107 KiB
107 KiB
In [1]:
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import LabelEncoder
from sklearn import metrics
from imblearn.over_sampling import RandomOverSampler
from imblearn.under_sampling import RandomUnderSampler
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.metrics import ConfusionMatrixDisplay
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.linear_model import LinearRegression, LogisticRegression
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor, RandomForestClassifier, GradientBoostingClassifier
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.metrics import (
precision_score, recall_score, accuracy_score, roc_auc_score, f1_score,
matthews_corrcoef, cohen_kappa_score, confusion_matrix
)
from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error
import numpy as np
import featuretools as ft
from sklearn.metrics import accuracy_score, classification_report
# Функция для применения oversampling
def apply_oversampling(X, y):
oversampler = RandomOverSampler(random_state=42)
X_resampled, y_resampled = oversampler.fit_resample(X, y)
return X_resampled, y_resampled
# Функция для применения undersampling
def apply_undersampling(X, y):
undersampler = RandomUnderSampler(random_state=42)
X_resampled, y_resampled = undersampler.fit_resample(X, y)
return X_resampled, y_resampled
def split_stratified_into_train_val_test(
df_input,
stratify_colname="y",
frac_train=0.6,
frac_val=0.15,
frac_test=0.25,
random_state=None,
):
"""
Splits a Pandas dataframe into three subsets (train, val, and test)
following fractional ratios provided by the user, where each subset is
stratified by the values in a specific column (that is, each subset has
the same relative frequency of the values in the column). It performs this
splitting by running train_test_split() twice.
Parameters
----------
df_input : Pandas dataframe
Input dataframe to be split.
stratify_colname : str
The name of the column that will be used for stratification. Usually
this column would be for the label.
frac_train : float
frac_val : float
frac_test : float
The ratios with which the dataframe will be split into train, val, and
test data. The values should be expressed as float fractions and should
sum to 1.0.
random_state : int, None, or RandomStateInstance
Value to be passed to train_test_split().
Returns
-------
df_train, df_val, df_test :
Dataframes containing the three splits.
"""
if frac_train + frac_val + frac_test != 1.0:
raise ValueError(
"fractions %f, %f, %f do not add up to 1.0"
% (frac_train, frac_val, frac_test)
)
if stratify_colname not in df_input.columns:
raise ValueError("%s is not a column in the dataframe" % (stratify_colname))
X = df_input # Contains all columns.
y = df_input[
[stratify_colname]
] # Dataframe of just the column on which to stratify.
# Split original dataframe into train and temp dataframes.
df_train, df_temp, y_train, y_temp = train_test_split(
X, y, stratify=y, test_size=(1.0 - frac_train), random_state=random_state
)
# Split the temp dataframe into val and test dataframes.
relative_frac_test = frac_test / (frac_val + frac_test)
df_val, df_test, y_val, y_test = train_test_split(
df_temp,
y_temp,
stratify=y_temp,
test_size=relative_frac_test,
random_state=random_state,
)
assert len(df_input) == len(df_train) + len(df_val) + len(df_test)
return df_train, df_val, df_test
df = pd.read_csv('../data/jio_mart_items.csv')
df.info()
df = df.sample(n=10000 , random_state=42)
In [2]:
print(df['sub_category'].unique())
Как бизнес-цели выделим следующие 2 варианта: 1) Регрессия - предсказание цены по категории (для аналитических систем или улучшения алгоритмов ценообразования) 2) Классификация - определение категории продукта по его подкатегории (для логистических или аналитических систем)
Однако данный датасет весьма плоо подходит для подобных задач.
Выполним подготовку данных
In [3]:
# df.fillna({"category": "NaN", "sub_category": "NaN", "href" : "NaN", "items" : "NaN", "price" : "NaN" }, inplace=True)
df = df.dropna()
data = df.copy()
value_counts = data["category"].value_counts()
rare = value_counts[value_counts < 100].index
data = data[~data["category"].isin(rare)]
data1 = pd.get_dummies(data, columns=['category', 'sub_category'], drop_first=True)
# label_encoder = LabelEncoder()
# data1['sub_category'] = label_encoder.fit_transform(data['sub_category'])
# data1['category'] = label_encoder.fit_transform(data['category'])
# data1['items'] = label_encoder.fit_transform(data['items'])
Определить достижимый уровень качества модели для каждой задачи. На основе имеющихся данных уровень качества моделей регрессии будет низким, поскольку цена слабо коррелирует с категорией (кроме некоторых исключений).
Построим конвейер.
In [4]:
print(data1.columns)
data1.info()
In [83]:
X_reg = data1.drop(['href', 'items', 'price'], axis=1)
y_reg = data1['price']
# Разделение данных
X_train_reg, X_test_reg, y_train_reg, y_test_reg = train_test_split(X_reg, y_reg, test_size=0.2, random_state=42)
# Выбор моделей для регрессии
models_reg = {
'Linear Regression': LinearRegression(),
'Random Forest Regressor': RandomForestRegressor(random_state=42),
'Gradient Boosting Regressor': GradientBoostingRegressor(random_state=42)
}
# Создание конвейера для регрессии
pipelines_reg = {}
for name, model in models_reg.items():
pipelines_reg[name] = Pipeline([
('scaler', StandardScaler()),
('model', model)
])
# Определение сетки гиперпараметров для регрессии
param_grids_reg = {
'Linear Regression': {},
'Random Forest Regressor': {
'model__n_estimators': [100, 200, 300],
'model__max_depth': [None, 10, 20, 30]
},
'Gradient Boosting Regressor': {
'model__n_estimators': [100, 200, 300],
'model__learning_rate': [0.01, 0.1, 0.2],
'model__max_depth': [3, 5, 7]
}
}
# Настройка гиперпараметров для регрессии
best_models_reg = {}
for name, pipeline in pipelines_reg.items():
grid_search = GridSearchCV(pipeline, param_grids_reg[name], cv=5, scoring='neg_mean_squared_error')
grid_search.fit(X_train_reg, y_train_reg)
best_models_reg[name] = {
'pipeline': grid_search.best_estimator_,
'best_params': grid_search.best_params_
}
print(f'Best parameters for {name}: {grid_search.best_params_}')
# Обучение моделей и оценка качества
for model_name in best_models_reg.keys():
print(f"Model: {model_name}")
model_pipeline = best_models_reg[model_name]['pipeline']
model_pipeline.fit(X_train_reg, y_train_reg)
y_train_predict = model_pipeline.predict(X_train_reg)
y_test_predict = model_pipeline.predict(X_test_reg)
best_models_reg[model_name]["preds_train"] = y_train_predict
best_models_reg[model_name]["preds_test"] = y_test_predict
best_models_reg[model_name]["MSE_train"] = mean_squared_error(y_train_reg, y_train_predict)
best_models_reg[model_name]["MSE_test"] = mean_squared_error(y_test_reg, y_test_predict)
best_models_reg[model_name]["R2_train"] = r2_score(y_train_reg, y_train_predict)
best_models_reg[model_name]["R2_test"] = r2_score(y_test_reg, y_test_predict)
best_models_reg[model_name]["MAE_train"] = mean_absolute_error(y_train_reg, y_train_predict)
best_models_reg[model_name]["MAE_test"] = mean_absolute_error(y_test_reg, y_test_predict)
In [11]:
# Кодирование категориальных данных через LabelEncoder
label_encoder = LabelEncoder()
data['sub_category_encoded'] = label_encoder.fit_transform(data['sub_category'])
# Определение признаков (X) и целевой переменной (y)
X = data[['sub_category_encoded']] # Используем закодированный sub_category
y = label_encoder.fit_transform(data['category']) # Кодируем category
# Разделение данных на тренировочную и тестовую выборки
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)
# Создание конвейера для классификатора
pipeline = Pipeline([
('scaler', StandardScaler()), # Масштабирование данных (хотя для категориальных признаков это не всегда нужно)
('classifier', RandomForestClassifier(random_state=42, n_estimators=100, max_depth=10))
])
# Обучение модели
pipeline.fit(X_train, y_train)
# Предсказание на тестовых данных
y_pred = pipeline.predict(X_test)
print("Accuracy:", accuracy_score(y_test, y_pred))
print("Classification Report:\n", classification_report(y_test, y_pred, target_names=label_encoder.inverse_transform(np.unique(y_test))))
# Матрица ошибок
cm = confusion_matrix(y_test, y_pred)
plt.figure(figsize=(10, 8))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues', xticklabels=label_encoder.classes_, yticklabels=label_encoder.classes_)
plt.xlabel('Predicted')
plt.ylabel('Actual')
plt.title('Confusion Matrix')
plt.show()
Модель классификации показывает неплохие результаты, что логично, учитывая структуру датасета.
In [8]:
data.info()
In [9]:
_, ax = plt.subplots(3, 2, figsize=(12, 10), sharex=False, sharey=False)
ax = ax.flatten()
for index, (name, model) in enumerate(best_models_reg.items()):
model_pipeline = model['pipeline']
y_pred_reg = model_pipeline.predict(X_test_reg)
# График фактических значений против предсказанных значений
ax[index * 2].scatter(y_test_reg, y_pred_reg, alpha=0.5)
ax[index * 2].plot([min(y_test_reg), max(y_test_reg)], [min(y_test_reg), max(y_test_reg)], color='red', linestyle='--')
ax[index * 2].set_xlabel('Actual Values')
ax[index * 2].set_ylabel('Predicted Values')
ax[index * 2].set_title(f'{name}: Actual vs Predicted')
# График остатков
residuals = y_test_reg - y_pred_reg
ax[index * 2 + 1].scatter(y_pred_reg, residuals, alpha=0.5)
ax[index * 2 + 1].axhline(y=0, color='red', linestyle='--')
ax[index * 2 + 1].set_xlabel('Predicted Values')
ax[index * 2 + 1].set_ylabel('Residuals')
ax[index * 2 + 1].set_title(f'{name}: Residuals vs Predicted')
plt.subplots_adjust(top=1, bottom=0, hspace=0.4, wspace=0.1)
plt.show()
Модель регресии демонстрирует ужасные результаты ввиду недостаточной корреляции между целеовй характеристикой и строковыми значениями.