NumPy ndarray — Complete Beginner to Interview Guide

1. What is NumPy?

Points to understand first:

1. NumPy stands for Numerical Python.
2. It is a third-party library — you install it with pip install numpy before using it.
3. Its main feature is the ndarray (n-dimensional array) — a fast, memory-efficient, multi-dimensional array.
4. Unlike a Python list, a NumPy array stores only one data type, as raw contiguous memory — this is what makes it so fast.
5. NumPy lets you do math on an entire array at once, without writing a manual loop — this is called vectorization.
6. Almost every major Python data-science and machine-learning library (pandas, scikit-learn, PyTorch) is built on top of NumPy.

2. Why is NumPy Faster Than a Python List?

3. Key Attributes of an Array

4. Creating Arrays — Every Way, One by One

4.1 np.array() — From a Python List

4.2 np.zeros() — An Array Full of Zeros

4.3 np.ones() — An Array Full of Ones

4.4 np.full() — An Array Filled With a Chosen Value

4.5 np.eye() — An Identity Matrix

4.6 np.arange() — Like Python’s range(), but Returns an Array

4.7 np.linspace() — Evenly Spaced Numbers

4.8 np.random.rand() and np.random.randint() — Random Arrays

4.9 np.zeros_like() — Same Shape as Another Array

5. Indexing and Slicing

5.1 Basic Indexing (1-D Array)

5.2 Indexing a 2-D Array (Matrix)

5.3 Boolean Masking — Picking Items by Condition

Points:

1. You can compare a whole array to a value — this gives back an array of True/False.
2. You can then use that True/False array to filter the original array.
3. This is one of the most useful NumPy tricks, used constantly in real code and interviews.

5.4 Fancy Indexing — Picking Multiple Specific Positions

6. Views vs Copies — An Important Beginner Trap

Points:

1. Basic slicing (a[1:3]) gives you a view — it shares the same memory as the original array.
2. Changing a view also changes the original array!
3. Use .copy() when you need a completely independent array.

7. Reshaping and Combining Arrays

7.1 reshape() — Change the Shape Without Changing the Data

7.2 flatten() vs ravel() — Back to 1-D

7.3 T — Transpose (Swap Rows and Columns)

7.4 np.concatenate(), np.vstack(), np.hstack() — Combining Arrays

8. Vectorized Math — The Whole Point of NumPy

Points:

1. “Vectorized” means you can apply an operation to every element at once, with no explicit loop.
2. This is both shorter to write and much faster to run than a Python for loop.

9. Broadcasting — Combining Different Shapes

Broadcasting is how NumPy applies an operation between two arrays of different shapes, without you writing a loop.

The rule (simplified):

1. Compare the shapes of the two arrays, starting from the right side.
2. Two dimensions are compatible if they are equal, or if one of them is 1.
3. NumPy “stretches” the smaller dimension to match — without actually copying data in memory.

10. Every Common Method — Explained One by One

10.1 np.sort() / a.sort() — Sorting

10.2 argsort() — Get the Sorted Indices

Why this matters: sometimes you need to know where the sorted values came from, not just the sorted values themselves.

10.3 sum(), mean(), std(), var()

10.4 min(), max(), argmin(), argmax()

10.5 np.unique() — Unique Values, Sorted

10.6 np.where() — Conditional Selection

10.7 cumsum() — Running (Cumulative) Sum

10.8 any() and all()

10.9 astype() — Convert to a Different Data Type

10.10 sum(axis=...) — Summing Along Rows or Columns

Points:

1. For 2-D arrays, axis=0 means “go down the columns.”
2. axis=1 means “go across the rows.”
3. This confuses almost every beginner at first — the diagram below should help.

11. All Methods in One Table (Quick Reference)

12. A Worked Example Combining Several Concepts

13. Interview Patterns (Question First, Then Solution)

13.1 Pattern: Check if an Array is Sorted (Vectorized)

❓ Interview Question:

1. Problem: Given a NumPy array, check whether it is sorted in ascending order — without writing a manual loop.
2. Example:
   • Input: [1, 2, 2, 5]
   • Expected Output: True
   • Input: [1, 5, 2]
   • Expected Output: False
3. Hint: Compare the array to a “shifted” version of itself.

✅ Approach:

1. Compare every item (except the last) to the item right after it.
2. If every item is less than or equal to its neighbor, the array is sorted.
3. np.all() checks that this is true for the entire array at once.

Complexity: Time = O(n), Space = O(n) for the comparison array.

13.2 Pattern: Find All Pairs That Sum to a Target (Vectorized Filtering)

❓ Interview Question:

1. Problem: Given an array of numbers and a target sum, find how many numbers in the array are needed to reach the target when paired with a specific value — here, count how many values are “positive after subtracting the target.”
2. Example:
   • Input: nums = [1, 2, 3, 4, 5], target = 3
   • Expected Output: values greater than target → [4, 5]
3. Hint: Boolean masking turns a filtering problem into one line.

✅ Approach:

1. Build a boolean mask comparing the whole array to the target at once.
2. Use that mask to select only the matching values.

Complexity: Time = O(n), Space = O(k) where k is the number of matches.

13.3 Pattern: Matrix Row and Column Sums (Common in Interviews)

❓ Interview Question:

1. Problem: Given a 2-D matrix, find the sum of each row and the sum of each column.
2. Example:
   • Input: [[1, 2, 3], [4, 5, 6]]
   • Expected Output: row sums [6, 15], column sums [5, 7, 9]
3. Hint: This is exactly what the axis parameter is for.

✅ Approach:

1. axis=1 sums across each row.
2. axis=0 sums down each column.

Complexity: Time = O(rows × columns), Space = O(rows + columns).

14. NumPy vs list vs array — When to Use Which

15. Practice Problems (Try These Yourself)

1. Given an array of exam scores, find how many students scored above the average.
2. Write a function that reverses each row of a 2-D matrix.
3. Given two arrays of the same shape, compute the elementwise maximum.
4. Find the index of the closest value to a target number in an array.
5. Compute the moving average of a 1-D array with a given window size.

16. Key Takeaways

1. numpy.ndarray stores one fixed data type, contiguously in memory — this is what makes it fast.
2. Vectorized operations (a + b, a ** 2, np.sum(a)) replace manual loops and run much faster.
3. Basic slicing returns a view, not a copy — use .copy() when you need independence.
4. Boolean masking (a[a > 0]) and fancy indexing (a[[0, 2, 4]]) are the two most useful indexing tricks.
5. Broadcasting lets you combine arrays of different shapes without writing loops, following the “match or 1” rule.
6. axis=0 means down the columns, axis=1 means across the rows — a common beginner confusion worth memorizing.
7. argsort, cumsum, where, and unique show up constantly in interview questions about arrays.

10. Every Important Method — Explained One by One

10.1 np.sort() — Sort the Array

10.2 argsort() — Get the Indexes That Would Sort the Array

This one confuses beginners — here is the difference from sort():

1. sort() gives you the sorted values.
2. argsort() gives you the indexes that would put the array in sorted order.

10.3 sum(), mean(), std(), var()

10.4 min(), max(), argmin(), argmax()

10.5 np.unique() — Get Unique Values, Sorted

10.6 np.where() — Choose a Value Based on a Condition

10.7 cumsum() — Running (Cumulative) Total

10.8 any() and all()

10.9 astype() — Convert to a Different Data Type

10.10 sum(axis=...) — Summing Along Rows or Columns

This trips up a lot of beginners with 2-D arrays:

1. axis=0 means “go down the columns” — sum each column.
2. axis=1 means “go across the rows” — sum each row.

11. All Methods in One Table (Quick Reference)

12. Interview Patterns (Question First, Then Solution)

12.1 Pattern: Normalize a List of Scores (Min-Max Scaling)

❓ Interview Question:

1. Problem: Given an array of numbers, rescale every value into the range 0 to 1.
2. Example:
   • Input: [65, 90, 72, 88, 55]
   • Expected Output: values between 0.0 and 1.0, where the smallest input becomes 0.0 and the largest becomes 1.0.
3. Hint: Subtract the minimum, then divide by the range (max - min).

✅ Approach:

1. Find the minimum and maximum values of the array.
2. For every value: subtract the minimum (shifts the range to start at 0).
3. Divide by (max - min) (scales the range down to fit between 0 and 1).

Complexity: Time = O(n), Space = O(n) for the result array.

12.2 Pattern: Distance Between Two Points

❓ Interview Question:

1. Problem: Given two points as arrays (e.g. [x, y]), calculate the straight-line (Euclidean) distance between them.
2. Example:
   • Input: p1 = [0, 0], p2 = [3, 4]
   • Expected Output: 5.0
3. Hint: This is the same formula as the Pythagorean theorem: square the differences, add them, take the square root.

✅ Approach:

1. Subtract the two points to get the difference in each dimension.
2. Square every difference.
3. Add the squares together, then take the square root of the total.

Complexity: Time = O(d) where d = number of dimensions, Space = O(1).

12.3 Pattern: Check if an Array is Sorted (Vectorized)

❓ Interview Question:

1. Problem: Given an array, check whether it is sorted in ascending order — without writing a manual loop.
2. Example:
   • Input: [1, 2, 2, 5]
   • Expected Output: True
   • Input: [1, 5, 2]
   • Expected Output: False
3. Hint: Compare the array to a “shifted” version of itself.

✅ Approach:

1. Compare every element (except the last) to the element right after it.
2. If every one of those comparisons is True (each item is less-than-or-equal to the next), the array is sorted.
3. np.all() checks that every comparison passed.

Complexity: Time = O(n), Space = O(n) for the comparison array.

12.4 Pattern: Find All Pairs That Add Up to a Target (Using a Set)

❓ Interview Question:

1. Problem: Given an array of numbers and a target sum, find all pairs of numbers that add up to the target.
2. Example:
   • Input: nums = [2, 7, 11, 15, -2], target = 9
   • Expected Output: [(2, 7)]
3. Hint: For each number, check whether target - number has already been seen.

✅ Approach:

1. Keep a set of numbers seen so far.
2. For each number, calculate what its “partner” would need to be (target - number).
3. If that partner is already in the seen set, you found a pair.
4. Add the current number to the seen set, then move on.

Complexity: Time = O(n), Space = O(n) for the seen set.

13. Practice Problems (Try These Yourself)

1. Given a 2-D array, find the sum of its diagonal elements.
2. Given an array of temperatures, count how many days were above the average.
3. Given two arrays of the same shape, compute their element-wise product (not the dot product).
4. Reverse a 2-D array both horizontally and vertically.
5. Find the index of the closest value to a given number in an array.

14. NumPy vs list vs array — When to Actually Use It

15. Key Takeaways

1. numpy.ndarray stores one fixed data type as contiguous memory, enabling fast, vectorized math with no manual loops.
2. shape, ndim, size, and dtype are the four attributes you’ll check constantly.
3. Basic slicing returns a view that shares memory with the original — use .copy() when you need independence.
4. Boolean masking (a[a > 0]) and fancy indexing (a[[0, 2, 4]]) are the two indexing tricks used everywhere.
5. Broadcasting lets you combine different shapes without loops — compare shapes from the right; dimensions must match, or one of them must be 1.
6. For interviews: know argsort, cumsum, where, unique, and axis=0 vs axis=1 — they appear constantly in array-manipulation questions.