NumPy Homework Help by Real Human Python Experts

Array creation and manipulation

np.array, np.zeros, np.ones, np.arange, np.linspace, np.eye, plus indexing, slicing, fancy indexing with integer and boolean masks, and conditional element selection.

Broadcasting

The broadcasting rules, dimension expansion with np.newaxis, common shape-mismatch fixes, and rewriting nested loops into broadcast-friendly operations.

Linear algebra

np.linalg.solve, np.linalg.inv, np.linalg.eig, np.linalg.svd, matrix multiplication with np.dot and the @ operator, determinants, and least-squares fitting.

Statistics and aggregation

sum, mean, median, std, var, percentiles, and correct use of the axis parameter so you aggregate the right dimension.

Random number generation

np.random.rand, np.random.randn, np.random.choice, np.random.seed for reproducible homework, plus the newer Generator API for modern courses.

Reshaping and transformation

reshape, transpose, ravel, flatten, swapaxes, expand_dims, and squeeze. Plus row-major vs column-major order when your course uses Fortran-style arrays.

Vectorization

Replacing Python for loops with NumPy operations. ufuncs, np.where, np.select, np.vectorize, and the trade-offs between each approach.

File I/O and serialization

np.save, np.load, np.savetxt, np.loadtxt, and np.genfromtxt for CSVs with missing values or mixed types.

Advanced NumPy

Structured arrays, masked arrays (np.ma), einsum for tensor contractions, np.fft for signal processing, np.polyfit for engineering courses, and memory mapping for datasets too large for RAM.

ValueError: shapes not aligned

The number-one NumPy error in student code. We diagnose the shape mismatch, explain the broadcasting rules, and rewrite the operation so the shapes align cleanly.

Python loops that need vectorizing

A 50-line loop becomes a 2-line NumPy operation. Typical speedup runs 100x to 500x on arrays larger than 10,000 elements.

MemoryError on large arrays

The default float64 dtype eats memory fast. We switch to float32 or int16 where the precision is enough, and the array footprint drops by 2x to 4x.

Confusion between views and copies

Slicing a NumPy array returns a view, not a copy. We add explicit .copy() calls where the assignment expects an independent array.

Reshape that loses or scrambles data

The order parameter (C vs F) controls how reshape walks the data. We pick the right order for your course conventions and verify the result against expected shapes.

IndexError on multi-dimensional arrays

Almost always an axis confusion. We rewrite the indexing with the correct axis, and add a short comment so the next assignment does not hit the same wall.

NaN propagation in computations

A single NaN poisons the entire result. We swap np.mean for np.nanmean, np.sum for np.nansum, and clean the array where the rubric expects no missing values.

Slow `np.linalg.solve` or unstable inverse

For ill-conditioned matrices, np.linalg.solve is preferred over computing the explicit inverse. We rewrite the linear-algebra step for numerical stability.

Courses we see most often

• DATA 100: Principles and Techniques of Data Science (UC Berkeley)
• MIT 18.06: Linear Algebra (NumPy-based assignments)
• Stanford CS231n: Convolutional Neural Networks
• CMU 10-601: Introduction to Machine Learning
• CSE 163: Intermediate Data Programming (UW Seattle)
• MIT 6.0001: Introduction to Computer Science and Programming
• CS50P: Introduction to Programming with Python (Harvard)
• Engineering and physics intros (ENGR 101 and equivalents) across US, UK, EU, and Australia

Textbooks our experts work from

• Guide to NumPy by Travis Oliphant (the NumPy creator)
• Python for Data Analysis by Wes McKinney
• Numerical Python by Robert Johansson
• Elegant SciPy by Juan Nunez-Iglesias, Stefan van der Walt, and Harriet Dashnow
• Python Scientific Computing by Hans Petter Langtangen

If your course uses a textbook not listed here, send us the syllabus and we match the conventions inside.