Task 2: Next Product Recommendation for Underrepresented Languages
Task 2 - Getting Started
Make your first submission on Task 2
Amazon KDD Cup 2023 - Task 2 - Next Product Recommendation¶
This notebook will contains instructions and example submission with random predictions.
Installations 🤖¶
aicrowd-cli
for downloading challenge data and making submissionspyarrow
for saving to parquet for submissions
!pip install aicrowd-cli pyarrow
Login to AIcrowd and download the data 📚¶
!aicrowd login
!aicrowd dataset download --challenge task-2-next-product-recommendation-for-underrepresented-languages
Setup data and task information¶
import os
import numpy as np
import pandas as pd
from functools import lru_cache
train_data_dir = '.'
test_data_dir = '.'
task = 'task2'
PREDS_PER_SESSION = 100
# Cache loading of data for multiple calls
@lru_cache(maxsize=1)
def read_product_data():
return pd.read_csv(os.path.join(train_data_dir, 'products_train.csv'))
@lru_cache(maxsize=1)
def read_train_data():
return pd.read_csv(os.path.join(train_data_dir, 'sessions_train.csv'))
@lru_cache(maxsize=3)
def read_test_data(task):
return pd.read_csv(os.path.join(test_data_dir, f'sessions_test_{task}.csv'))
Data Description¶
The Multilingual Shopping Session Dataset is a collection of anonymized customer sessions containing products from six different locales, namely English, German, Japanese, French, Italian, and Spanish. It consists of two main components: user sessions and product attributes. User sessions are a list of products that a user has engaged with in chronological order, while product attributes include various details like product title, price in local currency, brand, color, and description.
Each product as its associated information:¶
locale: the locale code of the product (e.g., DE)
id: a unique for the product. Also known as Amazon Standard Item Number (ASIN) (e.g., B07WSY3MG8)
title: title of the item (e.g., “Japanese Aesthetic Sakura Flowers Vaporwave Soft Grunge Gift T-Shirt”)
price: price of the item in local currency (e.g., 24.99)
brand: item brand name (e.g., “Japanese Aesthetic Flowers & Vaporwave Clothing”)
color: color of the item (e.g., “Black”)
size: size of the item (e.g., “xxl”)
model: model of the item (e.g., “iphone 13”)
material: material of the item (e.g., “cotton”)
author: author of the item (e.g., “J. K. Rowling”)
desc: description about a item’s key features and benefits called out via bullet points (e.g., “Solid colors: 100% Cotton; Heather Grey: 90% Cotton, 10% Polyester; All Other Heathers …”)
EDA 💽¶
def read_locale_data(locale, task):
products = read_product_data().query(f'locale == "{locale}"')
sess_train = read_train_data().query(f'locale == "{locale}"')
sess_test = read_test_data(task).query(f'locale == "{locale}"')
return products, sess_train, sess_test
def show_locale_info(locale, task):
products, sess_train, sess_test = read_locale_data(locale, task)
train_l = sess_train['prev_items'].apply(lambda sess: len(sess))
test_l = sess_test['prev_items'].apply(lambda sess: len(sess))
print(f"Locale: {locale} \n"
f"Number of products: {products['id'].nunique()} \n"
f"Number of train sessions: {len(sess_train)} \n"
f"Train session lengths - "
f"Mean: {train_l.mean():.2f} | Median {train_l.median():.2f} | "
f"Min: {train_l.min():.2f} | Max {train_l.max():.2f} \n"
f"Number of test sessions: {len(sess_test)}"
)
if len(sess_test) > 0:
print(
f"Test session lengths - "
f"Mean: {test_l.mean():.2f} | Median {test_l.median():.2f} | "
f"Min: {test_l.min():.2f} | Max {test_l.max():.2f} \n"
)
print("======================================================================== \n")
products = read_product_data()
locale_names = products['locale'].unique()
for locale in locale_names:
show_locale_info(locale, task)
products.sample(5)
train_sessions = read_train_data()
train_sessions.sample(5)
test_sessions = read_test_data(task)
test_sessions.sample(5)
Generate Submission 🏋️♀️¶
Submission format:
- The submission should be a parquet file with the sessions from all the locales.
- Predicted products ids per locale should only be a valid product id of that locale.
- Predictions should be added in new column named "next_item_prediction".
- Predictions should be a list of string id values
def random_predicitons(locale, sess_test_locale):
random_state = np.random.RandomState(42)
products = read_product_data().query(f'locale == "{locale}"')
predictions = []
for _ in range(len(sess_test_locale)):
predictions.append(
list(products['id'].sample(PREDS_PER_SESSION, replace=True, random_state=random_state))
)
sess_test_locale['next_item_prediction'] = predictions
sess_test_locale.drop('prev_items', inplace=True, axis=1)
return sess_test_locale
test_sessions = read_test_data(task)
predictions = []
test_locale_names = test_sessions['locale'].unique()
for locale in test_locale_names:
sess_test_locale = test_sessions.query(f'locale == "{locale}"').copy()
predictions.append(
random_predicitons(locale, sess_test_locale)
)
predictions = pd.concat(predictions).reset_index(drop=True)
predictions.sample(5)
Validate predictions ✅¶
def check_predictions(predictions, check_products=False):
"""
These tests need to pass as they will also be applied on the evaluator
"""
test_locale_names = test_sessions['locale'].unique()
for locale in test_locale_names:
sess_test = test_sessions.query(f'locale == "{locale}"')
preds_locale = predictions[predictions['locale'] == sess_test['locale'].iloc[0]]
assert sorted(preds_locale.index.values) == sorted(sess_test.index.values), f"Session ids of {locale} doesn't match"
if check_products:
# This check is not done on the evaluator
# but you can run it to verify there is no mixing of products between locales
# Since the ground truth next item will always belong to the same locale
# Warning - This can be slow to run
products = read_product_data().query(f'locale == "{locale}"')
predicted_products = np.unique( np.array(list(preds_locale["next_item_prediction"].values)) )
assert np.all( np.isin(predicted_products, products['id']) ), f"Invalid products in {locale} predictions"
check_predictions(predictions)
# Its important that the parquet file you submit is saved with pyarrow backend
predictions.to_parquet(f'submission_{task}.parquet', engine='pyarrow')
Submit to AIcrowd 🚀¶
# You can submit with aicrowd-cli, or upload manually on the challenge page.
!aicrowd submission create -c task-2-next-product-recommendation-for-underrepresented-languages -f "submission_task2.parquet"
Content
Comments
You must login before you can post a comment.
upload always failed. how should i fix it?