Stats, Cumulative & Rounding

After exploring a Series, you usually want answers: who top-scored, what is the typical score, how did the total build up over by over, and how do you cap or round values for a clean report. This topic covers the statistical and transformation methods you will use most.

Definition

idxmax/idxmin return the label of the largest/smallest value. describe returns a summary of statistics in one Series. Cumulative methods (cumsum, cumprod, cummax, cummin) return a Series of the same length where each position holds the running total/product/maximum/minimum up to that point. clip limits values to a range, and round rounds them to a number of decimals.

• Label vs position — idxmax() gives a label ("Pant"); argmax() gives a position (3).
• Running — each output value depends on all the values before and including it.
• Same length — cumulative methods, clip and round do not reduce the Series; they transform each element.

Syntax

s.idxmax()   s.idxmin()   s.argmax()   s.argmin()
s.describe(percentiles=None)
s.agg(["min", "max", "mean"])
s.cumsum()   s.cumprod()  s.cummax()   s.cummin()
s.clip(lower=None, upper=None)
s.round(decimals=0)

Example — idxmax / idxmin vs argmax / argmin

import pandas as pd

runs = pd.Series([45, 78, 12, 101, 33],
                 index=["Rohit", "Virat", "Gill", "Pant", "Hardik"])
print(runs.idxmax(), runs.idxmin())
print(runs.argmax(), runs.argmin())
print(runs.max(), runs[runs.idxmax()])

Pant Gill
3 2
101 101

How it works: max() tells you the top score (101), idxmax() tells you who scored it (Pant), and argmax() tells you where it is (position 3). Use idxmax with .loc, and argmax with .iloc.

Example — describe

print(runs.describe())

count      5.000000
mean      53.800000
std       35.604775
min       12.000000
25%       33.000000
50%       45.000000
75%       78.000000
max      101.000000
dtype: float64

For a text Series, describe shows different statistics:

teams = pd.Series(["CSK", "MI", "KKR", "MI", "CSK", "MI"])
print(teams.describe())

count      6
unique     3
top       MI
freq       3
dtype: object

unique = number of distinct values, top = most frequent value, freq = how often it appears.

Example — cumulative methods: an innings over by over

over_runs = pd.Series([8, 4, 12, 6, 15, 9], index=[1, 2, 3, 4, 5, 6])
print(pd.DataFrame({"runs": over_runs,
                    "cumsum": over_runs.cumsum(),
                    "cummax": over_runs.cummax(),
                    "cummin": over_runs.cummin()}))

   runs  cumsum  cummax  cummin
1     8       8       8       8
2     4      12       8       4
3    12      24      12       4
4     6      30      12       4
5    15      45      15       4
6     9      54      15       4

How it works:

• cumsum — the team score after each over: 8, 8+4=12, 12+12=24, …
• cummax — the best over so far: it only changes when a new record is set (12 in over 3, 15 in over 5).
• cummin — the quietest over so far.

cumprod multiplies instead of adding — useful for growth factors:

growth = pd.Series([1.10, 1.05, 0.95, 1.20])   # season-on-season change in followers
print(growth.cumprod().round(4))

0    1.1000
1    1.1550
2    1.0973
3    1.3167
dtype: float64

After four seasons the follower count is 1.3167 × the original (+31.67%).

Example — clip

print(runs.clip(lower=20, upper=80))

Rohit     45
Virat     78
Gill      20
Pant      80
Hardik    33
dtype: int64

Gill's 12 was raised to the lower limit 20, Pant's 101 was cut to the upper limit 80; values in between are unchanged. clip is used to cap outliers (e.g. treat any score above the 99th percentile as that percentile). Pass only one side to cap one end: runs.clip(upper=50).

Example — round

sr = pd.Series([138.4567, 145.2049, 129.9951])
print(sr.round(1))
print(sr.round())

0    138.5
1    145.2
2    130.0
dtype: float64
0    138.0
1    145.0
2    130.0
dtype: float64

round(1) keeps one decimal; round() (= round(0)) rounds to whole numbers but the dtype stays float (138.0). Convert with .astype(int) if you need integers.

Note: pandas (like NumPy) uses "round half to even": pd.Series([2.5, 3.5]).round() gives 2.0, 4.0. This avoids bias when rounding many values.

Comparison

Common mistakes

• ❌ Using max() when you need who scored most. ✅ idxmax() returns the label.
• ❌ Mixing argmax() with .loc. ✅ argmax → .iloc; idxmax → .loc.
• ❌ Expecting round() to produce integers. ✅ It stays float; add .astype(int).
• ❌ s.clip(20, 80) on a text Series. ✅ clip is for numbers (or other orderable types like dates).

Interview tip

Q: Difference between idxmax() and argmax() on a Series? A: idxmax() returns the index label of the maximum value; argmax() returns its integer position. They only coincide when the index is the default RangeIndex.