Mathematical ufuncs: Trigonometric, Exponential, Logarithmic

At the technical core of NumPy's blazing numerical speed lies the concept of Universal Functions (commonly abbreviated as ufuncs).

A ufunc is a function that operates on ndarrays in an element-by-element fashion. Unlike standard Python math functions from the math module (which accept only single scalar numbers and must be invoked sequentially inside loops), NumPy ufuncs are compiled C routines that support broadcasting, type coercion, and direct memory output targeting.

Click to Zoom

1. Trigonometric and Angle Routines

NumPy provides complete implementations of trigonometric functions. All trigonometric ufuncs expect angles expressed in radians (not degrees):

$$\text{radians} = \text{degrees} \times \frac{\pi}{180}$$

Output:

Inverse Trigonometric Functions and arctan2:

When computing angles from Cartesian coordinates $(x, y)$, standard arctan(y/x) loses quadrant information and risks division by zero. Always use np.arctan2(y, x):

2. Exponential and Logarithmic ufuncs

Logarithmic and exponential transformations are essential across data preprocessing (log-transforming skewed financial returns or heavy-tailed distributions):

Numerical Precision: log1p and expm1

When $x$ is near zero, evaluating $\ln(1 + x)$ using np.log(1 + x) suffers from catastrophic floating-point cancellation. NumPy provides np.log1p(x) and its inverse np.expm1(x):

3. High-Performance Optimization: The out Parameter

Every time you execute y = np.sin(x), NumPy allocates a brand new memory buffer for the result. When processing massive datasets (gigabytes of sensor streams), continuous memory allocations trigger garbage collection and waste bandwidth.

All ufuncs provide an optional out parameter to write results directly into an existing pre-allocated buffer:

Multiple Choice Questions

1. In what unit of measurement do trigonometric ufuncs like np.sin() and np.cos() expect input angles?

A. Degrees B. Radians C. Gradians D. Revolutions Answer: B Explanation: NumPy's trigonometric universal functions strictly expect angles in radians. Use np.deg2rad() or np.radians() to convert from degrees.

2. Why is np.arctan2(y, x) preferred over np.arctan(y / x) in computer vision and robotics?

A. np.arctan2 runs on GPUs B. np.arctan2 correctly identifies the full four-quadrant angle in [-pi, pi] and handles division by zero when x == 0 C. np.arctan only works with integer dtypes D. np.arctan2 converts the output to degrees automatically Answer: B Explanation: np.arctan2(y, x) takes signs of both arguments into account to determine the correct quadrant between $-pi$ and $+pi$, and cleanly avoids division-by-zero errors when $x = 0$.

3. What is the benefit of using np.log1p(x) instead of np.log(1 + x) when x is extremely small (e.g. 1e-15)?

A. It calculates base-10 logarithm instead of base-e B. It avoids catastrophic floating-point cancellation error when adding small numbers to 1.0 C. It rounds numbers to the nearest integer D. It accepts negative values without returning NaN Answer: B Explanation: Standard floating-point addition 1.0 + 1e-15 loses low-order bits due to precision limits. np.log1p(x) utilizes specialized Taylor series algorithms to retain precision for near-zero inputs.

4. What does the out parameter in a universal function like np.exp(x, out=buffer) accomplish?

A. It redirects printed outputs to a file B. It stores results into a pre-allocated memory array without allocating new heap memory C. It limits output values to positive numbers D. It flattens the output array Answer: B Explanation: Passing a pre-allocated array to out instructs the C loop to write results directly into that memory block, eliminating dynamic memory allocations and boosting performance.

5. Which NumPy ufunc computes the common base-10 logarithm?

A. np.ln() B. np.log() C. np.log10() D. np.logb(10) Answer: C Explanation: np.log() computes the natural (base $e$) logarithm, while np.log10() computes the common (base 10) logarithm.