What does reset_index do in pandas and when do I need drop=True?

The short version

reset_index() turns the current index into an ordinary column and gives the frame a fresh 0, 1, 2 index. Pass drop=True when the old index carries no information (typically after filtering), otherwise it is kept as a new column you did not want.

The two behaviours

import pandas as pd

df = pd.DataFrame({"value": [1, 2, 3]}, index=[10, 11, 12])

df.reset_index().columns.tolist()        # ['index', 'value'] , old labels kept
df.reset_index(drop=True).index.tolist() # [0, 1, 2] , old labels discarded

Without drop=True, the labels 10, 11, 12 become a column called index. With it, they are thrown away.

After filtering: use drop=True

Filtering keeps the original labels, so the index becomes gappy (0, 3, 7). Those numbers usually mean nothing once the rows are gone.

clean = df[df["value"] > 1].reset_index(drop=True)

Without drop=True you would end up with a stale column of old row numbers cluttering the frame. This is the most common use of reset_index.

After groupby: usually no drop

groupby puts the grouping keys in the index, and there you almost always want them back as columns, so drop=True would be wrong.

totals = df.groupby("city")["amount"].sum().reset_index()   # city becomes a column again

Dropping here would delete the very keys that identify each row, leaving anonymous totals.

The shortcut for groupby

You can skip the reset entirely by telling groupby not to use the index in the first place.

df.groupby("city", as_index=False)["amount"].sum()

This produces the same flat result directly, which many people find cleaner than grouping and then resetting.

Why the index matters at all

The index is how pandas aligns data. Two Series with different indexes will align on labels, not positions, when you add them, producing NaN where labels do not match. So a stale index from a filter can silently misalign a later operation, which is a genuinely confusing bug. Resetting it after filtering removes that risk.

The name-collision quirk

If a column called index already exists, reset_index names the new one level_0 to avoid a clash, which looks odd. That is a signal you probably wanted drop=True.

Reassign the result

Like most pandas methods, reset_index returns a new frame. Assign it back, or pass inplace=True. Reassignment is clearer and chains naturally, which is why most modern pandas code prefers it.