Python interviews mix language fundamentals with practical data handling. You will be asked to predict output, pick the right built-in data structure and explain behaviour like mutability and default arguments.
Below are 60 Python questions with the correct answer and a short explanation, grouped from fresher to experienced level. Try each one before you open the answer.
With from collections import Counter and c = Counter('apple'), what is c['z']?
Answer: A. 0
Counter returns 0 for keys it has never seen, without adding them, which makes counting code shorter. A plain dict raises KeyError for the same lookup.
In Python, what type is x after x = {}?
Answer: B. dict
Empty braces create an empty dict, for historical reasons. Use set() for an empty set. {1, 2} with values but no colons does create a set.
What happens with s.remove(9) and s.discard(9) when 9 is not in set s?
Answer: C. remove raises KeyError; discard does nothing
remove insists the item is present and raises KeyError otherwise. discard quietly ignores missing items. Pick remove when a missing item would be a bug you want to see.
Which expression turns codes = {'AE': 'UAE', 'IN': 'India'} into {'UAE': 'AE', 'India': 'IN'}?
Answer: B. {v: k for k, v in codes.items()}
A dict comprehension over items() swaps each pair. Iterating codes directly gives only keys, so unpacking fails, and dicts have no invert method. This is safe only when values are unique.
In Python, what does bisect.bisect_left([10, 20, 30, 40], 25) return?
import bisect print(bisect.bisect_left([10, 20, 30, 40], 25))
Answer: B. 2
bisect_left returns the index where 25 would be inserted to keep the list sorted, which is before 30, at index 2. It never returns -1; for a value larger than everything it returns len(list).
An invoice table stores amounts in rupees with paise, and GST must add up exactly. Which column type should you use?
Answer: B. numeric(12,2)
numeric stores exact decimal values, so ₹0.10 + ₹0.20 is exactly ₹0.30. Floating-point types like real and double store binary approximations and can produce tiny rounding errors in totals. text would allow invalid values and break arithmetic.
What is the main benefit of writing with open("sales.csv") as f: instead of f = open("sales.csv")?
Answer: C. The file is closed even if an error occurs
A with block uses a context manager, which closes the file when the block ends, whether it ends normally or because of an exception. Without it, an error before f.close() can leave the file open. The same pattern is used for database connections and locks.
Which call renames the column amt to amount and keeps all other columns?
Answer: B. df.rename(columns={"amt": "amount"})
rename with a columns dict changes only the listed names and returns a new DataFrame. Assigning df.columns or using set_axis replaces every name and fails when the count does not match. In the third option the mapping is backwards and, without columns=, it would apply to the index.
totals = df.groupby("city")["sales"].sum() returns a Series with city as the index. How do you get a DataFrame with city and sales columns?
Answer: B. totals.reset_index()
reset_index moves the index back into a normal column, giving a two-column DataFrame. Passing as_index=False to groupby gives the same result in one step. Transposing would turn cities into columns, which is not what you want.
Which line creates a column that says "high" when amount is above 1000 and "low" otherwise?
Answer: D. df["band"] = np.where(df["amount"] > 1000, "high", "low")
np.where(condition, a, b) works element by element and returns a for True and b for False. A Python if expression needs a single True or False and raises an ambiguous truth value error on a Series. For more than two bands, use np.select with lists of conditions and choices, or pd.cut.
Which call returns the three rows with the highest sales?
Answer: C. df.nlargest(3, "sales")
nlargest(n, column) returns the n rows with the largest values, already ordered from highest. sort_values without ascending=False puts the smallest values first, so head(3) gives the bottom three. There is no top method on a DataFrame.
Two lists of customer IDs are turned into sets a = {1, 2, 3} and b = {2, 3, 4}. What is a ^ b?
Answer: B. {1, 4}
^ is symmetric difference: elements in either set but not in both. Here 1 is only in a and 4 is only in b. & gives the intersection {2, 3}, | gives the union, and a - b gives {1}.
You want to print a ranked list starting at 1. What does this print first?
names = ["Asha", "Omar", "Ravi"]
for i, name in enumerate(names, start=1):
print(i, name)Answer: C. 1 Asha
enumerate yields (index, item) pairs, and start=1 makes the counter begin at 1 instead of 0. The first pair is (1, 'Asha'). This is cleaner than keeping a manual counter or using range(len(names)).
A DataFrame has a datetime column order_date and an amount column. Which expression gives total amount per calendar month?
Answer: B. df.groupby(df['order_date'].dt.to_period('M'))['amount'].sum()
dt.to_period('M') turns each date into a monthly period like 2026-03, so all days of a month group together. Grouping by the raw date gives daily totals, and dt.day would mix the 5th of every month. resample('MS') on a date index is another option.
What does this print in Python 3.8 or later?
total = 2500
print(f"{total=}")Answer: C. total=2500
Adding = inside an f-string field prints the expression text, an equals sign and its value. It is a quick way to debug values without typing the name twice. You can combine it with format specs, as in f"{total=:,}".
After importing a CSV into Excel, employee IDs like 00451 now show as 451. What is the best fix?
Answer: A. Import the ID column as Text so it keeps its exact characters
Excel guessed the column was a number and dropped the leading zeros. IDs, phone numbers and postal codes are labels, not quantities, so they should be stored as text. Set the column type to Text during import (for example in Power Query) so the original value is kept.
A survey asks customers to rate satisfaction from 1 (very unhappy) to 5 (very happy). What type of variable is this?
Answer: D. Ordinal
The values have a clear order, but the gap between 1 and 2 is not guaranteed to equal the gap between 4 and 5. That makes it ordinal. Medians and percentage of 4s and 5s are often safer summaries than a precise mean.
A data scientist has a pandas DataFrame df with columns city and delivery_min. Which seaborn call draws one box plot per city?
Answer: D. sns.boxplot(data=df, x="city", y="delivery_min")
sns.boxplot with a categorical x and numeric y draws one box per category. countplot counts rows per value, lineplot draws lines, and heatmap needs a 2D matrix of numbers.
Names imported from a form look like " Anil Sharma ". Which function turns this into "Anil Sharma"?
Answer: C. TRIM
TRIM removes spaces at the start and end, and reduces runs of spaces between words to one. CLEAN removes non-printing control characters, and PROPER changes capitalisation. Extra spaces are a very common reason lookups fail.
A staging table has repeated rows for the same order_id. Which pattern keeps one row per order_id?
Answer: A. Number rows per order_id with ROW_NUMBER and keep rn = 1
ROW_NUMBER over PARTITION BY order_id numbers each copy 1, 2, 3, so keeping rn = 1 leaves one row per order. Keeping COUNT = 1 throws away every duplicated order entirely. Order the window by a load timestamp so you keep the copy you want.
In Python, seen[[r, c]] = True raises TypeError: unhashable type: 'list'. What is the usual fix?
Answer: C. Use a tuple: seen[(r, c)] = True
Lists are mutable, so Python does not let them be hashed. Tuples of hashable values are immutable and hashable, which makes them the normal choice for grid coordinates and composite keys.
What is Counter(a=3, b=1) - Counter(a=1, b=2) in Python?
Answer: B. Counter({'a': 2})
The - operator keeps only positive results, so b (1 - 2 = -1) is dropped. If you need negatives, use c.subtract(other), which updates in place and keeps zero and negative counts.
An automation script sometimes hangs for hours on this line. What is the fix?
resp = requests.get("https://inventory.internal/api/hosts")Answer: C. Pass a timeout, for example requests.get(url, timeout=(3, 10)), and handle the timeout exception
The Python requests library has no default timeout, so a server that accepts the connection but never answers can block forever. Passing a connect and read timeout makes the call fail fast, and then the script can retry with backoff or alert. Every network call in automation should have a timeout.
You run ALTER TABLE orders ALTER COLUMN id TYPE bigint on a 500 GB table. What should you expect?
Answer: C. A full rewrite while all access is blocked
int and bigint have different on-disk sizes, so every row and every index containing the column must be rewritten, and an ACCESS EXCLUSIVE lock blocks all access the whole time. On large tables teams add a new bigint column, backfill in batches, then swap in a short window.
What does print(a) show?
a = [[1, 2], [3]] b = a.copy() b[0].append(9) b.append([4]) print(a)
Answer: A. [[1, 2, 9], [3]]
list.copy() is a shallow copy: b is a new outer list, but its items are the same inner list objects as in a. Appending to b[0] changes the inner list that a also holds, while appending a new item to b does not touch a. Use copy.deepcopy when you need independent nested data.
You want students sorted by score from highest to lowest, and alphabetically by name when scores tie. Which call does it?
Answer: A. sorted(rows, key=lambda r: (-r["score"], r["name"]))
A tuple key sorts by the first item, then uses the second to break ties. Negating the numeric score gives descending order while names stay ascending. Using reverse=True would reverse both parts, putting names in Z to A order on ties.
What does this print?
def f(a, *args, **kwargs):
print(a, args, kwargs)
f(1, 2, 3, city="Pune")Answer: B. 1 (2, 3) {'city': 'Pune'}
The first positional argument fills a, extra positional arguments are collected into the tuple args, and keyword arguments not matched by name go into the dict kwargs. So args is (2, 3), a tuple not a list, and kwargs is {'city': 'Pune'}.
What is the result of this comprehension?
d = {"a": 1, "b": 2, "c": 1}
{v: k for k, v in d.items()}Answer: A. {1: 'c', 2: 'b'}
Keys in a dict are unique, so when the value 1 appears twice, the later pair overwrites the earlier one. Items are visited in insertion order, so "c" wins over "a". If you need all keys per value, build a dict of lists with defaultdict(list).
Which call returns one row per city with columns called total and orders?
Answer: A. df.groupby("city").agg(total=("sales", "sum"), orders=("order_id", "nunique"))
Named aggregation takes keyword arguments where each value is a (column, function) pair, and the keyword becomes the output column name. The dict form expects existing column names as keys, so "total" would fail. Named aggregation avoids messy multi-level column names.
Sales are recorded on Jan 1 (1), Jan 3 (2) and Jan 9 (3), with the date as the index. What does rolling("7D").sum() give on Jan 9?
Answer: B. 5
A "7D" window covers the 7 days ending at each row, here Jan 3 to Jan 9, so it includes 2 and 3. Jan 1 is more than 7 days back and drops out. rolling(3) would count rows instead of days and give 6, which is wrong when dates have gaps.
event_time holds UTC times with no timezone attached. Which .dt steps convert it to Dubai local time?
Answer: A. tz_localize("UTC"), then tz_convert("Asia/Dubai")
tz_localize says what timezone naive times are already in, and tz_convert moves aware times to another zone. Localising straight to Asia/Dubai would label UTC times as Dubai times without shifting them. tz_convert on naive times raises an error, and Dubai is UTC+4, so subtracting 4 hours goes the wrong way.
After orders.merge(products, on="sku"), the result has price_x and price_y. What does that mean?
Answer: D. Both tables had a non-key column called price
When both sides share a column name that is not a key, pandas keeps both and adds _x (left) and _y (right). Use suffixes=("_order", "_list") to give clear names, or drop or rename the column on one side first. Here it is likely the sale price versus the list price.
Scores are 100, 90, 90 and 80. With ascending=False, what ranks does method="dense" give?
Answer: D. 1, 2, 2, 3
Dense ranking gives ties the same rank and the next score the next whole number, so 80 is 3. method="min" also gives ties rank 2 but skips to 4 for the next score, like sports rankings. The default method="average" gives each tie 2.5.
A loop builds a CSV line for each of 1 million rows and adds each to a big string with +=. What is the most reliable way to make this fast?
Answer: D. Collect the lines in a list and call ''.join(lines)
Strings are immutable, so += may copy the whole string each time, which can turn the loop quadratic. Collecting pieces in a list and joining once does a single pass. For files, writing lines directly or using the csv module avoids the big string entirely.
What does this print?
words = ["bb", "a", "cc", "d"] print(sorted(words, key=len))
Answer: B. ['a', 'd', 'bb', 'cc']
Python's sort is stable: items with equal keys keep their original order. 'a' and 'd' both have length 1 and 'a' came first; 'bb' came before 'cc'. Stability is what lets you sort by a second key first, then by the main key.
For order values, Q1 = 10 and Q3 = 20. Using the 1.5 x IQR rule, above what value is an order flagged as a high outlier?
Answer: C. 35
The interquartile range is Q3 - Q1 = 10. The upper fence is Q3 + 1.5 x IQR = 20 + 15 = 35, and the lower fence is 10 - 15 = -5. The IQR rule uses quartiles, so it is not pulled around by the outliers themselves the way a mean plus standard deviation rule is.
In pandas, which line matches LAG(sales) OVER (PARTITION BY store ORDER BY date)?
Answer: D. df.sort_values('date').groupby('store')['sales'].shift(1)
shift(1) moves values down one row, like LAG, and groupby keeps each store separate. You must sort by date first or the "previous" row is arbitrary. shift(-1) is LEAD, and without groupby the shift crosses stores.
A partner file has dates like 03/04/2026, and some rows were loaded as 4 March while others as 3 April. What is the best fix?
Answer: B. Parse one agreed format; prefer ISO 8601
dd/MM and MM/dd look the same for days 1–12, so automatic inference mixes them. Parse with the format the partner confirms, and prefer ISO 8601 (yyyy-MM-dd) in contracts because it is unambiguous and sorts correctly.
An API returns "deliveryAt": "2026-05-10 18:00:00" with no offset. Clients in India and the UAE show different times. What should the contract require?
Answer: D. ISO 8601 with an offset, e.g. ...T18:00:00+04:00
A time with no zone is ambiguous; each client guesses its own zone. ISO 8601 with Z or an explicit offset makes the instant clear. Tests should check the format and that the value is right for the business zone.
Each row of df holds one quarter: quarter-end receivables and that quarter's revenue. What is wrong with this DSO calculation?
df['dso'] = df['receivables'] / df['revenue'] * 365
Answer: D. It uses 365 days with one quarter's revenue
Quarterly revenue covers about 91 days, so multiplying by 365 overstates DSO roughly fourfold. Use the days in the quarter (about 91), or annualise revenue first.
Which statement about iteration order of Python dicts and sets is correct today?
Answer: A. Dicts keep insertion order by spec; sets do not
Since Python 3.7 the language guarantees that dicts preserve insertion order. Sets make no ordering promise, and string hash randomization can change their order between runs. Use a dict with None values, or sort, when order matters.
A Python script prints list(set(names)) and the order differs between runs, breaking a snapshot test. Why?
Answer: A. String hashes are randomized per process
Python seeds string hashing randomly at startup (PYTHONHASHSEED) to resist hash flooding, so set iteration order for strings changes between runs. Sort the output, or use dict.fromkeys to keep insertion order.
What is {1: 'a', 1.0: 'b', True: 'c'} in Python?
Answer: C. {1: 'c'}
1 == 1.0 == True and they have the same hash, so they are one key. The first inserted key object, 1, is kept, while each later assignment replaces the value, so the final value is 'c'.
What does this code print?
import pandas as pd s = pd.Series([10, 20, 30, 40, 50]) print(s.rolling(3).mean().iloc[[1, 4]].tolist())
Answer: B. [nan, 40.0]
rolling(3) needs three values by default (min_periods equals the window), so position 1 has only two and returns NaN. Position 4 averages 30, 40 and 50 to give 40.0. The window is trailing, which is right for features, since it never looks ahead.
What happens when show() is called?
count = 10
def show():
print(count)
count = 5
show()Answer: C. It raises UnboundLocalError
Because count is assigned somewhere inside show, Python treats count as a local variable for the whole function. The print runs before the local has a value, so it raises UnboundLocalError. Declare global count if you really mean the global, or better, pass and return values.
What happens when Python reaches this class definition?
from dataclasses import dataclass
@dataclass
class Cart:
items: list = []Answer: D. It raises a ValueError right away
dataclasses refuse list, dict and set defaults and raise ValueError when the class is defined, because one shared list would leak between instances. The fix is items: list = field(default_factory=list), which calls list() for each new object. This is the dataclass version of the mutable default argument trap.
What does print(round(2.5), round(3.5)) show?
Answer: C. 2 4
Python's round uses "round half to even", also called banker's rounding: ties go to the nearest even number, so 2.5 becomes 2 and 3.5 becomes 4. This avoids an upward bias when rounding many values. For invoice rounding that must always go half up, use decimal.Decimal with ROUND_HALF_UP.
What are s.nunique() and len(s.unique())?
s = pd.Series([1, 1, np.nan, 2, np.nan])
Answer: B. 2 and 3
nunique() skips missing values by default, so it counts only 1 and 2. unique() returns every distinct value including NaN once, giving [1, 2, nan], which has length 3. Use nunique(dropna=False) if NaN should count as its own value.
How many rows does df.explode("tags") return?
df = pd.DataFrame({"id": [1, 2, 3],
"tags": [["a", "b"], ["c"], []]})Answer: C. 4
explode makes one row per list item, so id 1 gives two rows and id 2 gives one. An empty list does not disappear; it becomes a single row with NaN in tags. That gives 4 rows, and the repeated index labels show which original row each came from.
df holds daily sales for many stores, sorted by store then date. df["sales"].pct_change() gives odd values on the first day of each store. Why, and what is the fix?
Answer: D. It crosses store boundaries; use groupby("store") first
pct_change compares each row with the row above, ignoring which store it belongs to. At each store boundary, it divides by the previous store's last value. Grouping by store restarts the calculation per store, giving NaN on each store's first day, which is correct.
What does print(a) show?
a = np.arange(5) b = a[1:3] b[0] = 99 c = a[[3, 4]] c[0] = -1 print(a)
Answer: C. [ 0 99 2 3 4]
Basic slicing returns a view that shares memory with a, so writing to b changes a. Fancy indexing with a list of positions returns a new copy, so writing to c leaves a alone. Call .copy() on a slice when you need independent data.
orders.cust_id is int64, but customers.cust_id was read as text like "101". What does orders.merge(customers, on="cust_id") do?
Answer: C. It raises a ValueError about int64 and str columns
pandas refuses to merge numeric keys with string keys and raises a ValueError telling you the types differ. Fix the types first, for example customers["cust_id"] = customers["cust_id"].astype("int64"), after stripping spaces. Checking df.dtypes before a merge saves a lot of debugging.
For x = [2, 4, 4, 4, 5, 5, 7, 9], np.std(x) gives 2.0 but pd.Series(x).std() gives about 2.14. Why?
Answer: B. NumPy divides by n, pandas divides by n - 1
The squared deviations sum to 32. NumPy's default ddof=0 gives the population formula, sqrt(32 / 8) = 2.0, while pandas defaults to ddof=1, the sample formula, sqrt(32 / 7) which is about 2.14. Pass ddof explicitly so results match across tools.
Group "a" has rows (p=NaN, q=1) and (p="x", q=2). What does df.groupby("c").first() return for group "a"?
Answer: D. p = "x" and q = 1
first() returns the first non-missing value of each column separately, so p comes from the second row and q from the first. The output row never existed in the data. Use .head(1) or .nth(0) when you need the actual first row.
What does pd.Series([10, 0]) / pd.Series([0, 0]) return?
Answer: B. inf and NaN
pandas follows floating-point rules: a positive number divided by 0 gives inf, and 0 / 0 gives NaN, with no error. Infinite values then wreck averages and charts. Replace zero denominators first, or clean the result with .replace([np.inf, -np.inf], np.nan).
A score column holds [70, NaN, 90, 80]. Where does NaN end up after df.sort_values('score', ascending=False)?
Answer: B. Last, because na_position defaults to 'last'
pandas puts missing values at the end by default whatever the sort direction, because na_position='last'. The result is 90, 80, 70, NaN. Pass na_position='first' to put them at the top, or dropna() before ranking.
An attendance export has times as text like "09:45". =B2-A2 works for some rows but others give #VALUE! because they hold "9.45 am". Which approach is most reliable?
Answer: D. Standardise the text, then convert with TIMEVALUE
Excel can convert some text times on the fly, but not non-standard ones like "9.45 am". Clean them first, for example SUBSTITUTE the dot with a colon, then use TIMEVALUE to get a real time. Formatting alone never changes text into numbers.
In Snowflake, which expression fills each NULL reading with the latest earlier non-NULL reading?
Answer: C. LAST_VALUE(reading) IGNORE NULLS OVER (ORDER BY ts ROWS UNBOUNDED PRECEDING)
With a frame from the start to the current row, LAST_VALUE IGNORE NULLS returns the most recent non-NULL value. FIRST_VALUE IGNORE NULLS returns the very first reading in the series. Without IGNORE NULLS, LAST_VALUE just returns the current row, NULL included.
A templated sql field is built as below. The query fails because the date filter is wrong. What did the operator actually receive?
table = "sales"
sql = f"SELECT * FROM {table} WHERE dt = '{{ ds }}'"Answer: A. SELECT * FROM sales WHERE dt = '{ ds }'
In a Python f-string, {{ and }} are escapes for literal single braces, so the string becomes { ds }, which Jinja does not treat as a template. Write {{{{ ds }}}} inside the f-string, or build the SQL with normal string joining.
An API sends "amount": 1999.99 as a JSON number. A tester suggests 199999 in paise, or the string "1999.99". Why?
Answer: C. Binary floats cannot hold many decimals exactly
Many JSON parsers turn decimals into binary floats, which can add tiny errors that grow in sums. Integer minor units (paise, fils) or decimal strings keep money exact. Tests should check the type and value match the contract.
Take a 15-question Python MCQ test with a score at the end, or practise the full Python question bank and earn points on the leaderboard.