Scikit or NumPy for Machine Learning

Course Q&A Machine Learning Specialization Supervised ML: Regression and Classification
21

From your overall experience, what would you recommend?

22

All of them.

23

Thanks.
For someone who wants to eventually focus on computer vision, what would you recommend as a path to follow

24

I suggest to complete DLS specialization first, then you can take or audit CS231n. Videos from previous offerings of the course are available on YouTube.

25

The DLS specialization includes Course 4 on CNN’s (Convolutional Neural Networks). These are a common tool for visual analysis.

26

Thanks for this

27

Thanks you for the information

28

@AIChef in my quick assessment, you can think of Numpy as all the little contained cells of, say, an arm muscle-- Scikit-learn (or similar) is what does all the ‘heavy lifting’, so to speak (or more like the mind controlled by a ‘brain’). Yet neither is separate.

1 Like
29

How can I visualize a multilinear variable?

30

If it has more than three dimensions, you can’t.

31

so what do you do?

32

Rely on the cost history during training.
Try a lot of different models to get the lowest possible cost.

33

@TMosh mentions an important point-- Maybe you can ‘kind of’ get there if you do a ‘heat map’; But the better lesson you’ll learn through these courses (especially DLS) is that is exactly why we do SGD, or ADAM or your optimizer of your choice.

We just ‘can’t see it’, and this comes not only from the visual side, as you express, but also the mathematical one.

Otherwise we’d ‘see’ the optimal solution and the day is done (*I mean by this back-prop is a process, not just one single equation). Instead we’ve found you kind of have to ‘poke’ at it to make the equations work.

34

Okay, so how do i visualize 3 dimensions

35

A 3D plot seems useful.

36

Based on the conversation so far, I’d recommend @AIChef the following:

  1. If you can’t find an interesting dataset on Kaggle, then you’d have to create it yourself. This involves the data acquisition or collection part of the data life cycle.
  2. About platforms, Google Colab uses the Compute Engine behind the scenes, which is a virtual machine (VM). If you want to create a serious ML project, especially if you intend to focus on computer vision, then you’d have to use one of the 3 major cloud providers – AWS, Microsoft Azure or Google Cloud. I’ve provided the links of their free tiers, which last for one year. In my experience, the best way to get started is to study for their introductory certifications. For example, I passed the AZ-900: Azure Fundamentals exam ~3 months ago.
  3. Once you have decided the environment (Colab or a cloud provider) and the question to answer, which is the goal of the project, you can focus on the tools and algorithms to use. As it is an iterative process, you most likely would have to update or change your approach, be it the data, the methods or both.
37

thanks