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 Data Science

4 Pandas Concepts That Quietly Break Your Data Pipelines

Master data types, index alignment, and defensive Pandas practices to prevent silent bugs in real data pipelines. Ibrahim Salami March 23, 20269 min read Generated with Nano Banana When I first started using Pandas, I thought I was doing pretty well. I could clean datasets, run groupby, merge tables, and build quick analyses in a Jupyter notebook. Most tutorials made it feel straightforward: load data, transform it, visualize it, and you’re done. And to be fair, my code usually worked. Until it didn’t. At some point, I started running into strange issues that were hard to explain. Numbers didn’t add up the way I expected. A column that looked numeric behaved like text. Sometimes a transformation ran without errors but produced results that were clearly wrong. The frustrating part was that Pandas rarely complained.
There were no obvious exceptions or crashes. The code executed just fine — it simply produced incorrect results. That’s when I realized something important: most Pandas tutorials focus on what you can do , but they rarely explain how Pandas actually behaves under the hood. Things like:

These concepts don’t feel exciting when you’re first learning Pandas. They’re not as flashy as groupby tricks or fancy visualizations.
But they’re exactly the things that prevent silent bugs in real-world data pipelines. In this article, I’ll walk through four Pandas concepts that most tutorials skip — the same ones that kept causing subtle bugs in my own code. If you understand these ideas, your Pandas workflows become far more reliable, especially when your analysis starts turning into production data pipelines instead of one-off notebooks.
Let’s start with one of the most common sources of trouble: data types.

A Small Dataset (and a Subtle Bug)

To make these ideas concrete, let’s work with a small e-commerce dataset. Imagine we’re analyzing orders from an online store. Each row represents an order and includes revenue and discount information. python

import pandas as pd
orders = pd.DataFrame({
"order_id": [1001, 1002, 1003, 1004],
"customer_id": [1, 2, 2, 3],
"revenue": ["120", "250", "80", "300"], # looks numeric
"discount": [None, 10, None, 20]
})
orders

Output: At first glance, everything looks normal. We have revenue values, some discounts, and a few missing entries. Now let’s answer a simple question: What is the total revenue? python

orders["revenue"].sum()

You might expect something like: markup

750

Instead, Pandas returns: text

'12025080300'

This is a perfect example of what I mentioned earlier: Pandas often fails silently. The code runs successfully, but the output isn’t what you expect. The reason is subtle but incredibly important: The revenue column appears to be numeric, but Pandas actually stores it as text. We can confirm this by checking the dataframe’s data types. python

orders.dtypes

This small detail introduces one of the most common sources of bugs in Pandas workflows: data types. Let’s fix that next.

1. Data Types: The Hidden Source of Many Pandas Bugs

The issue we just saw comes down to something simple: data types.
Even though the revenue column looks numeric, Pandas interpreted it as an object (essentially text).
We can confirm that: python

orders.dtypes

Output: markup

order_id int64 customer_id int64 revenue object discount float64 dtype: object

Because revenue is stored as text, operations behave differently. When we asked Pandas to sum the column earlier, it concatenated strings instead of adding numbers:

This kind of issue shows up surprisingly often when working with real datasets. Data exported from spreadsheets, CSV files, or APIs frequently stores numbers as text. The safest approach is to explicitly define data types instead of relying on Pandas’ guesses. We can fix the column using astype(): python

orders["revenue"] = orders["revenue"].astype(int)

Now if we check the types again: python

orders.dtypes

We get: markup

order_id int64 customer_id int64 revenue int64 discount float64 dtype: object

And the calculation finally behaves as expected: python

orders["revenue"].sum()

Output: markup

750

A Simple Defensive Habit

Whenever I load a new dataset now, one of the first things I run is:
orders.info() It gives a quick overview of:

This simple step often reveals subtle issues before they turn into confusing bugs later. But data types are only one part of the story. Another Pandas behavior causes even more confusion — especially when combining datasets or performing calculations.
It’s something called index alignment.

Index Alignment: Pandas Matches Labels, Not Rows

One of the most powerful — and confusing — behaviors in Pandas is index alignment. When Pandas performs operations between objects (like Series or DataFrames), it does not match rows by position. Instead, it matches them by index labels. At first, this seems subtle. But it can easily produce results that look correct at a glance while actually being wrong. Let’s see a simple example. python

revenue = pd.Series([120, 250, 80], index=[0, 1, 2])
discount = pd.Series([10, 20, 5], index=[1, 2, 3])
revenue + discount

The result looks like this: markup

0 NaN1 2602 1003 NaNdtype: float64

At first glance, this might feel strange. Why did Pandas produce four rows instead of three? The reason is that Pandas aligned the values based on index labels.
Pandas aligns values using their index labels. Internally, the calculation looks like this:

Which produces: markup

0 NaN1 2602 1003 NaNdtype: float64

Rows without matching indices produce missing values,basically.
This behavior is actually one of Pandas’ strengths because it allows datasets with different structures to combine intelligently. But it can also introduce subtle bugs.

How This Shows Up in Real Analysis

Let’s go back to our orders dataset. Suppose we filter orders with discounts: python

discounted_orders = orders[orders["discount"].notna()]

Now imagine we try to calculate net revenue by subtracting the discount. python

orders["revenue"] - discounted_orders["discount"]

You might expect a straightforward subtraction. Instead, Pandas aligns rows using the original indices. The result will contain missing values because the filtered dataframe no longer has the same index structure. This can easily lead to:

And again — Pandas will not raise an error.

A Defensive Approach

If you want operations to behave row-by-row , a good practice is to reset the index after filtering. python

discounted_orders = orders[orders["discount"].notna()].reset_index(drop=True)

Now the rows are aligned by position again. Another option is to explicitly align objects before performing operations: python

orders.align(discounted_orders)

Or in situations where alignment is unnecessary, you can work with raw arrays: python

orders["revenue"].values

In the end, it all boils down to this. In Pandas, operations align by index labels , not row order. Understanding this behavior helps explain many mysterious NaN values that appear during analysis. But there’s another Pandas behavior that has confused almost every data analyst at some point. You’ve probably seen it before:
**SettingWithCopyWarning** Let’s unpack what’s actually happening there. Great — let’s continue with the next section.

The Copy vs View Problem (and the Famous Warning)

If you’ve used Pandas for a while, you’ve probably seen this warning before: SettingWithCopyWarning When I first encountered it, I mostly ignored it. The code still ran, and the output looked fine, so it didn’t seem like a big deal. But this warning points to something important about how Pandas works: sometimes you’re modifying the original dataframe , and sometimes you’re modifying a temporary copy. The tricky part is that Pandas doesn’t always make this obvious. Let’s look at an example using our orders dataset. Suppose we want to adjust revenue for orders where a discount exists. A natural approach might look like this: python

discounted_orders = orders[orders["discount"].notna()]
discounted_orders["revenue"] = discounted_orders["revenue"] - discounted_orders["discount"]

This often triggers the warning: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame
The problem is that discounted_orders may not be an independent dataframe. It might just be a view into the original orders dataframe. So when we modify it, Pandas isn’t always sure whether we intend to modify the original data or modify the filtered subset. This ambiguity is what produces the warning. Even worse, the modification might not behave consistently depending on how the dataframe was created. In some situations, the change affects the original dataframe; in others, it doesn’t. This kind of unpredictable behavior is exactly the sort of thing that causes subtle bugs in real data workflows.

The Safer Way: Use .loc

A more reliable approach is to modify the dataframe explicitly using.loc. python

orders.loc[orders["discount"].notna(), "revenue"] = (
orders["revenue"] - orders["discount"]
)

This syntax clearly tells Pandas which rows to modify and which column to update. Because the operation is explicit, Pandas can safely apply the change without ambiguity.

Another Good Habit: Use .copy()

Sometimes you really do want to work with a separate dataframe. In that case, it’s best to create an explicit copy. python

discounted_orders = orders[orders["discount"].notna()].copy()

Now discounted_orders is a completely independent object, and modifying it won’t affect the original dataset.

So far we’ve seen how three behaviors can quietly cause problems:

But there’s one more habit that can dramatically improve the reliability of your data workflows. It’s something many data analysts rarely think about: defensive data manipulation.

Defensive Data Manipulation: Writing Pandas Code That Fails Loudly

One thing I’ve slowly realized while working with data is that most problems don’t come from code crashing. They come from code that runs successfully but produces the wrong numbers. And in Pandas, this happens surprisingly often because the library is designed to be flexible. It rarely stops you from doing something questionable. That’s why many data engineers and experienced analysts rely on something called defensive data manipulation. Here's the idea. Instead of assuming your data is correct, you actively validate your assumptions as you work. This helps catch issues early before they quietly propagate through your analysis or pipeline. Let’s look at a few practical examples.

Validate Your Data Types

Earlier we saw how the revenue column looked numeric but was actually stored as text. One way to prevent this from slipping through is to explicitly check your assumptions. For example: python

assert orders["revenue"].dtype == "int64"

If the dtype is incorrect, the code will immediately raise an error.
This is much better than discovering the problem later when your metrics don’t add up. Prevent Dangerous Merges Another common source of silent errors is merging datasets. Imagine we add a small customer dataset: python

customers = pd.DataFrame({
"customer_id": [1, 2, 3],
"city": ["Lagos", "Abuja", "Ibadan"]
})

A typical merge might look like this: orders.merge(customers, on="customer_id") This works fine, but there’s a hidden risk. If the keys aren’t unique, the merge could accidentally create duplicate rows , which inflates metrics like revenue totals. Pandas provides a very useful safeguard for this: python

orders.merge(customers, on="customer_id", validate="many_to_one")

Now Pandas will raise an error if the relationship between the datasets isn’t what you expect. This small parameter can prevent some very painful debugging later.

Check for Missing Data Early

Missing values can also cause unexpected behavior in calculations.
A quick diagnostic check can help reveal issues immediately: python

orders.isna().sum()

This shows how many missing values exist in each column.
When datasets are large, these small checks can quickly surface problems that might otherwise go unnoticed.

A Simple Defensive Workflow

Over time, I’ve started following a small routine whenever I work with a new dataset:

These steps only take a few seconds, but they dramatically reduce the chances of introducing silent bugs.

Final Thoughts

When I first started learning Pandas, most tutorials focused on powerful operations like groupby, merge, or pivot_table. Those tools are important, but I’ve come to realize that reliable data work depends just as much on understanding how Pandas behaves under the hood. Concepts like:

may not feel exciting at first, but they’re exactly the things that keep data workflows stable and trustworthy. The biggest mistakes in data analysis rarely come from code that crashes. They come from code that runs perfectly — while quietly producing the wrong results. And understanding these Pandas fundamentals is one of the best ways to prevent that. Thanks for reading! If you found this article helpful, feel free to let me know. I truly appreciate your feedback Medium LinkedIn Twitter YouTube

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

Ibrahim Salami

See all from Ibrahim Salami Data AnalysisData TypePandasProgrammingPython 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

Using Classical ML to Empower AI Agents

Machine Learning On the value of building on existing foundations Stephanie Kirmer July 17, 20268 min read

Creating Charts in Google Slides with Python

Data Science A guide to easily creating elegantly formatted charts & slides using Google APIs Mikey Gracie August 21, 20216 min read

Human vs. Machine: Comparing My Performance in Wordle Against an Algorithm

Data Science A comparison based on statistics and linear programming Nicolas Garcia Aramouni April 19, 202210 min read

Classical classifier combination techniques: Voting approaches, Borda Counts and...

Artificial Intelligence In this article, we review classical classifier combination techniques that are usually used as benchmarks against a newly proposed method. Mahmoud Albardan November 1, 20206 min read

Bias Variance Trade-off

Artificial Intelligence Learn about bias, variance and total error relationship to model complexity and overfitting/underfitting. Manpreet Singh Minhas January 16, 20217 min read

Can Algorithms be Racist?

Artificial Intelligence A brief exploration of juristic and social dilemmas within language models Jillian Chambers December 22, 20215 min read

Understanding and Choosing the Right Probability Distributions with Examples

Data Science The Most Common Discrete Probability Distributions Explained with Examples Kessie Zhang September 14, 20206 min read

Ridge Regression: A Robust Path to Reliable Predictions

Machine Learning Learn how regularization reduces overfitting and improves model stability in linear regression. Niklas Lang January 30, 202510 min read

Best Practices for Writing Good SQL Code

Programming Introduction Terence Shin, MSc, MBA April 14, 20226 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