A DataFrame is Pandas' 2D labeled data structure — rows and columns, both with labels (an index and column names) — built from one or more Series (its 1D building block).
DataFrame vs Series
import pandas as pd
s = pd.Series([25, 30, 35], name="age") # 1D, labeled
print(type(s)) # <class 'pandas.core.series.Series'>
df = pd.DataFrame({
"age": [25, 30, 35],
"income": [40000, 55000, 62000]
})
print(type(df)) # <class 'pandas.core.frame.DataFrame'>
print(type(df["age"])) # Series — a single column of a DataFrame IS a SeriesThe Index — Often Overlooked, Frequently the Source of Bugs
df = pd.DataFrame({"score": [88, 92, 79]}, index=["Amit", "Priya", "Rahul"])
print(df.loc["Priya"]) # label-based lookup using the index
print(df.reset_index()) # turns the index back into a normal column, numeric index restoredAfter filtering a DataFrame, the index keeps the original row numbers (it doesn't reset to 0,1,2...) — this trips up beginners who then try to use .iloc[] with the old numbers.
.loc vs .iloc
.loc | .iloc | |
|---|---|---|
| Selects by | Label (index name, column name) | Integer position (0-based) |
| Example | df.loc["Priya", "score"] | df.iloc[1, 0] |
| Slicing end | Inclusive of the end label | Exclusive, like normal Python slicing |
dtypes Matter More Than They Look
df.dtypes
# age int64
# income int64
# city object -> usually strings
df["income"] = df["income"].astype(float) # explicit conversion when neededA column stored as object when it should be numeric (e.g. "45000" as a string instead of 45000) will silently break most ML preprocessing until explicitly converted — always check df.dtypes after loading new data.
Practical Use Cases
- Every dataset loaded for an ML project starts life as a DataFrame
- Filtering, joining and reshaping data before feature engineering
- Feeding
X(DataFrame) andy(Series) directly into scikit-learn
Common Mistakes
- Assuming the index is always 0,1,2,... after filtering — use
.reset_index(drop=True)if you need a clean sequential index. - Mixing
.locand.ilocsemantics —.loc[0:5]includes row label 5;.iloc[0:5]stops before position 5.
Interview Relevance
Q: "What's the difference between .loc and .iloc?" .loc selects by label (index/column names, inclusive slicing), .iloc selects by integer position (exclusive slicing, like standard Python) — this distinction is one of the most commonly asked Pandas interview questions.
Practice Question
Given a DataFrame df with a default integer index, write the code to select rows 10 through 20 (inclusive) using both .loc and .iloc, and explain why the results differ.
