Data Analysis in Python
A DataCamp track teaching pandas, NumPy and data cleaning techniques for analysts moving from spreadsheets or SQL into Python-based.
pandas is the Python library for working with tables of data. Its core object, the DataFrame, is a table with named columns that you can load, filter, group, join and reshape with code. Most day-to-day analysis in pandas uses a small set of about 15 operations. Learn those well and you can handle the majority of real analyst tasks.
pandas has hundreds of functions, and beginners often try to learn them all. That isn't necessary. This guide covers the operations that show up in almost every analysis, in the order you'll usually need them, with the Excel or SQL equivalent where it helps.
The running example is a small orders file with the columns order_id, customer_id, order_date, region, product, quantity and revenue.
import pandas as pd
pd is the universal alias. You'll see it in every tutorial and code sample.
df = pd.read_csv("orders.csv")
# also: pd.read_excel("orders.xlsx", sheet_name="2026")
read_csv and read_excel cover most files. pandas can also read directly from SQL databases, JSON and Parquet.
df.head() # first 5 rows
df.shape # (rows, columns)
df.info() # column names, types, non-null counts
df.describe() # summary stats for numeric columns
Always run these first. info() alone catches most problems: numbers stored as text, dates not recognized, unexpected missing values.
df["revenue"] # one column (a Series)
df[["region", "revenue"]] # several columns (a DataFrame)
Note the double brackets for multiple columns. It's a list of column names inside the selection brackets.
df[df["region"] == "North"]
df[(df["region"] == "North") & (df["revenue"] > 1000)]
df[df["region"].isin(["North", "South"])]
This is the equivalent of Excel's filter or SQL's WHERE. Use & for AND and | for OR, and wrap each condition in parentheses.
.locdf.loc[df["revenue"] > 1000, ["order_id", "revenue"]]
.loc[rows, columns] is the clearest way to filter and pick columns at the same time. It's also the safe way to change values in a subset of rows:
df.loc[df["region"] == "Nth", "region"] = "North"
df.sort_values("revenue", ascending=False)
df.sort_values(["region", "revenue"], ascending=[True, False])
df["unit_price"] = df["revenue"] / df["quantity"]
df["is_large"] = df["revenue"] > 1000
Calculations run on the whole column at once. You don't write a loop, and you don't "drag the formula down" as in Excel.
df["region"].value_counts()
df["region"].value_counts(normalize=True) # as shares
This is one of the most useful quick checks in pandas: it shows the categories, how often each appears, and any messy spellings that need cleaning.
df.groupby("region")["revenue"].sum()
df.groupby("region").agg(
total_revenue=("revenue", "sum"),
orders=("order_id", "count"),
avg_order=("revenue", "mean"),
)
This is the pandas version of a PivotTable or SQL's GROUP BY. The named-aggregation style in the second example gives readable column names in one step.
df.isna().sum() # missing per column
df["region"] = df["region"].fillna("Unknown")
df = df.dropna(subset=["revenue"]) # drop rows missing revenue
Decide column by column. Filling and dropping both change your results, so note what you did.
df["order_date"] = pd.to_datetime(df["order_date"], dayfirst=True)
df["month"] = df["order_date"].dt.to_period("M")
df["weekday"] = df["order_date"].dt.day_name()
Once a column is a real datetime, the .dt accessor gives you year, month, weekday and more. Use dayfirst=True for day/month/year formats common in Vietnam and Europe.
customers = pd.read_csv("customers.csv")
merged = df.merge(customers, on="customer_id", how="left")
how takes "inner", "left", "right" or "outer", with exactly the same meaning as SQL joins. The same traps apply too: duplicate keys multiply rows, and unmatched rows get missing values. See SQL Joins Explained. Check len(df) before and after.
pd.pivot_table(df, index="region", columns="month",
values="revenue", aggfunc="sum", fill_value=0)
This produces a region × month grid, just like an Excel PivotTable.
df.duplicated(subset=["order_id"]).sum()
df = df.drop_duplicates(subset=["order_id"], keep="last")
Check first, then drop, and know why the duplicates exist.
df.to_csv("orders_clean.csv", index=False)
df.to_excel("summary.xlsx", index=False)
index=False stops pandas from writing its row numbers as an extra column, a common beginner annoyance.
"Which region had the highest revenue per order last quarter?"
df["order_date"] = pd.to_datetime(df["order_date"], dayfirst=True)
q = df[df["order_date"].dt.to_period("Q") == "2026Q2"]
summary = q.groupby("region").agg(revenue=("revenue", "sum"), orders=("order_id", "nunique"))
summary["rev_per_order"] = summary["revenue"] / summary["orders"]
summary.sort_values("rev_per_order", ascending=False)
Load, fix types, filter, group, derive, sort: most real analyses follow this same pattern.
| Task | Excel | SQL | pandas |
|---|---|---|---|
| Filter | AutoFilter | WHERE |
df[condition] |
| Group totals | PivotTable | GROUP BY |
groupby().agg() |
| Lookup / join | XLOOKUP |
JOIN |
merge() |
| New column | Formula dragged down | Expression in SELECT |
df["new"] = ... |
| Pivot | PivotTable | Conditional aggregation | pivot_table() |
If you already know one of these tools, most pandas concepts will feel familiar. Our Excel to Python guide covers the mindset shift, and SQL vs Python explains when to use which.
df.dropna() returns a new DataFrame. It doesn't change df unless you write df = df.dropna().and / or in filters. Use & and |, with parentheses around each condition.df[df["region"] == "North"]["revenue"] = 0. It may not change anything. Use .loc instead.for i in range(len(df)), there's almost always a better way.Keep our Python cheat sheet for data analysts open while you practice, and apply the 12-step data cleaning checklist to a real dataset. For structured learning, Data Analysis with Python (DataCamp) teaches these operations through interactive exercises. The DataCamp Data Scientist Career Track continues into visualization, statistics and machine learning. Browse all Data Analysis courses.
Is pandas hard to learn? The basics above take a week or two of regular practice for someone comfortable with Excel. Fluency, meaning writing these without looking things up, comes from using them on real data for a few months.
Should I learn pandas or SQL first? For most analyst roles, SQL first, then pandas. They share the same concepts (filter, group, join), so learning the second is much faster.
Is pandas still relevant with newer libraries like Polars? Yes. Polars is faster for large data and growing quickly, but pandas remains the most widely used library in tutorials, job postings and existing code. The concepts transfer, so learning pandas first is a safe choice.
How big a dataset can pandas handle? Comfortably, data that fits in your computer's memory, typically up to a few million rows on a normal laptop. Beyond that, look at Polars, DuckDB or Spark.
You don't need all of pandas, just 15 operations: load, inspect, select, filter, .loc, sort, new columns, value counts, group-by, missing values, dates, merge, pivot, deduplicate and save. They map directly to what you already do in Excel or SQL. Practice them on a real, messy dataset and most everyday analysis tasks become routine.
A DataCamp track teaching pandas, NumPy and data cleaning techniques for analysts moving from spreadsheets or SQL into Python-based.
A 23-course Python learning path covering data manipulation, visualization, statistics, and machine learning for aspiring data scientists.