet"
from pycaret.classification import ClassificationExperiment

#instaciation et initialisation
session = ClassificationExperiment()
session.setup(donnees,target="diabete",train_size=0.7,data_split_stratify=True,normalize=True,fold=5,session_id=2023)

#verif.
print(session)

	Description	Value
0	Session id	2023
1	Target	diabete
2	Target type	Binary
3	Target mapping	negative: 0, positive: 1
4	Original data shape	(724, 5)
5	Transformed data shape	(724, 5)
6	Transformed train set shape	(506, 5)
7	Transformed test set shape	(218, 5)
8	Numeric features	4
9	Preprocess	True
10	Imputation type	simple
11	Numeric imputation	mean
12	Categorical imputation	mode
13	Normalize	True
14	Normalize method	zscore
15	Fold Generator	StratifiedKFold
16	Fold Number	5
17	CPU Jobs	-1
18	Use GPU	False
19	Log Experiment	False
20	Experiment Name	clf-default-name
21	USI	1cd6

<pycaret.classification.oop.ClassificationExperiment object at 0x000001BEF717E350>

#lister les modèles dispnibles pour le classement
#aka classification supervisée
algos = session.models()
print(algos)

                                     Name  \
ID                                          
lr                    Logistic Regression   
knn                K Neighbors Classifier   
nb                            Naive Bayes   
dt               Decision Tree Classifier   
svm                   SVM - Linear Kernel   
rbfsvm                SVM - Radial Kernel   
gpc           Gaussian Process Classifier   
mlp                        MLP Classifier   
ridge                    Ridge Classifier   
rf               Random Forest Classifier   
qda       Quadratic Discriminant Analysis   
ada                  Ada Boost Classifier   
gbc          Gradient Boosting Classifier   
lda          Linear Discriminant Analysis   
et                 Extra Trees Classifier   
lightgbm  Light Gradient Boosting Machine   
catboost              CatBoost Classifier   
dummy                    Dummy Classifier   

                                                  Reference  Turbo  
ID                                                                  
lr        sklearn.linear_model._logistic.LogisticRegression   True  
knn       sklearn.neighbors._classification.KNeighborsCl...   True  
nb                           sklearn.naive_bayes.GaussianNB   True  
dt             sklearn.tree._classes.DecisionTreeClassifier   True  
svm       sklearn.linear_model._stochastic_gradient.SGDC...   True  
rbfsvm                             sklearn.svm._classes.SVC  False  
gpc       sklearn.gaussian_process._gpc.GaussianProcessC...  False  
mlp       sklearn.neural_network._multilayer_perceptron....  False  
ridge           sklearn.linear_model._ridge.RidgeClassifier   True  
rf          sklearn.ensemble._forest.RandomForestClassifier   True  
qda       sklearn.discriminant_analysis.QuadraticDiscrim...   True  
ada       sklearn.ensemble._weight_boosting.AdaBoostClas...   True  
gbc         sklearn.ensemble._gb.GradientBoostingClassifier   True  
lda       sklearn.discriminant_analysis.LinearDiscrimina...   True  
et            sklearn.ensemble._forest.ExtraTreesClassifier   True  
lightgbm                    lightgbm.sklearn.LGBMClassifier   True  
catboost                   catboost.core.CatBoostClassifier   True  
dummy                         sklearn.dummy.DummyClassifier   True  

Premier modèle avec les paramètres par défaut

#modélisation sur le train avec la régression logistique
modele_first = session.create_model("lr")
print(modele_first)

	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
Fold
0	0.7549	0.8111	0.7549	0.7488	0.7499	0.4371	0.4399
1	0.7426	0.7682	0.7426	0.7344	0.7356	0.3974	0.4013
2	0.7129	0.7615	0.7129	0.7077	0.7096	0.3530	0.3538
3	0.7921	0.8398	0.7921	0.7920	0.7807	0.5043	0.5222
4	0.7822	0.8662	0.7822	0.7793	0.7717	0.4845	0.4983
Mean	0.7569	0.8094	0.7569	0.7524	0.7495	0.4353	0.4431
Std	0.0284	0.0404	0.0284	0.0304	0.0255	0.0555	0.0617

LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
                   intercept_scaling=1, l1_ratio=None, max_iter=1000,
                   multi_class='auto', n_jobs=None, penalty='l2',
                   random_state=2023, solver='lbfgs', tol=0.0001, verbose=0,
                   warm_start=False)

#évaluation sur l'échantillon test
res = session.predict_model(modele_first)

#affichage - on a le résumé + un data frame des observé vs. prédictions
print(res[['diabete','prediction_label','prediction_score']])

	Model	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
0	Logistic Regression	0.7752	0.8546	0.7752	0.7698	0.7672	0.4736	0.4817

      diabete prediction_label  prediction_score
330  negative         negative            0.9112
190  negative         negative            0.8536
699  positive         negative            0.5047
17   negative         negative            0.5992
660  positive         positive            0.7929
..        ...              ...               ...
672  positive         positive            0.6954
361  negative         negative            0.8807
1    negative         negative            0.9301
544  positive         positive            0.9355
410  negative         negative            0.6067

[218 rows x 3 columns]

Recherche du meilleur modèle

#optimiser les paramètres de l'algo
modele_best = session.tune_model(modele_first,optimize="Accuracy")

	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
Fold
0	0.7549	0.8124	0.7549	0.7488	0.7499	0.4371	0.4399
1	0.7426	0.7678	0.7426	0.7344	0.7356	0.3974	0.4013
2	0.7327	0.7645	0.7327	0.7257	0.7273	0.3892	0.3918
3	0.8020	0.8398	0.8020	0.8057	0.7897	0.5245	0.5472
4	0.7822	0.8662	0.7822	0.7793	0.7717	0.4845	0.4983
Mean	0.7629	0.8101	0.7629	0.7588	0.7548	0.4465	0.4557
Std	0.0257	0.0398	0.0257	0.0297	0.0230	0.0516	0.0591

Fitting 5 folds for each of 10 candidates, totalling 50 fits

#affichage du modèle
modele_best

LogisticRegression(C=0.179, class_weight={}, dual=False, fit_intercept=True,
                   intercept_scaling=1, l1_ratio=None, max_iter=1000,
                   multi_class='auto', n_jobs=None, penalty='l2',
                   random_state=2023, solver='lbfgs', tol=0.0001, verbose=0,
                   warm_start=False)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

#vérif. de nouveau en test
session.predict_model(modele_best)

	Model	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
0	Logistic Regression	0.7798	0.8542	0.7798	0.7750	0.7713	0.4826	0.4920

	diastolic	bodymass	age	plasma	diabete	prediction_label	prediction_score
330	82	34.400002	46	61	negative	negative	0.9007
190	68	27.299999	32	108	negative	negative	0.8444
699	94	32.700001	45	140	positive	negative	0.5031
17	88	39.299999	27	126	negative	negative	0.5953
660	88	35.000000	52	168	positive	positive	0.7775
...	...	...	...	...	...	...	...
672	78	33.799999	31	173	positive	positive	0.6795
361	54	22.299999	24	119	negative	negative	0.8736
1	66	26.600000	31	85	negative	negative	0.9220
544	70	34.700001	62	197	positive	positive	0.9217
410	75	29.900000	28	147	negative	negative	0.6063

218 rows × 7 columns

#affichage de l'importance des variables
session.plot_model(modele_best,plot="feature")

#coefficients
temp_df = pandas.DataFrame({'var':modele_best.feature_names_in_,'coef':modele_best.coef_[0]})
temp_df

	var	coef
0	diastolic	-0.049967
1	bodymass	0.578019
2	age	0.338531
3	plasma	0.918690

Modèle définitif entraîné sur la totalité des données

#modele définitif
modele_final = session.finalize_model(modele_best)

#c'est un pipeline en réalité
#avec un preprocessing par défaut
modele_final

Pipeline(memory=Memory(location=None),
         steps=[('label_encoding',
                 TransformerWrapperWithInverse(exclude=None, include=None,
                                               transformer=LabelEncoder())),
                ('numerical_imputer',
                 TransformerWrapper(exclude=None,
                                    include=['diastolic', 'bodymass', 'age',
                                             'plasma'],
                                    transformer=SimpleImputer(add_indicator=False,
                                                              copy=True,
                                                              fill_value=None,
                                                              keep_empty_features=False,
                                                              mis...
                 TransformerWrapper(exclude=None, include=None,
                                    transformer=StandardScaler(copy=True,
                                                               with_mean=True,
                                                               with_std=True))),
                ('actual_estimator',
                 LogisticRegression(C=0.179, class_weight={}, dual=False,
                                    fit_intercept=True, intercept_scaling=1,
                                    l1_ratio=None, max_iter=1000,
                                    multi_class='auto', n_jobs=None,
                                    penalty='l2', random_state=2023,
                                    solver='lbfgs', tol=0.0001, verbose=0,
                                    warm_start=False))],
         verbose=False)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Déploiement

#création d'une API pour le déploiement
mon_api = session.create_api(modele_final,"api_ricco")

API successfully created. This function only creates a POST API, it doesn't run it automatically. To run your API, please run this command --> !python api_ricco.py