保留正文与补采附件。源页面当前未重新核验；具体限制见交付清单。

Revisit consent button We value your privacy We use cookies to enhance your browsing experience, serve personalised ads or content, and analyse our traffic. By clicking "Accept All", you consent to our use of cookies. Customise Reject All Accept All Customise Consent Preferences We use cookies to help you navigate efficiently and perform certain functions. You will find detailed information about all cookies under each consent category below. The cookies that are categorised as "Necessary" are stored on your browser as they are essential for enabling the basic functionalities of the site. ... Show more NecessaryAlways Active Necessary cookies are required to enable the basic features of this site, such as providing secure log-in or adjusting your consent preferences. These cookies do not store any personally identifiable data.

Functional Functional cookies help perform certain functionalities like sharing the content of the website on social media platforms, collecting feedback, and other third-party features.

Analytics Analytical cookies are used to understand how visitors interact with the website. These cookies help provide information on metrics such as the number of visitors, bounce rate, traffic source, etc.

Performance Performance cookies are used to understand and analyse the key performance indexes of the website which helps in delivering a better user experience for the visitors.

Advertisement Advertisement cookies are used to provide visitors with customised advertisements based on the pages you visited previously and to analyse the effectiveness of the ad campaigns.

Uncategorised Other uncategorised cookies are those that are being analysed and have not been classified into a category as yet. No cookies to display. Reject All Save My Preferences Accept All Toggle SearchToggle Mobile Navigation Publish AI, ML & data-science insights to a global community of data professionals. Sign inSubmit an Article LatestEditor's PicksNewsletterShipAI

Write For TDS LinkedInXToggle Search Programming

Simplifying the Python Code for Data Engineering Projects

Python tricks and techniques for data ingestion, validation, processing, and testing: a practical walkthrough John Leung June 12, 202411 min read Raw data comes from various sources and formats. Before the data can be available to answer critical business questions, substantial effort and time are required to perform data engineering. While the underlying data infrastructure can vary based on the data volume, velocity, and analytics requirements, some fundamental code design techniques are still relevant to simplify and streamline various tasks across time. This article will explore different critical parts of general data engineering projects, from data ingestion to pipeline testing. Python is the most widely used programming language for data engineering, and we will learn how to deal with these use cases using built-in functionalities and efficient libraries in Python. Photo by Katerina Pavlyuchkova on Unsplash Photo by Katerina Pavlyuchkova on Unsplash

Imagine you have an online retail shop that sells unique, all-occasion gifts. The online shop is so popular that it has a high volume of transactions every minute and second. You have the ambition to satisfy more current customers' needs and serve more new customers through analyzing buying habits about the current transactions, so this motivates you to dive into the data processing of the transaction records as the preparation. Learn this step by step with the interactive Computer Science and Data Engineer roadmaps.

0 Mock data

We first mock some transaction data into a file using the JSON Lines (JSONL) text format, where each line is a separate JSON object. This format is appealing for data streaming in areas, such as web/ app analytics and log management. In our file, the data fields belong to various data types. They include the customer and product identifiers (in integer/ array format), the payment method (in string format), and the total transaction amount (in float number). python

import json
import random
import numpy as np
import datetime

# Remove existing 'retail_transactions.jsonl' file, if any
! rm -f /p/a/t/h retail_transactions.jsonl

# Set the no of transactions
no_of_iteration = 500000

# Open a file in write mode
with open('retail_transactions.jsonl', 'w') as f:
  for num in range(no_of_iteration):
    if (random.randint(1, 10000) != 5000):
      # Create a valid transaction
      new_txn = {
        'orderID': num,
        'customerID': random.randint(1, 100000),
        'productID': np.random.randint(10000, size=random.randint(1, 5)).tolist(),
        'paymentMthd': random.choice(['Credit card', 'Debit card', 'Digital wallet', 'Cash on delivery', 'Cryptocurrency']),
        'totalAmt': round(random.random() * 5000, 2),
        'invoiceTime': datetime.datetime.now().isoformat()
      }
    else:
      # Create an invalid transaction
      new_txn = {
        'orderID': "",
        'customerID': "",
        'productID': "",
        'paymentMthd': "",
        'totalAmt': "",
        'invoiceTime': ""
       }

     # Write the transaciton as a JSON line to the file
     f.write(json.dumps(new_txn) + "n")

You may discover several single transactions with blank data fields written to the file. This mimics the missing data issue, as one of the data quality issues often encountered in the real world.

1 Data ingestion - Yield

To read the transaction records from the file, one of the simplest approaches is to loop through the dataset into a list and then convert it into a Pandas DataFrame. This method will work like a charm for the 500,000 transactions in our demo dataset. But what if the real-world datasets range from millions to even billions of rows? We may sit still for long to wait until the whole computation completes, if not leading to memory issues. Sometimes we don't care about the entire results and want to process the initial results instead before the last record is loaded. In such cases, we can alternatively use yield to control the flow of a generator.

The generator does not store the entire records in memory. Instead, it gives a value one at a time and pauses the function execution until the next value is requested. There are routines of interleaving user codes with library codes. It also enforces sequencing, which means you cannot access the second record before reaching the first. You can learn more about this concept in the Pydata talk video, which provides a detailed explanation. The yield statement has different practical usages. For example, we can go through each line of the file and yield only the non-blank records. Below shows how we can execute real-time data filtering: python

import json

def read_json_file(file_name):
  # Read the JSONL file
  with open(file_name) as f:
    for line in f:
      txn = json.loads(line)
      # Yield valid transactions only
      if (txn['orderID'] != ""):
        yield(txn)

txn_generator = read_json_file('retail_transactions.jsonl')

The output of these codes gives a Python generator, a special type of iterator. You can use the next function in a loop to return the subsequent items one by one. Apart from real-time data filtering, another idea is to design a generator function that pre-processes data and yields it in a pre-defined batch size, which can be parsed straightforwardly to feed a machine-learning model for training. And more, we can use it to asynchronously handle web requests and responses, when crawling web pages.

2 Data validation - Pydantic

Assume you have a list of JSON data that covers the information of transaction records after data ingestion. Here is a sample transaction: json

{
 'orderID': 10000,
 'customerID': 48316,
 'productID': [5620],
 'paymentMthd': 'Cash on delivery',
 'totalAmt': 9301.2,
 'invoiceTime': '2024-06-10T23:30:29.608443',
 'price': -1
}

For each incoming data, we want to ensure it is being validated, otherwise, we will easily hit different types of errors when running the subsequent data processing functions. This can be achieved using pydantic library.

We first define the schema of our data fields using the Pydantic model, then validate our JSON data using model_validate() function. python

from datetime import datetime
from pydantic import BaseModel, ValidationError

# Define the data model for a transaction record
class TxnModel(BaseModel):
  orderID: int
  customerID: int
  productID: list[int]
  paymentMthd: str
  totalAmt: float
  invoiceTime: datetime

try:
  # Validate the sample case against the schema
  TxnModel.model_validate(sample_txn)
  print("Validated successfully!")
except ValidationError as exc:
  # Print error messages for any validation error
  print("Validation Error:")
  print(exc.errors())

# Output:
# Validated successfully

Sometimes, we find the necessity to apply stricter validation rules. For example, the Pydantic base model attempts to coerce string data to an integer if possible. To avoid this, you can set strict=True at the model level or field level. Besides, we can apply custom validation rules to data fields. For example, we may want to check if the payment method value is within our expectations. To facilitate testing, we manually set the payment method of the sample case to 'Bitcoin', which is a non-existent option in the online shop, and then use AfterValidator to embed a function for further checking. python

from typing import Annotated
from pydantic.functional_validators import AfterValidator

# Customize the validation rule
def validate_payment_mthd(paymentMthd: str):
  possible_values = ['Credit card', 'Debit card', 'Digital wallet', 'Cash on delivery', 'Cryptocurrency']
  if paymentMthd not in possible_values:
    raise ValueError(f"Invalid paymentMthd, payment type must be one of {possible_values}")
  return storage

# Define the data model for a transaction record
class TxnModel(BaseModel):
  orderID: int = Field(strict=True)
  customerID: int
  productID: list[int]
  paymentMthd: Annotated[str, AfterValidator(validate_payment_mthd)]
  totalAmt: Annotated[float, Field(strict=True, gt=0)]
  invoiceTime: datetime

# Manually define a non-existent payment method
sample_txn['paymentMthd'] = 'Bitcoin'

try:
  # Validate the sample case against the schema
  TxnModel.model_validate(sample_txn)
  print("Validated successfully!")
except ValidationError as exc:
  # Print error messages for any validation error
  print("Validation Error:")
  print(exc.errors()[0]['ctx'])

# Output
# Validation Error:
# {'error': ValueError("Invalid paymentMthd, payment type must be one of ['Credit card', 'Debit card', 'Digital wallet', 'Cash on delivery', 'Cryptocurrency']")}

The validator successfully identifies that the payment method is not within the list of possible values. This is done by applying Pydantic's inner validation logic, followed by custom validation functions. The code raises a ValueError, which populates ValidationError. When the error is triggered, we can have follow-up actions for rectification. These features help eliminate data errors, thus ensuring the accuracy and completeness of our data.

3 Data processing

(1) Python decorator After data validation, we start working with data-intensive functions. There is a high chance of encountering lengthy execution times as the data pipeline becomes complex. We wish to identify the root cause and optimize the time performance of functions. One simple method is to collect two timestamps at the beginning and end of every function, and then calculate the time differences one by one. To ensure the code is less cluttered throughout the data pipeline, we can leverage the Python decorator.

We first design a Python decorator that measures the execution time. Afterward, we annotate any function that requires this feature. For example, you can measure the time taken to categorize prices for all transactions. python

import time

# Measure the excution time of a given function
def time_decorator(func):
  def wrapper(*args, **kwargs):
    begin_time = time.time()
    output = func(*args, **kwargs)
    end_time = time.time()
    print(f"Execution time of function {func.__name__}: {round(end_time - begin_time, 2)} seconds.")
    return output
  return wrapper

# Categorize the total amount of each transaction
@time_decorator
def group_txn_price(data):
  for txn in data:
    price = txn['totalAmt']
    if 0 <= price <= 1500:
      txn['totalAmtCat'] = 'Low'
    elif 1500 < price <= 3500:
      txn['totalAmtCat'] = 'Moderate'
    elif 3500 < price:
      txn['totalAmtCat'] = 'High'
    return data

txn_list = group_txn_price(txn_list)

# Output
# Execution time of function group_txn_price: 0.26 seconds.

The decorator approach makes the code reusable without changing the source code of our original functions. Similarly, we can apply decorator ideas for logging function completion or email alerting when jobs encounter failure. (2) Map, reduce, filter These are commonly used Python array methods that many developers may be familiar with. But I still think they are worth mentioning due to several reasons: (1) immutable - the functions do not modify values of the original lists; (2) the chain flexibility - can apply a combination of functions simultaneously; and (3) concise and readable - with only one line of code. Assume we have a list of JSON objects with only two keys: payment method and total amount. Let's explore some examples of how these functions work. Map: Perform the same operation on all elements in the list (e.g. adding a suffix to the values of the payment method). python

updated_txn_list = list(map(lambda x: {
                      'paymentMthd': f"{x['paymentMthd']}_2024",
                      "totalAmt": x["totalAmt"]
                   }, txn_list))

print(updated_txn_list)

# Output
# [{'paymentMthd': 'Cryptocurrency_2024', 'totalAmt': 3339.85},
# {'paymentMthd': 'Cash on delivery_2024', 'totalAmt': 872.52},
# ...]

Filter : Obtain a subset of elements that meet a certain condition(s) (e.g. only records with cryptocurrency as the payment method). python

updated_txn_list = list(map(lambda x: x, filter(lambda y: y["paymentMthd"] == "Cryptocurrency", txn_list)))

print(updated_txn_list)

# Output
# [{'paymentMthd': 'Cryptocurrency', 'totalAmt': 3339.85},
# {'paymentMthd': 'Cryptocurrency', 'totalAmt': 576.15},
# ...]

Reduce : Get a single-valued outcome (e.g. summing or multiplying all elements). python

from functools import reduce

total_amt_crypto = reduce(lambda acc, x: acc + x["totalAmt"], updated_txn_list, 0)

print(total_amt_crypto)

# Output
# 250353984.67000002

We can leverage these functions during the transformation steps in data science projects. For example, use map() to scale or normalize the data, use filter() to remove outliers and irrelevant data points, and reduce() to generate summary statistics.

4 Data pipeline testing - Pytest

Data pipelines often involve data ingestion, data cleansing, and Extract-Transform-Load (ETL) operations. The scope of potential errors can be broad and easily overlooked, especially as the model flow and result are hard for users to interpret. This leads to a heavier reliance on testing efforts by the development team.

It is common to conduct unit testing to ascertain each component of the machine learning system performs as expected. One of the most popular Python testing frameworks is [Pytest](https://docs.pytest.org/en/stable/contents.html). Imagine we want to ensure the high-quality of transformed data that both technical teams and decision-makers can trust. We can test for the function that we have gone through about categorizing transaction prices. To achieve this, we need to prepare two Python files:

python

# Categorize the total amount of each transaction
def add_features(sample_cases):
  for txn in sample_cases:
    price = txn['totalAmt']
  if 0 <= price <= 1500:
    txn['totalAmtCat'] = 'Low'
  elif 1500 < price <= 3500:
    txn['totalAmtCat'] = 'Moderate'
  elif 3500 < price:
    txn['totalAmtCat'] = 'High'

return sample_cases

python

from feature_engineering import add_features

def test_add_features():
  sample_cases = [{
      'orderID': 1,
      'customerID': 36536,
      'productID': [2209, 2262, 4912, 3162, 5734],
      'paymentMthd': 'Cryptocurrency',
      'totalAmt': 576.15,
      'invoiceTime': '2024–06–10T23:53:25.329928'
    }]

  # Call the function with the sample cases
  sample_cases = add_features(sample_cases)

  # Check the assertations
  for txn in sample_cases: 
    assert 'totalAmtCat' in list(txn.keys())
    assert len(txn) == 7
    assert len(txn['totalAmtCat']) != 0

The assert statements above ensure that the new 'totalAmtCat' data field is added with a non-blank value, and the original data fields are not affected. By executing the command Pytest, we can know that our test has passed! Pytest result - Test passed Image by author Pytest result - Test passed (Image by author) For a more advanced case, let's say we have three functions with the following sequences: load_data, clean_data, and add_features. How should we design the test file to validate the output of these functions one by one? python

import pytest
import json
from feature_engineering import load_data, clean_data, add_features

# Set up a temporary JSONL file
@pytest.fixture
def jsonl_file(tmp_path):
  sample_cases = [{'orderID': 10000,
    'customerID': 48316,
    'productID': [5620],
    'paymentMthd': 'Cash on delivery',
    'totalAmt': 9301.2,
    'invoiceTime': '2024-06-10T23:30:29.608443',
    'price': -1
  }]

  file_path = tmp_path + "/test_transactions.jsonl"

  with open(file_path, 'w') as f:
    for txn in sample_cases:
        f.write(json.dumps(txn) + "n")

  return file_path

# Test function to validate the `load_data` function
def test_load_data(jsonl_file):
  data = load_data(jsonl_file)
  # assert statements here

# Test function to validate the `clean_data` function
def test_clean_data(jsonl_file):
  data = load_data(jsonl_file)
  data = clean_data(data)
  # assert statements here

# Test function to validate the `add_features` function
def test_add_features(jsonl_file):
  data = load_data(jsonl_file)
  data = clean_data(data)
  data = add_features(data)
  # assert statements here

We should define a fixed baseline for initialization, such as a JSON Lines file with sample test cases. Here we use @pytest.fixturedecorator, which works similarly to the time_decoratorwe discussed earlier in the Python decorator section. This decorator helps prevent repeatedly initializing the sample files. For the remaining codes, we involve several test functions to run the pipeline functions and use assert statements to detect logical errors.

Wrapping it up

We came across several critical aspects of data engineering projects, and explored how to simplify and streamline the Python code for efficiency and readability:

Before you go

If you enjoy this reading, I invite you to **** follow my Medium page and LinkedIn page. By doing so, you can stay updated with exciting content related to data science side projects, Machine Learning Operations (MLOps) demonstrations, and project management methodologies.

Performing Customer Analytics with LangChain and LLMs Feature Engineering for Time-Series Using PySpark on Databricks

Register Now to Continue Reading

Unlock Towards Data Science with a Free Account

All articles on Towards Data Science are available to registered users. Create a free account today to access this article and thousands more. Register for freeSign in to your account

Written By

John Leung

See all from John Leung Big DataData EngineeringProgrammingPythonTips And Tricks Share This Article

Towards Data Science is a community publication. Submit your insights to reach our global audience and earn through the TDS Author Payment Program. Write for TDS

Related Articles

How to Read and Write Data from/to the Quip Spreadsheet using Quip Python APIs

Enable end-users to provide their inputs that can then be used as overrides/additional context in the final analytical solution Isha Garg November 17, 20237 min read

Behind the scenes: Delta Lake Clones

Programming How to clone your delta lake tables easily using Spark. Prashanth Xavier March 14, 20215 min read

Python Pandas Reading a CSV

Programming Learn how to read a CSV file and create a Pandas DataFrame Dean McGrath February 2, 20218 min read

A Practical Guide on Missing Values with Pandas

Artificial Intelligence A comprehensive guide to handle missing Soner Yıldırım June 13, 20205 min read

Practical Guide for Virtual Environments in Python

Artificial Intelligence Using virtualenv and pipenv tools Soner Yıldırım January 3, 20214 min read

A Gentle Introduction to Self-Training and Semi-Supervised Learning

Machine Learning Coding an example of self-training in Python to utilize unlabeled data for classification Doug Steen August 30, 20208 min read

How To Setup HashiCorp Vault with Airflow

Programming Integrating HashiCorp Vault with Apache Airflow Giorgos Myrianthous June 3, 20224 min read

Surviving a Coding Bootcamp

Programming Tips, Tricks, and Advice from a Coding Bootcamp Survivor Jesse Neumann July 22, 20208 min read

Apache Hadoop: A Review on Security Issues and Solutions for HDFS

Security Big Data Solution A deep dive into the security issues that occur in HDFS structure,... KahEm Chu December 4, 202015 min read

Your home for data science and AI. The world's leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals. Subscribe to Our NewsletterWrite For TDS•About•Advertise•Privacy Policy•Terms of Use © Insight Media Group, LLC 2026•The New Stack•roadmap.sh•Creator Network

Recommended Articles

Close