242 KiB
242 KiB
In [7]:
import re
import pandas as pd
from sklearn import set_config
from sklearn.model_selection import train_test_split
import pymorphy3
set_config(transform_output="pandas")
morph = pymorphy3.MorphAnalyzer()
def _preprocess_text(text):
text = text.lower()
text = re.sub(r'\s+', ' ', text).strip()
return text
def _remove_punctuation(text):
return re.sub(r'[^\w\s]', ' ', text)
def _remove_emojis(text):
emoji_pattern = re.compile(
"[\U0001F600-\U0001F64F"
"\U0001F300-\U0001F5FF"
"\U0001F680-\U0001F6FF"
"\U0001F700-\U0001F77F"
"\U0001F780-\U0001F7FF"
"\U0001F800-\U0001F8FF"
"\U0001F900-\U0001F9FF"
"\U0001FA70-\U0001FAFF"
"\U00002702-\U000027B0"
"\U000024C2-\U0001F251"
"]+", flags=re.UNICODE
)
return emoji_pattern.sub(r'', text)
def _lemmatize(text):
words = text.split()
lemmas = [morph.parse(word)[0].normal_form for word in words]
return ' '.join(lemmas)
def preprocess_for_prediction(text):
text = _remove_punctuation(text)
text = _remove_emojis(text)
text = _preprocess_text(text)
text = _lemmatize(text)
return text
preprocess_for_prediction("Купили эти наушники пару недель назад — звук просто потрясающий! 🎧 Но через несколько дней начали трещать на высокой громкости. Немного разочарованы 😕, хотя поначалу всё нравилось.")
Out[7]:
In [8]:
from sklearn.utils import shuffle
def load_and_prepare_data(file_path):
data = pd.read_csv(file_path, sep=',', encoding='utf-8')
data['text'] = data['text'].astype(str).apply(preprocess_for_prediction)
data = data[data['text'].str.strip() != '']
data['class'] = data['Полученный класс'].apply(lambda x: 0 if x == 1 else 1)
data = shuffle(data, random_state=42).reset_index(drop=True)
return data
In [9]:
import numpy as np
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.utils.class_weight import compute_class_weight
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.naive_bayes import MultinomialNB
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import LinearSVC
from sklearn.metrics import (
classification_report, accuracy_score, precision_score,
recall_score, f1_score, confusion_matrix
)
from imblearn.over_sampling import RandomOverSampler
df = load_and_prepare_data("data/wb_reviews.csv")
print("Распределение классов до балансировки:")
print(df['class'].value_counts())
X = df['text']
y = df['class']
In [10]:
from imblearn.under_sampling import RandomUnderSampler
rus = RandomUnderSampler(random_state=42)
X_resampled, y_resampled = rus.fit_resample(X.values.reshape(-1,1), y)
X_resampled = pd.Series(X_resampled.flatten())
print("\nРаспределение классов после балансировки (undersampling):")
print(pd.Series(y_resampled).value_counts())
In [11]:
df_class_0 = df[df['class'] == 0]
print(df_class_0)
In [12]:
X_trainval, X_test, y_trainval, y_test = train_test_split(
X_resampled, y_resampled, test_size=0.2, random_state=42, stratify=y_resampled)
X_train, X_val, y_train, y_val = train_test_split(
X_trainval, y_trainval, test_size=0.25, random_state=42, stratify=y_trainval)
In [13]:
classes = np.array(sorted(set(y_train)))
class_weights = compute_class_weight(class_weight='balanced', classes=classes, y=y_train)
class_weights_dict = dict(zip(classes, class_weights))
print("\nClass Weights:", class_weights_dict)
In [14]:
models = {
"Logistic Regression": LogisticRegression(max_iter=1000, class_weight=class_weights_dict),
"SVM (Linear)": LinearSVC(class_weight=class_weights_dict),
"Naive Bayes": MultinomialNB(),
"Random Forest": RandomForestClassifier(class_weight=class_weights_dict, random_state=42)
}
results = {}
for name, model in models.items():
pipe = Pipeline([
('tfidf', TfidfVectorizer()),
('clf', model)
])
pipe.fit(X_train, y_train)
y_pred = pipe.predict(X_val)
report = classification_report(y_val, y_pred, output_dict=True)
cm = confusion_matrix(y_val, y_pred)
y_pred_test = pipe.predict(X_test)
report_test = classification_report(y_test, y_pred_test, output_dict=True)
cm_test = confusion_matrix(y_test, y_pred_test)
results[name] = {
'val_report': report,
'val_cm': cm,
'test_report':report_test,
'test_cm':cm
}
display("\n[Валидация]")
display(classification_report(y_val, y_pred))
display("Confusion matrix (validation):\n", cm)
display("\n[Тест]")
display(classification_report(y_test, y_pred_test))
display("Confusion matrix (validation):\n", cm_test)
In [15]:
fig, axes = plt.subplots(2, 2, figsize=(12, 10))
axes = axes.flatten()
for i, (name, data) in enumerate(results.items()):
cm = data['test_cm']
ax = axes[i]
im = ax.imshow(cm, cmap='Blues')
ax.set_title(name)
ax.set_xlabel('Predicted')
ax.set_ylabel('Actual')
for (j, k), val in np.ndenumerate(cm):
ax.text(k, j, str(val), ha='center', va='center')
fig.suptitle("Матрицы ошибок моделей", fontsize=16)
plt.tight_layout()
plt.show()
metric_names = ['accuracy', 'precision', 'recall', 'f1-score']
metrics_summary = {metric: [] for metric in metric_names}
model_names = list(results.keys())
for model in model_names:
report = results[model]['test_report']
metrics_summary['f1-score'].append(report['weighted avg']['f1-score'])
metrics_summary['precision'].append(report['weighted avg']['precision'])
metrics_summary['recall'].append(report['weighted avg']['recall'])
acc = accuracy_score(
y_test,
Pipeline([('tfidf', TfidfVectorizer()), ('clf', models[model])])
.fit(X_train, y_train).predict(X_test)
)
metrics_summary['accuracy'].append(acc)
for metric in metric_names:
plt.figure(figsize=(8, 4))
plt.plot(model_names, metrics_summary[metric], marker='o', linestyle='-')
plt.title(f'{metric.capitalize()} по моделям (валидация)')
plt.xlabel('Модель')
plt.ylabel(metric.capitalize())
plt.ylim(0, 1.05)
plt.grid(True, linestyle='--', alpha=0.5)
plt.xticks(rotation=15)
plt.tight_layout()
plt.show()
