TelegramOpen channel
Anatolii Palash
Merged · Machine Learning

Класифікація за допомогою sklearn

Розглянемо, як розв’язувати задачі класифікації з використанням бібліотеки sklearn. На двох практичних прикладах - текстові дані та класи з сильним дисбалансом - розберемо повний підхід до побудови, навчання й коректної оцінки моделей.

Code and results

Published version after Merge. Results were saved by the author; code does not run on this page.

HTML, SVG, віджети та інтерактивні результати не переносяться. Збережено безпечний текст або зображення.

Python · 1
# В даному ноуті виконав пару завдань - два різних датасети
# У першому датасеті дизбаланс даних несуттєвий тому після нормалізації данних
# та застосуванні моделі GradientBoostingClassifier метрики виглядають добре
# на мій недосвідчений погляд :)

import kagglehub
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
import os
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import (
    classification_report,
    f1_score,
    precision_score,
    recall_score,
    roc_auc_score,
    precision_recall_curve
    )

# Перший датасет - передбачення хвороби серця - дизбаланс несуттєвий

path = kagglehub.dataset_download("fedesoriano/heart-failure-prediction")
df = pd.read_csv(os.path.join(path, "heart.csv"), sep=",")
Python · 2
df.head(3)
Results saved by the author in Colab
   Age Sex ChestPainType  RestingBP  Cholesterol  FastingBS RestingECG  MaxHR  \
0   40   M           ATA        140          289          0     Normal    172   
1   49   F           NAP        160          180          0     Normal    156   
2   37   M           ATA        130          283          0         ST     98   

  ExerciseAngina  Oldpeak ST_Slope  HeartDisease  
0              N      0.0       Up             0  
1              N      1.0     Flat             1  
2              N      0.0       Up             0  
Python · 3
df["HeartDisease"].value_counts()
Results saved by the author in Colab
HeartDisease
1    508
0    410
Name: count, dtype: int64
Python · 4
TARGET_NAME="HeartDisease"
X = df.drop(columns=[TARGET_NAME])
y=df[TARGET_NAME]

X_train, X_valid, y_train, y_valid = train_test_split(
    X, y,
    test_size=0.33,
    random_state=42
)


Python · 5
# normalization of data
float_cols = X_train.select_dtypes(include="float").columns.tolist()
int_cols = X_train.select_dtypes(include="int").columns.tolist()
cat_cols = X_train.select_dtypes(include="object").columns.tolist()
Python · 6
X_train[cat_cols]
Results saved by the author in Colab
    Sex ChestPainType RestingECG ExerciseAngina ST_Slope
585   M           ATA         ST              N     Flat
169   M           NAP     Normal              N       Up
602   M           NAP     Normal              Y     Flat
140   M           ASY     Normal              Y     Flat
173   M           NAP     Normal              N       Up
..   ..           ...        ...            ...      ...
106   F           ASY         ST              N       Up
270   M           ASY     Normal              N       Up
860   M           ASY     Normal              Y       Up
435   M           ASY         ST              Y       Up
102   F           ASY     Normal              N     Flat

[615 rows x 5 columns]
Python · 7
ohe = OneHotEncoder(handle_unknown="ignore", sparse_output=False)
X_train_ohe = ohe.fit_transform(X_train[cat_cols])
X_valid_ohe = ohe.transform(X_valid[cat_cols])

X_train_ohe_df = pd.DataFrame(X_train_ohe, columns=ohe.get_feature_names_out(cat_cols), index=X_train.index)
X_valid_ohe_df = pd.DataFrame(X_valid_ohe, columns=ohe.get_feature_names_out(cat_cols), index=X_valid.index)
Python · 8
X_train_ohe_df.head(3)
Results saved by the author in Colab
     Sex_F  Sex_M  ChestPainType_ASY  ChestPainType_ATA  ChestPainType_NAP  \
585    0.0    1.0                0.0                1.0                0.0   
169    0.0    1.0                0.0                0.0                1.0   
602    0.0    1.0                0.0                0.0                1.0   

     ChestPainType_TA  RestingECG_LVH  RestingECG_Normal  RestingECG_ST  \
585               0.0             0.0                0.0            1.0   
169               0.0             0.0                1.0            0.0   
602               0.0             0.0                1.0            0.0   

     ExerciseAngina_N  ExerciseAngina_Y  ST_Slope_Down  ST_Slope_Flat  \
585               1.0               0.0            0.0            1.0   
169               1.0               0.0            0.0            0.0   
602               0.0               1.0            0.0            1.0   

     ST_Slope_Up  
585          0.0  
169          1.0  
602          0.0  
Python · 9
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train[float_cols])
X_valid_scaled = scaler.transform(X_valid[float_cols])

X_train_scaled_df = pd.DataFrame(X_train_scaled, columns=float_cols, index=X_train.index)
X_valid_scaled_df = pd.DataFrame(X_valid_scaled, columns=float_cols, index=X_valid.index)

X_train_scaled_df.head(3)
Results saved by the author in Colab
      Oldpeak
585 -0.106065
169 -0.826835
602 -0.826835
Python · 10
df.describe()
Results saved by the author in Colab
              Age   RestingBP  Cholesterol   FastingBS       MaxHR  \
count  918.000000  918.000000   918.000000  918.000000  918.000000   
mean    53.510893  132.396514   198.799564    0.233115  136.809368   
std      9.432617   18.514154   109.384145    0.423046   25.460334   
min     28.000000    0.000000     0.000000    0.000000   60.000000   
25%     47.000000  120.000000   173.250000    0.000000  120.000000   
50%     54.000000  130.000000   223.000000    0.000000  138.000000   
75%     60.000000  140.000000   267.000000    0.000000  156.000000   
max     77.000000  200.000000   603.000000    1.000000  202.000000   

          Oldpeak  HeartDisease  
count  918.000000    918.000000  
mean     0.887364      0.553377  
std      1.066570      0.497414  
min     -2.600000      0.000000  
25%      0.000000      0.000000  
50%      0.600000      1.000000  
75%      1.500000      1.000000  
max      6.200000      1.000000  
Python · 11
X_train_df_final = pd.concat([X_train_ohe_df, X_train_scaled_df, X_train[int_cols]], axis=1)
X_valid_df_final = pd.concat([X_valid_ohe_df, X_valid_scaled_df, X_valid[int_cols]], axis=1)
Python · 12
X_train_df_final.head(2)
Results saved by the author in Colab
     Sex_F  Sex_M  ChestPainType_ASY  ChestPainType_ATA  ChestPainType_NAP  \
585    0.0    1.0                0.0                1.0                0.0   
169    0.0    1.0                0.0                0.0                1.0   

     ChestPainType_TA  RestingECG_LVH  RestingECG_Normal  RestingECG_ST  \
585               0.0             0.0                0.0            1.0   
169               0.0             0.0                1.0            0.0   

     ExerciseAngina_N  ExerciseAngina_Y  ST_Slope_Down  ST_Slope_Flat  \
585               1.0               0.0            0.0            1.0   
169               1.0               0.0            0.0            0.0   

     ST_Slope_Up   Oldpeak  Age  RestingBP  Cholesterol  FastingBS  MaxHR  
585          0.0 -0.106065   57        180          285          1    120  
169          1.0 -0.826835   58        140          179          0    160  
Python · 13
print(X_train.shape[0] == X_train_df_final.shape[0])
Results saved by the author in Colab
True
Python · 14
print(X_train_df_final.isna().sum())
Results saved by the author in Colab
Sex_F                0
Sex_M                0
ChestPainType_ASY    0
ChestPainType_ATA    0
ChestPainType_NAP    0
ChestPainType_TA     0
RestingECG_LVH       0
RestingECG_Normal    0
RestingECG_ST        0
ExerciseAngina_N     0
ExerciseAngina_Y     0
ST_Slope_Down        0
ST_Slope_Flat        0
ST_Slope_Up          0
Oldpeak              0
Age                  0
RestingBP            0
Cholesterol          0
FastingBS            0
MaxHR                0
dtype: int64
Python · 15
gb_model = GradientBoostingClassifier(
    n_estimators=1000
)

gb_model.fit(X_train_df_final, y_train)
y_valid_pred = gb_model.predict(X_valid_df_final)
Python · 16
print(classification_report(y_valid, y_valid_pred))
Results saved by the author in Colab
              precision    recall  f1-score   support

           0       0.78      0.89      0.83       123
           1       0.92      0.83      0.87       180

    accuracy                           0.85       303
   macro avg       0.85      0.86      0.85       303
weighted avg       0.86      0.85      0.86       303

Python · 17
# Другий датасет - передбачення виживання пасажирів Титаніку. Дані більш дизбалансовані ніж в попередній задачі.
# Було використано підходи з уроку - нормалізація даних - застосування кількох моделей та файнтюнинг моделі
# з підбором таргетованого реколу


import kagglehub
path = kagglehub.dataset_download("heptapod/titanic")
df = pd.read_csv(os.path.join(path, "train_and_test2.csv"), sep=",")
Results saved by the author in Colab
Using Colab cache for faster access to the 'titanic' dataset.
Python · 18
df.head(3)
Results saved by the author in Colab
   Passengerid   Age     Fare  Sex  sibsp  zero  zero.1  zero.2  zero.3  \
0            1  22.0   7.2500    0      1     0       0       0       0   
1            2  38.0  71.2833    1      1     0       0       0       0   
2            3  26.0   7.9250    1      0     0       0       0       0   

   zero.4  ...  zero.12  zero.13  zero.14  Pclass  zero.15  zero.16  Embarked  \
0       0  ...        0        0        0       3        0        0       2.0   
1       0  ...        0        0        0       1        0        0       0.0   
2       0  ...        0        0        0       3        0        0       2.0   

   zero.17  zero.18  2urvived  
0        0        0         0  
1        0        0         1  
2        0        0         1  

[3 rows x 28 columns]
Python · 19

df.isna().sum()
Results saved by the author in Colab
Passengerid    0
Age            0
Fare           0
Sex            0
sibsp          0
zero           0
zero.1         0
zero.2         0
zero.3         0
zero.4         0
zero.5         0
zero.6         0
Parch          0
zero.7         0
zero.8         0
zero.9         0
zero.10        0
zero.11        0
zero.12        0
zero.13        0
zero.14        0
Pclass         0
zero.15        0
zero.16        0
Embarked       2
zero.17        0
zero.18        0
2urvived       0
dtype: int64
Python · 20
df.dropna(subset="Embarked", inplace=True)
Python · 21
plt.bar(counts.index[0].astype(str), counts.values[0], label="Клас 0")
plt.bar(counts.index[1].astype(str), counts.values[1], label="Class 1")

plt.show()
Results saved by the author in ColabCell 21 output
Python · 22
df.describe()
Results saved by the author in Colab
       Passengerid          Age         Fare          Sex        sibsp  \
count  1307.000000  1307.000000  1307.000000  1307.000000  1307.000000   
mean    655.319816    29.471821    33.209595     0.355011     0.499617   
std     377.922205    12.881592    51.748768     0.478700     1.042273   
min       1.000000     0.170000     0.000000     0.000000     0.000000   
25%     328.500000    22.000000     7.895800     0.000000     0.000000   
50%     655.000000    28.000000    14.454200     0.000000     0.000000   
75%     982.500000    35.000000    31.275000     1.000000     1.000000   
max    1309.000000    80.000000   512.329200     1.000000     8.000000   

         zero  zero.1  zero.2  zero.3  zero.4  ...  zero.12  zero.13  zero.14  \
count  1307.0  1307.0  1307.0  1307.0  1307.0  ...   1307.0   1307.0   1307.0   
mean      0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   
std       0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   
min       0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   
25%       0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   
50%       0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   
75%       0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   
max       0.0     0.0     0.0     0.0     0.0  ...      0.0      0.0      0.0   

            Pclass  zero.15  zero.16     Embarked  zero.17  zero.18  \
count  1307.000000   1307.0   1307.0  1307.000000   1307.0   1307.0   
mean      2.296863      0.0      0.0     1.492731      0.0      0.0   
std       0.836942      0.0      0.0     0.814626      0.0      0.0   
min       1.000000      0.0      0.0     0.000000      0.0      0.0   
25%       2.000000      0.0      0.0     1.000000      0.0      0.0   
50%       3.000000      0.0      0.0     2.000000      0.0      0.0   
75%       3.000000      0.0      0.0     2.000000      0.0      0.0   
max       3.000000      0.0      0.0     2.000000      0.0      0.0   

          2urvived  
count  1307.000000  
mean      0.260138  
std       0.438877  
min       0.000000  
25%       0.000000  
50%       0.000000  
75%       1.000000  
max       1.000000  

[8 rows x 28 columns]
Python · 23
df.drop(list(df.filter(regex='zero')), axis=1, inplace=True)
df.drop(columns=["Passengerid"], inplace=True)
df.rename(columns={"2urvived": "Survived"}, inplace=True)
Python · 24
df.describe()
Results saved by the author in Colab
               Age         Fare          Sex        sibsp        Parch  \
count  1309.000000  1309.000000  1309.000000  1309.000000  1309.000000   
mean     29.503186    33.281086     0.355997     0.498854     0.385027   
std      12.905241    51.741500     0.478997     1.041658     0.865560   
min       0.170000     0.000000     0.000000     0.000000     0.000000   
25%      22.000000     7.895800     0.000000     0.000000     0.000000   
50%      28.000000    14.454200     0.000000     0.000000     0.000000   
75%      35.000000    31.275000     1.000000     1.000000     0.000000   
max      80.000000   512.329200     1.000000     8.000000     9.000000   

            Pclass     Embarked     Survived  
count  1309.000000  1307.000000  1309.000000  
mean      2.294882     1.492731     0.261268  
std       0.837836     0.814626     0.439494  
min       1.000000     0.000000     0.000000  
25%       2.000000     1.000000     0.000000  
50%       3.000000     2.000000     0.000000  
75%       3.000000     2.000000     1.000000  
max       3.000000     2.000000     1.000000  
Python · 25
TARGET_NAME="Survived"
X = df.drop(columns=[TARGET_NAME])
y = df[TARGET_NAME]

X_train, X_valid, y_train, y_valid = train_test_split(
    X, y,
    test_size=0.33,
    shuffle=True,
    random_state=42
)
Python · 26
float_cols = X_train.select_dtypes(include="float").columns.tolist()
int_cols = X_train.select_dtypes(include="int").columns.tolist()

scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train[float_cols])
X_valid_scaled = scaler.transform(X_valid[float_cols])

X_train_scaled_df = pd.DataFrame(X_train_scaled, columns=float_cols, index=X_train.index)
X_valid_scaled_df = pd.DataFrame(X_valid_scaled, columns=float_cols, index=X_valid.index)
Python · 27
int_cols
Results saved by the author in Colab
['Sex', 'sibsp', 'Parch', 'Pclass']
Python · 28
X_train_scaled_df.head(3)
Results saved by the author in Colab
           Age      Fare  Embarked
377  -0.228783  3.285404 -1.881209
1169  0.009351 -0.243911  0.612818
495  -0.149405 -0.365106 -1.881209
Python · 29
X_train_df_final = pd.concat([X_train[int_cols], X_train_scaled_df], axis=1)
X_valid_df_final = pd.concat([X_valid[int_cols], X_valid_scaled_df], axis=1)
Python · 30
X_train_df_final.head(3)
X_train_df_final.isna().sum()
Results saved by the author in Colab
Sex         0
sibsp       0
Parch       0
Pclass      0
Age         0
Fare        0
Embarked    0
dtype: int64
Python · 31
log_model = LogisticRegression()
log_model.fit(X_train_df_final, y_train)
y_valid_pred = log_model.predict(X_valid_df_final)

print(classification_report(y_valid, y_valid_pred))
Results saved by the author in Colab
              precision    recall  f1-score   support

           0       0.79      0.94      0.86       308
           1       0.72      0.37      0.49       124

    accuracy                           0.78       432
   macro avg       0.75      0.66      0.67       432
weighted avg       0.77      0.78      0.75       432

Python · 32
gb_model = GradientBoostingClassifier(
    n_estimators=1000
)

gb_model.fit(X_train_df_final, y_train)
y_valid_pred = gb_model.predict(X_valid_df_final)

print(classification_report(y_valid, y_valid_pred))
Results saved by the author in Colab
              precision    recall  f1-score   support

           0       0.80      0.86      0.83       308
           1       0.57      0.46      0.51       124

    accuracy                           0.75       432
   macro avg       0.68      0.66      0.67       432
weighted avg       0.73      0.75      0.74       432

Python · 33
from imblearn.over_sampling import SMOTE

smote = SMOTE(random_state=42)

X_train_smote, y_train_smote = smote.fit_resample(
    X_train_df_final, y_train
)

y_train_smote.value_counts()
Results saved by the author in Colab
Survived
0    659
1    659
Name: count, dtype: int64
Python · 34
import lightgbm as lgbm

lgbm_model = lgbm.LGBMClassifier(num_leaves = 3,
                  n_estimators = 500,
                  min_child_samples = 20,
                  max_depth = 15,
                  learning_rate = 0.1,
                  random_state=42,
                  verbose=-1
                )
lgbm_model.fit(X_train_smote, y_train_smote)
y_valid_pred = lgbm_model.predict(X_valid_df_final)

print(classification_report(y_valid, y_valid_pred))
Results saved by the author in Colab
              precision    recall  f1-score   support

           0       0.85      0.81      0.83       308
           1       0.57      0.65      0.61       124

    accuracy                           0.76       432
   macro avg       0.71      0.73      0.72       432
weighted avg       0.77      0.76      0.76       432

Python · 35
y_valid_pred_proba = lgbm_model.predict_proba(X_valid_df_final)[:, 1]

precision, recall, threshold = precision_recall_curve(y_valid, y_valid_pred_proba)
f1_score = (2 * precision * recall) / (precision + recall)

idx = np.argmax(f1_score)

idx
Results saved by the author in Colab
np.int64(182)
Python · 36
y_valid_pred = [1 if el >= threshold[idx] else 0 for el in y_valid_pred_proba]
print(classification_report(y_valid, y_valid_pred))
Results saved by the author in Colab
              precision    recall  f1-score   support

           0       0.91      0.67      0.77       308
           1       0.50      0.84      0.63       124

    accuracy                           0.72       432
   macro avg       0.71      0.75      0.70       432
weighted avg       0.79      0.72      0.73       432

Python · 37
target_recall = 0.9
idxs = np.where(recall[:-1] >= target_recall)[0]
if len(idxs) == 0:
    print(f"Recall {target_recall} недосяжний для даної моделі/даних")
else:
    i = idxs[-1]  # самий високий поріг при recall>=target
    thr = threshold[i]
    y_pred = (y_valid_pred_proba >= thr).astype(int)
    print("thr:", thr)
    print(classification_report(y_valid, y_pred))
Results saved by the author in Colab
thr: 0.27045589755621346
              precision    recall  f1-score   support

           0       0.94      0.57      0.71       308
           1       0.46      0.90      0.61       124

    accuracy                           0.66       432
   macro avg       0.70      0.74      0.66       432
weighted avg       0.80      0.66      0.68       432

Community reviews

Anar LavrenovApprove

Відмінна робота. Використання оверсемплінгу окремо заслуговує відзнаки. Можна мерджати