Building a DataFrame

The most common way to construct a small DataFrame is from a dictionary, where each key becomes a column name and each value is the data for that column:

import pandas as pd

df = pd.DataFrame({
    "name": ["Ada", "Lin", "Sam"],
    "score": [88, 92, 79],
})
print(df)

   name  score
0   Ada     88
1   Lin     92
2   Sam     79

That leftmost unlabeled column is the index — a row identifier pandas adds automatically (here, just 0, 1, 2) unless you set something more meaningful, like a student ID.

A handful of methods answer almost every "what does this look like" question you'll have early on:

df.shape        # (3, 2) — rows, columns, same idea as NumPy
df.columns      # Index(['name', 'score'], dtype='object')
df.dtypes       # the type pandas inferred for each column
df.head(2)      # first 2 rows — essential once a table has thousands
df.describe()   # count, mean, std, min/max for numeric columns

df.describe() in particular is worth running on any new dataset immediately — it's a fast sanity check for obviously wrong data (a negative age, a score of 1000 out of 100) before you do anything else with it.

df["score"]              # one column, returned as a pandas Series
df[["name", "score"]]    # multiple columns, returned as a DataFrame (note the double brackets)

The double-bracket form is a common early stumbling block: df["name", "score"] (single brackets, comma inside) raises an error, while df[["name", "score"]] (a list inside brackets) is the correct way to select several columns at once.

New columns are created just by assigning to a name that doesn't exist yet:

df["pass"] = df["score"] >= 80
print(df)

   name  score   pass
0   Ada     88   True
1   Lin     92   True
2   Sam     79  False

This is the same vectorized comparison from the previous page — df["score"] >= 80 — applied to a DataFrame column (which is really a pandas Series, a labeled 1D array built on NumPy underneath) instead of a plain array.

Once a table exists, a common next question is "who's on top?" — sort_values handles this directly:

df.sort_values("score", ascending=False)

This returns a new DataFrame ordered by score, highest first, without changing df itself unless you pass inplace=True (generally worth avoiding while learning — an explicit reassignment like df = df.sort_values(...) is easier to reason about later when reading your own code back). Sorting by more than one column works the same way with a list: df.sort_values(["class", "score"], ascending=[True, False]) sorts by class first, then by score (highest first) within each class.

Real datasets often arrive with column names that don't match your code's conventions — Student Name instead of name, or an all-caps SCORE. Renaming avoids either fighting the original names everywhere or rebuilding the DataFrame from scratch:

df = df.rename(columns={"Student Name": "name", "SCORE": "score"})

This is worth doing immediately after loading a messy real-world file, before writing any other code against it — consistent column names make every later line easier to read and less error-prone to type.