Ranker Regressor

In [1]:

pip install catboost -qqq

pip install catboost -qqq

Note: you may need to restart the kernel to use updated packages.

In [2]:

# import necessary libraries
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

from sklearn.metrics import mean_absolute_percentage_error as mape
from catboost import CatBoostRegressor
import warnings
warnings.filterwarnings('ignore')

# set the color palette
sns.set_palette(sns.color_palette('deep'))
sns.set(rc = {'figure.figsize': (10, 5)})
sns.set_style('whitegrid')

# import necessary libraries import numpy as np import pandas as pd import seaborn as sns import matplotlib.pyplot as plt from sklearn.metrics import mean_absolute_percentage_error as mape from catboost import CatBoostRegressor import warnings warnings.filterwarnings('ignore') # set the color palette sns.set_palette(sns.color_palette('deep')) sns.set(rc = {'figure.figsize': (10, 5)}) sns.set_style('whitegrid')

1 | Вводная Часть¶

▎Ранжирования Книг¶

Сегодня посмотрим на задачу ранжирования товаров. По сути это процесс упорядочивания товаров по определенным критериям.

Зачем нам это нужно; это влияние на пользовательский опыт:

• Удобство поиска и фильтрации товаров
• Индивидуальные рекомендации на основе поведения пользователя
• Персонализированные результаты на основе истории покупок и предпочтений (рассмотрим в следующем ноутбуке)

Подход который мы рассмотрим; Эту задачу мы можем решить с помощью машинного обучения. Имея исторические данные о приоритетах пользователей мы можем предсказывать implicit/explicit метрики оценки товара.

2 | Amazon Best Seller Dataset¶

▎Чтение данных¶

Рассмотрим ранжирование товаров на примере датасета Amazon Top 50 Bestselling Books 2009 – 2019, Датасет содержит 550 книг

Признаки:

1. Name: Название книги
   
2. Author: Автор книги
   
3. User Rating: Рейтинг
   
4. Reviews: Количество ревью
   
5. Price: Стоимость книги
   
6. Year: Год когда книга попала в бестселлеры
   
7. Genre: Жанр Fiction или Non-fiction
   

In [3]:

path = '/kaggle/input/bestseller/bestsellers with categories.csv'
df = pd.read_csv(path)
df.head()

path = '/kaggle/input/bestseller/bestsellers with categories.csv' df = pd.read_csv(path) df.head()

Out[3]:

	Name	Author	User Rating	Reviews	Price	Year	Genre
0	10-Day Green Smoothie Cleanse	JJ Smith	4.7	17350	8	2016	Non Fiction
1	11/22/63: A Novel	Stephen King	4.6	2052	22	2011	Fiction
2	12 Rules for Life: An Antidote to Chaos	Jordan B. Peterson	4.7	18979	15	2018	Non Fiction
3	1984 (Signet Classics)	George Orwell	4.7	21424	6	2017	Fiction
4	5,000 Awesome Facts (About Everything!) (Natio...	National Geographic Kids	4.8	7665	12	2019	Non Fiction

In [4]:

df.shape

df.shape

Out[4]:

(550, 7)

In [5]:

df['Name'].value_counts()
df[df['Name'] == "Oh, the Places You'll Go!"]

df['Name'].value_counts() df[df['Name'] == "Oh, the Places You'll Go!"]

Out[5]:

	Name	Author	User Rating	Reviews	Price	Year	Genre
245	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2012	Fiction
246	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2013	Fiction
247	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2014	Fiction
248	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2015	Fiction
249	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2016	Fiction
250	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2017	Fiction
251	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2018	Fiction
252	Oh, the Places You'll Go!	Dr. Seuss	4.9	21834	8	2019	Fiction

In [6]:

df.drop_duplicates(subset=['Name', 'Author'], inplace=True)
df.shape

df.drop_duplicates(subset=['Name', 'Author'], inplace=True) df.shape

Out[6]:

(351, 7)

3 | Разведовательный Анализ¶

▎Колонки для ранжирования¶

У нас в датасете две основные колонки которые мы можем использлвать для ранжирования товаров; это explicit и implicit feedback; количество ревьюеров Reviews и их совокупная оценка пользователей User Rating; эти колонки будут нашими таргетами при обучении моделей

In [7]:

# ratings distribution
plt.title('Rating Distribution')
sns.histplot(x = 'User Rating', data = df);

# ratings distribution plt.title('Rating Distribution') sns.histplot(x = 'User Rating', data = df);

In [8]:

# ratings distribution
plt.title('Number of Reviews')
sns.histplot(x = 'Reviews', data = df);

# ratings distribution plt.title('Number of Reviews') sns.histplot(x = 'Reviews', data = df);

In [9]:

df['Name Length'] = df['Name'].apply(lambda name: len(name))

df['Name Length'] = df['Name'].apply(lambda name: len(name))

In [10]:

train = df[df['Year'] < 2019]
test = df[df['Year'] == 2019]

train = df[df['Year'] < 2019] test = df[df['Year'] == 2019]

In [11]:

train.head()

train.head()

Out[11]:

	Name	Author	User Rating	Reviews	Price	Year	Genre	Name Length
0	10-Day Green Smoothie Cleanse	JJ Smith	4.7	17350	8	2016	Non Fiction	29
1	11/22/63: A Novel	Stephen King	4.6	2052	22	2011	Fiction	17
2	12 Rules for Life: An Antidote to Chaos	Jordan B. Peterson	4.7	18979	15	2018	Non Fiction	39
3	1984 (Signet Classics)	George Orwell	4.7	21424	6	2017	Fiction	22
5	A Dance with Dragons (A Song of Ice and Fire)	George R. R. Martin	4.4	12643	11	2011	Fiction	45

4 | Ранжирования Товаров¶

▎Модель Предсказания Совокупного Рейтинга Фильма¶

In [12]:

# Фичи и целевая переменная 
features = ['Name','Author','Price','Genre','Name Length']
target = 'User Rating'

# Для catboost можно указать категориальные фичи и текстовые 
categorical_features = ['Author', 'Genre'] 
text_features = ['Name']

# Фичи и целевая переменная features = ['Name','Author','Price','Genre','Name Length'] target = 'User Rating' # Для catboost можно указать категориальные фичи и текстовые categorical_features = ['Author', 'Genre'] text_features = ['Name']

In [13]:

X_train = train[features]
y_train = train[target]
X_test = test[features]
y_test = test[target]

X_train = train[features] y_train = train[target] X_test = test[features] y_test = test[target]

In [14]:

model = CatBoostRegressor(loss_function='RMSE', 
                          verbose=False)
model.fit(X_train, 
          y_train,
          categorical_features,
          text_features)

preds = model.predict(X_test)
print(f'Mean Absolute Percentage Error (MAPE): {round(mape(y_test, preds) * 100, 2)}')

model = CatBoostRegressor(loss_function='RMSE', verbose=False) model.fit(X_train, y_train, categorical_features, text_features) preds = model.predict(X_test) print(f'Mean Absolute Percentage Error (MAPE): {round(mape(y_test, preds) * 100, 2)}')

Mean Absolute Percentage Error (MAPE): 3.21

In [15]:

rating_pred = test.copy()
rating_pred['Pred Rating'] = preds
rating_pred.sort_values(by='Pred Rating', ascending=False, inplace=True)
rating_pred.head()

rating_pred = test.copy() rating_pred['Pred Rating'] = preds rating_pred.sort_values(by='Pred Rating', ascending=False, inplace=True) rating_pred.head()

Out[15]:

	Name	Author	User Rating	Reviews	Price	Year	Genre	Name Length	Pred Rating
86	Dog Man: For Whom the Ball Rolls: From the Cre...	Dav Pilkey	4.9	9089	8	2019	Fiction	85	4.872677
85	Dog Man: Fetch-22: From the Creator of Captain...	Dav Pilkey	4.9	12619	8	2019	Fiction	70	4.865559
545	Wrecking Ball (Diary of a Wimpy Kid Book 14)	Jeff Kinney	4.9	9413	8	2019	Fiction	44	4.785464
155	Harry Potter and the Goblet of Fire: The Illus...	J. K. Rowling	4.9	7758	18	2019	Fiction	87	4.771625
150	Guts	Raina Telgemeier	4.8	5476	7	2019	Non Fiction	4	4.733733

In [16]:

model.feature_importances_

model.feature_importances_

Out[16]:

array([44.15442025, 17.23367217, 18.09923482,  2.28755613, 18.22511664])

In [17]:

imp_catboost = pd.DataFrame(X_train.columns, 
                            columns = ['feature'])
imp_catboost['importance'] = model.feature_importances_

sns.set_theme(rc={'figure.figsize':(5,3)},style='whitegrid')
ax = sns.barplot(data=imp_catboost.sort_values(by='importance', ascending=False), 
            x='importance', y='feature', palette="BuGn_r")
plt.title('Top feature importances');

imp_catboost = pd.DataFrame(X_train.columns, columns = ['feature']) imp_catboost['importance'] = model.feature_importances_ sns.set_theme(rc={'figure.figsize':(5,3)},style='whitegrid') ax = sns.barplot(data=imp_catboost.sort_values(by='importance', ascending=False), x='importance', y='feature', palette="BuGn_r") plt.title('Top feature importances');

▎Модель предсказания количества рейтингов¶

Для второй модели попробуем предсказывать Reviews

In [18]:

target = 'Reviews'

target = 'Reviews'

In [19]:

X_train = train[features]
y_train = train[target]
X_test = test[features]
y_test = test[target]

X_train = train[features] y_train = train[target] X_test = test[features] y_test = test[target]

In [20]:

model = CatBoostRegressor(loss_function='RMSE', 
                          verbose=False)
model.fit(X_train, 
          y_train,
          categorical_features,
          text_features)

preds = model.predict(X_test)
print(f'Mean Absolute Percentage Error (MAPE): {round(mape(y_test, preds) * 100, 2)}')

model = CatBoostRegressor(loss_function='RMSE', verbose=False) model.fit(X_train, y_train, categorical_features, text_features) preds = model.predict(X_test) print(f'Mean Absolute Percentage Error (MAPE): {round(mape(y_test, preds) * 100, 2)}')

Mean Absolute Percentage Error (MAPE): 58.13

In [21]:

reviews_pred = test.copy()
reviews_pred['Pred Reviews'] = preds
reviews_pred.sort_values(by='Pred Reviews', ascending=False, inplace=True)
reviews_pred.head()

reviews_pred = test.copy() reviews_pred['Pred Reviews'] = preds reviews_pred.sort_values(by='Pred Reviews', ascending=False, inplace=True) reviews_pred.head()

Out[21]:

	Name	Author	User Rating	Reviews	Price	Year	Genre	Name Length	Pred Reviews
150	Guts	Raina Telgemeier	4.8	5476	7	2019	Non Fiction	4	36632.891881
462	The Silent Patient	Alex Michaelides	4.5	27536	14	2019	Fiction	18	15312.348405
534	Where the Crawdads Sing	Delia Owens	4.8	87841	15	2019	Fiction	23	14982.666031
176	I Am Confident, Brave & Beautiful: A Coloring ...	Hopscotch Girls	4.8	9737	7	2019	Non Fiction	60	14977.741160
345	The Body Keeps the Score: Brain, Mind, and Bod...	Bessel van der Kolk M.D.	4.8	12361	12	2019	Non Fiction	72	11709.129867

In [22]:

pd.DataFrame({'rating': rating_pred['Name'].tolist(),
              'review': reviews_pred['Name'].tolist(),
})

pd.DataFrame({'rating': rating_pred['Name'].tolist(), 'review': reviews_pred['Name'].tolist(), })

Out[22]:

	rating	review
0	Dog Man: For Whom the Ball Rolls: From the Cre...	Guts
1	Dog Man: Fetch-22: From the Creator of Captain...	The Silent Patient
2	Wrecking Ball (Diary of a Wimpy Kid Book 14)	Where the Crawdads Sing
3	Harry Potter and the Goblet of Fire: The Illus...	I Am Confident, Brave & Beautiful: A Coloring ...
4	Guts	The Body Keeps the Score: Brain, Mind, and Bod...
5	It's Not Supposed to Be This Way: Finding Unex...	Strange Planet (Strange Planet Series)
6	The Mueller Report	What Should Danny Do? (The Power to Choose Ser...
7	Difficult Riddles For Smart Kids: 300 Difficul...	The Guardians: A Novel
8	Where the Crawdads Sing	Harry Potter and the Goblet of Fire: The Illus...
9	National Geographic Kids Why?: Over 1,111 Answ...	Unfreedom of the Press
10	The Body Keeps the Score: Brain, Mind, and Bod...	Girl, Stop Apologizing: A Shame-Free Plan for ...
11	Can't Hurt Me: Master Your Mind and Defy the Odds	Good Days Start With Gratitude: A 52 Week Guid...
12	Howard Stern Comes Again	Instant Pot Pressure Cooker Cookbook: 500 Ever...
13	I Am Confident, Brave & Beautiful: A Coloring ...	The Mueller Report
14	Good Days Start With Gratitude: A 52 Week Guid...	The Total Money Makeover: Classic Edition: A P...
15	Strange Planet (Strange Planet Series)	Howard Stern Comes Again
16	The Silent Patient	National Geographic Little Kids First Big Book...
17	5,000 Awesome Facts (About Everything!) (Natio...	5,000 Awesome Facts (About Everything!) (Natio...
18	Instant Pot Pressure Cooker Cookbook: 500 Ever...	The Unofficial Harry Potter Cookbook: From Cau...
19	Unicorn Coloring Book: For Kids Ages 4-8 (US E...	Difficult Riddles For Smart Kids: 300 Difficul...
20	The Total Money Makeover: Classic Edition: A P...	It's Not Supposed to Be This Way: Finding Unex...
21	The Unofficial Harry Potter Cookbook: From Cau...	National Geographic Kids Why?: Over 1,111 Answ...
22	National Geographic Little Kids First Big Book...	Dog Man: Fetch-22: From the Creator of Captain...
23	What Should Danny Do? (The Power to Choose Ser...	Dog Man: For Whom the Ball Rolls: From the Cre...
24	Girl, Stop Apologizing: A Shame-Free Plan for ...	Can't Hurt Me: Master Your Mind and Defy the Odds
25	Unfreedom of the Press	Unicorn Coloring Book: For Kids Ages 4-8 (US E...
26	The Guardians: A Novel	Wrecking Ball (Diary of a Wimpy Kid Book 14)