Why Pandas Questions Reveal More Than Any Resume Line

import pandas as pd

s = pd.Series(['UPI', 'Card', 'UPI', 'Cash', 'UPI', 'Card'])
print(s.value_counts(normalize=True))

# Prediction a daily user gives:
# UPI     0.500000
# Card    0.333333
# Cash    0.166667
# (shares, sorted from most to least frequent)

# Follow-up: what if there are missing values?
s2 = pd.Series(['UPI', None, 'UPI'])
print(s2.value_counts())               # NaN not counted
print(s2.value_counts(dropna=False))   # NaN counted as its own value

import pandas as pd
import numpy as np

df = pd.DataFrame({'city': ['Delhi', np.nan, 'Delhi', 'Pune'],
                   'sales': [100, 250, 50, 80]})

print(df.groupby('city')['sales'].sum())
# city
# Delhi    150
# Pune      80
# The 250 from the missing city is silently gone.

print(df.groupby('city', dropna=False)['sales'].sum())
# Delhi    150
# Pune      80
# NaN      250

print(df['sales'].sum())   # 480, so totals no longer match

import pandas as pd

df = pd.DataFrame({'amount': [500, 7000, 9000],
                   'flag': [0, 0, 0]})

# Chained: filter, then set on the result
df[df['amount'] > 5000]['flag'] = 1
# SettingWithCopyWarning (or ChainedAssignment warning)
print(df['flag'].tolist())   # [0, 0, 0]  - nothing changed

# Correct: one .loc call with row mask and column
df.loc[df['amount'] > 5000, 'flag'] = 1
print(df['flag'].tolist())   # [0, 1, 1]

import pandas as pd

sales = pd.DataFrame({
    'city':  ['Delhi', 'Delhi', 'Pune'],
    'month': ['Jan', 'Jan', 'Jan'],
    'amt':   [100, 50, 80]
})

# sales.pivot(index='city', columns='month', values='amt')
# ValueError: Index contains duplicate entries, cannot reshape

print(sales.pivot_table(index='city', columns='month',
                        values='amt', aggfunc='sum'))
# month   Jan
# city
# Delhi   150
# Pune     80

# Note: pivot_table's default aggfunc is 'mean', so
# always pass aggfunc explicitly.

import pandas as pd

# A strong candidate can explain each argument
orders = pd.read_csv(
    'orders.csv',
    usecols=['order_id', 'customer_id', 'city', 'order_date', 'amount'],
    dtype={'customer_id': 'string', 'city': 'category'},
    parse_dates=['order_date'],
    dayfirst=True
)

print(orders.dtypes)
print(orders.memory_usage(deep=True).sum() / 1024**2, 'MB')

# Very large file: process in chunks
total = 0
for chunk in pd.read_csv('orders.csv', usecols=['amount'], chunksize=100_000):
    total += chunk['amount'].sum()
print(total)

BEFORE
Skills: Python (Pandas, NumPy, Matplotlib),
Data Cleaning, EDA, Large Datasets

AFTER
Project: E-commerce Orders Analysis (Python, Pandas)
- Loaded orders and customers CSVs with explicit dtypes
  and dayfirst date parsing; converted city to category
  to reduce memory
- Fixed duplicate customer IDs before merging and
  verified row counts with validate='many_to_one'
- Built month-wise revenue and repeat-purchase metrics
  with groupby, transform and pivot_table

Every bullet invites a question you can answer.

WRITE YOUR PREDICTION, THEN RUN IT

1. pd.Series(['a','b','a']).value_counts(normalize=True)
2. df.groupby('k')['v'].sum()  where k has NaN
3. df[df.v > 0]['w'] = 1  then print df['w']
4. df.pivot(...) with duplicate index/column pairs
5. df.pivot_table(...) with no aggfunc given
6. pd.to_datetime('05/03/2025')
7. df.sort_values('v'); df.head()
8. pd.Series([1, None, 3]).mean()

Score: ___ / 8 correct before running.
Below 6 -> practise before listing Pandas.

ANSWER OUT LOUD, 60 SECONDS EACH

1. What did the raw data look like?
2. Which column needed the most cleaning, and how?
3. How did you load the file, and why those options?
4. Which merge did you use, and what did you check?
5. One number in your results: how did you verify it?
6. One thing that broke and how you found it.
7. What would you do differently now?

If any answer starts with 'I think I...' -> reopen
the notebook and refresh your memory.

BUILD EACH PANDAS BULLET LIKE THIS

[Action verb] + [data] + [Pandas method] + [check/result]

Examples:
- Cleaned 3 messy date and amount columns with
  to_datetime and to_numeric, logging rows that failed
- Merged orders with customers using validate to
  catch duplicate keys
- Built monthly retention table with groupby and
  pivot_table

Rule: no bullet you cannot explain line by line.