Neste desafio, aplicar o fluxo de trabalho completo aprendido no curso — desde o pré-processamento de dados até o treinamento e avaliação do modelo.


import unittest
import pandas as pd
import numpy as np

def _dynamic_test(test_case, condition, success_message, failure_message):
    if condition:
        test_case._testMethodName = success_message
        test_case.assertTrue(True, success_message)
    else:
        test_case._testMethodName = failure_message
        test_case.fail(failure_message)

class TestPipelineWithGridSearch(unittest.TestCase):

    @classmethod
    def setUpClass(cls):
        cls.df = pd.read_csv(
            'https://codefinity-content-media.s3.eu-west-1.amazonaws.com/a65bbc96-309e-4df9-a790-a1eb8c815a1c/penguins.csv'
        )
        cls.df = cls.df[cls.df.isna().sum(axis=1) < 2]
        import user_code
        cls.uc = user_code  # очікується: label_enc, y, X_train, X_test, y_train, y_test, ct, param_grid, grid_search, pipe

    def test_target_encoded_with_labelencoder(self):
        from sklearn.preprocessing import LabelEncoder
        uc = self.uc
        y_arr = np.asarray(uc.y)
        cond = isinstance(getattr(uc, 'label_enc', None), LabelEncoder) \
               and hasattr(uc.label_enc, 'classes_') \
               and y_arr.ndim == 1 and np.issubdtype(y_arr.dtype, np.integer)
        _dynamic_test(
            self,
            cond,
            "Target is encoded with LabelEncoder to integer dtype",
            "Target must be encoded with LabelEncoder to integer dtype"
        )

    def test_train_test_split_33_percent(self):
        uc = self.uc
        has_vars = all(hasattr(uc, v) for v in ['X_train', 'X_test', 'y_train', 'y_test'])
        if not has_vars:
            cond = False
        else:
            n_total = len(uc.X_train) + len(uc.X_test)
            expected_test = int(round(0.33 * n_total))
            cond = len(uc.y_train) + len(uc.y_test) == n_total and abs(len(uc.X_test) - expected_test) <= 2
        _dynamic_test(
            self,
            cond,
            "Data split uses approximately 33% for the test set",
            "Data must be split with test_size=0.33"
        )

    def test_columntransformer_ohe_passthrough(self):
        from sklearn.compose import ColumnTransformer
        from sklearn.preprocessing import OneHotEncoder
        uc = self.uc
        ct = getattr(uc, 'ct', None)
        cond_ct = isinstance(ct, ColumnTransformer)
        found_ohe = False
        if cond_ct:
            for _, trans, cols in ct.transformers:
                if isinstance(trans, OneHotEncoder):
                    cols_set = set(cols) if isinstance(cols, (list, tuple)) else {cols}
                    if cols_set == {'island', 'sex'}:
                        found_ohe = True
                        break
        cond = cond_ct and found_ohe and getattr(ct, 'remainder', None) == 'passthrough'
        _dynamic_test(
            self,
            cond,
            "ColumnTransformer applies OneHotEncoder to ['island', 'sex'] with remainder='passthrough'",
            "ColumnTransformer must OneHotEncode ['island', 'sex'] and set remainder='passthrough'"
        )

    def test_param_grid_values(self):
        uc = self.uc
        pg = getattr(uc, 'param_grid', None)
        
        cond = False
        if isinstance(pg, dict) and 'n_neighbors' in pg:
            n_neighbors_vals = pg['n_neighbors']
            # Перевіряємо, що це список (або кортеж), він не порожній, і всі його елементи — непарні числа
            if len(n_neighbors_vals) > 0:
                cond = all(isinstance(x, int) and x % 2 != 0 for x in n_neighbors_vals)
                
        _dynamic_test(
            self,
            cond,
            "param_grid defines odd values for n_neighbors, allowing experimentation",
            "param_grid must be a dict and include 'n_neighbors' with a list of odd integer values"
        )

    def test_gridsearchcv_with_knn(self):
        from sklearn.model_selection import GridSearchCV
        from sklearn.neighbors import KNeighborsClassifier
        uc = self.uc
        gs = getattr(uc, 'grid_search', None)
        cond = isinstance(gs, GridSearchCV) and isinstance(getattr(gs, 'estimator', None), KNeighborsClassifier)
        _dynamic_test(
            self,
            cond,
            "GridSearchCV is initialized with KNeighborsClassifier and given param_grid",
            "GridSearchCV must be initialized with KNeighborsClassifier and given param_grid"
        )

    def test_pipeline_structure_and_order(self):
        # Очікуваний порядок: columntransformer -> simpleimputer -> standardscaler -> gridsearchcv
        uc = self.uc
        step_names = [name for name, _ in getattr(uc, 'pipe', getattr(uc, 'pipeline', object())).steps] \
                     if hasattr(uc, 'pipe') else []
        expected = ['columntransformer', 'simpleimputer', 'standardscaler', 'gridsearchcv']
        cond = step_names == expected
        _dynamic_test(
            self,
            cond,
            "Pipeline steps are in order: ColumnTransformer, SimpleImputer, StandardScaler, GridSearchCV",
            "Pipeline steps must be: ColumnTransformer, SimpleImputer, StandardScaler, GridSearchCV"
        )

    def test_fitted_on_train_and_scores_on_test(self):
        uc = self.uc
        try:
            score = uc.pipe.score(uc.X_test, uc.y_test)
            cond = isinstance(score, (float, np.floating)) and np.isfinite(score)
        except Exception:
            cond = False
        _dynamic_test(
            self,
            cond,
            "Pipeline is fitted on train and computes a finite score on the test set",
            "Pipeline must be fitted on train and compute a finite score on the test set"
        )

    def test_predictions_returned_for_X_test(self):
        uc = self.uc
        try:
            y_pred = uc.pipe.predict(uc.X_test)
            cond = isinstance(y_pred, (np.ndarray, list)) and len(y_pred) == len(uc.X_test)
        except Exception:
            cond = False
        _dynamic_test(
            self,
            cond,
            "Pipeline returns predictions for X_test",
            "Pipeline must return predictions for X_test"
        )

    def test_best_estimator_available(self):
        uc = self.uc
        try:
            be = uc.grid_search.best_estimator_
            cond = be is not None
        except Exception:
            cond = False
        _dynamic_test(
            self,
            cond,
            "Best estimator is available via grid_search.best_estimator_",
            "Best estimator must be available via grid_search.best_estimator_"
        )

if __name__ == "__main__":
    unittest.main()

test_code.py

O aprendizado de máquina está presente em todos os lugares atualmente. Quer aprender por conta própria? Este curso é uma introdução ao mundo do aprendizado de máquina para que você compreenda os conceitos básicos, trabalhe com o Scikit-learn – a biblioteca mais popular para ML – e desenvolva seu primeiro projeto de aprendizado de máquina.
Este curso é destinado a estudantes com conhecimentos básicos em Python, Pandas e Numpy.

Aprenda os conceitos de Machine Learning e o fluxo de trabalho de projetos de ML.

O pré-processamento é provavelmente a etapa mais importante de um projeto de ML. Este capítulo aborda as etapas de pré-processamento necessárias para praticamente qualquer conjunto de dados.

Um pipeline é uma maneira organizada de combinar todas as etapas de pré-processamento, bem como um modelo. Pipelines facilitam muito o treinamento e a utilização de um modelo.

A modelagem é a etapa mais divertida de um projeto de ML. Vamos aprender a construir, ajustar e avaliar o modelo!

Desafio: Integrando Tudo

Solução