Feature engineering для time series на собеседовании Data Scientist

Lag features

Past values как feature.

df['lag_1'] = df['sales'].shift(1)
df['lag_7'] = df['sales'].shift(7)
df['lag_30'] = df['sales'].shift(30)

Daily, weekly, monthly lags — common.

Caveat. Не leak future. На time t — only data до t.

Rolling statistics

Aggregate over window.

df['rolling_mean_7'] = df['sales'].rolling(7).mean()
df['rolling_std_30'] = df['sales'].rolling(30).std()
df['rolling_max_14'] = df['sales'].rolling(14).max()

Useful — trend, volatility.

Expanding window. From start до t.

Seasonal features

Cyclic encoding — see «feature engineering».

df['hour_sin'] = np.sin(2 * np.pi * df['hour'] / 24)
df['day_of_week'] = df['ts'].dt.dayofweek
df['day_of_month'] = df['ts'].dt.day
df['week_of_year'] = df['ts'].dt.isocalendar().week

Holiday / event features

import holidays
df['is_holiday'] = df['date'].apply(lambda d: d в holidays.RU())
df['is_weekend'] = df['day_of_week'].isin([5, 6])

Custom events. Marketing campaigns, product launches.

df['campaign_active'] = df['date'].between('2026-05-01', '2026-05-15')

Cross-series features

Aggregate from related series.

Hierarchical. Total category sales (related items).

Geo. Average across nearby stores.

Cluster. Mean внутри similar items group.

Helps когда individual series sparse.

Связанные темы

• ARIMA на собесе DS
• Holt-Winters для DS
• Forecasting system design для DS
• Feature engineering для DS
• Подготовка к собесу Data Scientist

FAQ

Это официальная информация?

Нет. Статья основана на time series ML practices.

Тренируйте Data Science — откройте тренажёр с 1500+ вопросами для собесов.