C2W2 softmax and multiclass labs dataset generation

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

In C2W2 labs Softmax and Multiclass we create a dataset by using make_blob

CleanShot 2025-05-19 at 13.52.10@2x
CleanShot 2025-05-19 at 13.52.10@2x1982×390 48.1 KB

Until now, we’ve been defining the dataset by giving X,y value pairs or e.g pictures representing zeroes and ones. Here we give ‘centers’ instead and generate a dataset. What are ‘centers’ and how is the dataset generated?

What does [[-5, 2], [-2, -2], [1, 2], [5, -2]] represent?

Thank you!

2

Hi @Delacraft

This is for creating toy datasets. This function will generate 2000 samples of data, around each of the provided centers with a standard deviation of 1.0 (and random state of 30). Each point in the dataset is generated by sampling from a Gaussian (normal) distribution centered at one of these coordinates.

The below image is an example when you provide 3 centers (e.g. [-10, 8], [6, 8], and [8, -3]) with a smaller std.

image
image932×606 68.7 KB

Hope it helps! Feel free to ask if you need further assistance.

1 Like
3

Thanks @Alireza_Saei ! What is a random state? How random is 30?

Here we generate this example set because we are learning about ML and I assume it’s an easy way of getting a dataset, without having to collect the data, label it etc. Outside of classroom, in what cases do we generate these kind of artificial datasets? What are they used for?

4

You’re welcome! happy to help :raised_hands:

Random state is just a seed number that controls randomness and using 30 or 42 or … ensures reproducibility (you get the same random data every time).

Exactly.

Outside class, we use synthetic datasets to:

  • Test/debug models quickly
  • Benchmark algorithms
  • Simulate rare or controlled scenarios when real data is hard to get

Hope it helps!

1 Like