20% of NumPy Functions that Data Scientists use 80% of the Time

Plus a free NumPy CheatSheet from DataCamp

Image created by Author

As a data scientist who’s worked on over 30 projects of different sizes, scopes, and purposes… I’ll be real with you.

You may rarely use NumPy ‘directly’ for the major parts of your workflow, instead you’d mostly be leveraging other libraries that have been built on NumPy such as Pandas, MatplotLib, and Scikit-Learn.

I know, I know what you’re thinking. But hear me out.

Despite not using NumPy functions explicitly in every project, having a solid understanding of NumPy can be a real time saver.

Many underlying operations in other libraries are implemented using NumPy so understanding how to work with arrays efficiently will make you a better data scientist cos you’ll be able to optimize your code and improve your model’s performance effortlessly.

Okay, okay, this isn’t some ‘justice for NumPy’ post; it’s more like a little list I’ve compiled that contains some NumPy functions I find myself reaching for frequently in my data science work.

But before we get to it…

What is NumPy?

NumPy which stands for Numerical Python is a top Python library that’s used for working with arrays (vectors and matrices — which form the very basis of the datasets you’ll be working with).

For more info about NumPy, you can visit the NumPy Documentation.

While NumPy provides a whole lot of functions that you can use in your workflow, I often substitute a handful of them with Pandas and Matplotlib for more common tasks.

For instance, instead of using np.mean, np.std, etc., I find it more intuitive to employ Pandas' describe() for a comprehensive summary.

So in your own work process, you may find that you mostly need to consult NumPy functions when you need a bit more fine-tuning or when working on operations that demand a lower-level approach.

To use the NumPy library, first ensure you’ve imported it into your code like this:

import numpy as np

For the sake of this blog, assume we have loaded a dataset (of 10 rows) into our code as data that contains:

• a numerical column: numeric_column
• another numerical column: numeric_column1

It’s time to bake the biscuits!

(In Ernie’s voice)

1. np.unique(): Identifying unique values in one or more numerical columns of a dataset.

unique_nums = np.unique(data[['numeric_column', 'numeric_column1']])

2. np.array(): For converting a column in a dataset into a NumPy array for numerical operations.

num_array = np.array(data['numeric_column'])

3. np.arange(): For generating a sequence of numbers for an index or time steps. i.e. Creating an array representing time intervals for time series analysis.

# Generate a sequence of time intervals (e.g., days, hours, etc.)
time_intervals = np.arange(1, 366, 7) 

4. np.reshape() : For preparing data for input into machine learning models. I.e. reshaping a 1D array of pixel values into a 2D array for image processing.

# Reshape a 1D array into a 2D array (e.g., representing a 2x5 image)
image_2d = np.reshape(data['numeric_column'], (2, 5))

5. np.ceil(): Rounding up elements to the nearest integer.

rounded_up_nums = np.ceil(data['numeric_column'])

6. np.floor(): Rounding down elements to the nearest integer.

rounded_down_nums = np.floor(data['numeric_column'])

7. np.exp(): For exponential transformation for feature engineering.

exp_trans_data = np.exp(data['numeric_column'])

8. np.log(): For calculating the natural logarithm for proportional relationships. I.e. Transforming data to achieve linear relationships for regression analysis.

log_trans_data = np.log(data['numeric_column'])

9. np.power(): Raising elements to a specified power. I.e. Applying power transformations to features in a machine learning model.

power_trans_data = np.power(data['numeric_column'], 2)

10. np.sqrt(): For calculating square roots for scaling purposes. I.e. Scaling data using square root transformations for better model performance.

sqrt_trans_array = np.sqrt(data['numeric_column'])

And finally as promised…

Your Free NumPy Cheatsheet from DataCamp → link.

Call-to-Action

Please leave as many claps as you please (up to 50) if you enjoyed this article and let me know in the comments what other NumPy functions you use regularly.

I’ll be very happy if you

Follow GDSC Babcock DataVerse for more data-related articles.

Bye for now :)