Existing material review

[svenvanderburg]

Review

Goal

We want to see if we should turn our SICS Odissei lesson material into a carpentries lesson. We compare existing material for teaching about 'Introduction to applied machine learning with scikitlearn for researchers', similar to our SICS Odissei lesson material.

Introduction to Machine Learning with scikitlearn 😃

https://carpentries-incubator.github.io/machine-learning-novice-sklearn/

• More about understanding than applying Compared to our SICS-Odissei lesson this is a bit more about understanding machine learning than applying it with scikitlearn. For example, we first code up a least squares regression ourselves, before starting to use scikitlearn. I quite like this, but it limits you to cover less powerful machine learning methods.
• Neural networks with scikitlearn I would say the episode on neural networks is quite heavy on the theory of neural networks and would fit better in the deep learning lesson, where we then use keras to implement neural networks instead of scikitlearn. In scikitlearn it is just another estimator.
• Misses machine learning validation There is not much about validation of your approach, in general I think our focus on the machine learning workflow is very nice.
• Quite heavy on unsupervised learning 50% of the lesson is dimensionality reduction and clustering

Concluding

• Too much overlap to ignore There is a lot of overlap with our SICS-Odissei lesson, and the general aim is the same. It would thus make sense to bundle our forces instead of creating a parallel lesson with very similar aim.
• It would be a shame to throw away our material The SICS-Odissei material is actually pretty good. If I had to choose I would definitely pick our material over the novice-sklearn material. It feels more mature, there is more a red line in the material, and I think it is more complete. But of course it is just a collection of notebooks.
• So: best to contribute the best of our material to this repo and improve it!

Introduction to Machine Learning in Python 😞

https://carpentries-incubator.github.io/machine-learning-novice-python/

• Very heavy on understanding instead of applying All machine learning code is hard-coded in python from scratch. This is very good for understanding machine learning, but you can cover much less. I think this is usually not what our audience wants.

Machine learning for biologists 😞

https://carpentries-incubator.github.io/ml4bio-workshop/

• Mature Complete and mature lesson
• Uses specific ml4bio software The lesson evolves around ml4bio, a not very generic GUI-based software.

Intro to AI for GLAM 😞

https://carpentries-incubator.github.io/machine-learning-librarians-archivists/

• Not mature and more conceptual Material for less than an hour, more on a very conceptual level, understanding the main concepts in ML for librarians and archivists.

Responsible Machine Learning in Python 😐

https://carpentries-incubator.github.io/machine-learning-responsible-python/

• Nice lesson Very nice lesson, but:
• Topic is on responsible AI Topic is responsible AI, not how to actually do machine learning in python.

[CunliangGeng]

It would cost more time/resource to merge lessons than just incubating our own star🌟 lesson. Also, we can fully control the direction of our own lesson, e.g. adapting it to meet new demands of our partners.

The ML novice sklearn lesson was started from 2013, with rare commits to it in the past 9 years(!!!), and now it's still in alpha version. It's hard to convince me that it's a popular or successful lesson. Then why should we invest in it?

[svenvanderburg]

@CunliangGeng those old commits are from the carpentries lesson template, development on the ML lesson has started in 2019 (see commits).

But you are right the development of the lesson is not super active, and it is in alpha for quite long now.

I decided to go ahead with porting our materials to carpentries-style lesson. But I also reached out to the contributors of the ML novice sklearn lesson. See more details in https://github.com/carpentries-incubator/machine-learning-novice-sklearn/issues/37