Jehoshaphat I. Abu
std::steam

std::steam

100 Days Of ML Code — Day 026

100 Days Of ML Code — Day 026

Jehoshaphat I. Abu's photo
Jehoshaphat I. Abu
·Aug 4, 2018·

2 min read

Recap from Day 025

In day 025, we continued with how machine learning in the arts is different. We saw that for lots of creative applications, making changes to the data is often the most efficient and easy to understand way for you to improve your models, bringing them more into line with your goals for whatever it is you’re building.

Today we will continue from where we stopped in day 025.

Training Sets May Be Very Small.

“If you’re the one creating the training data for your algorithm, it may be difficult for you to create massive training sets of thousands or millions of examples. There’s a lot of hype right now about big data. Having such large data sets can be a godsend for many applications. Big data can make it possible to build much better models leading to much more accurate predictions or even new discoveries about the world.

Big Date — [Source](https://cdn.hashnode.com/res/hashnode/image/upload/v1632827312382/d0E13z7ny.jpeg)Big Date — Source

They are a few big data sets out there in the arts, for example, collections of audio or text related to classical or popular music. But it’s less likely that you’ll find big data sets of gestures made by musicians or dancers, and pretty certain you won’t find a big data set of exactly the type of gesture you want to use in a customized musical instrument or installation, and that’s fine.

[Source](https://cdn.hashnode.com/res/hashnode/image/upload/v1632827314778/_zvRNEaP1.octet-stream)Source

As you’ve seen, many algorithms allow us to build good models from just a few examples, and again, these models are often suitable for the type of goals we have in creating systems for artistic or musical expression. More data wouldn't necessarily help us build a better game track controlled blowtar, for instance.

Now, when you’re using the training data to try to push your model toward particular behaviors, you may find that you prefer learning algorithms that are inclined to over-fit to your training examples.

[Source](https://cdn.hashnode.com/res/hashnode/image/upload/v1632827318465/To63E2XOU.html)Source

In most machine learning applications, you don’t want an algorithm that draws unnecessarily complicated decision boundaries. However, if you know that you have adequate features for the problem you’re trying to learn and you don’t have much noise, you may want an algorithm that’s like nearest neighbor that’s happy to fit very complicated models to small training sets, because this means that you can use very small numbers of examples to nudge your model in a useful direction.”

Wow! You made it to the end of day 026. I hope you found this informative. Thank you for taking time out of your schedule and allowing me to be your guide on this journey.

Reference

https://www.kadenze.com/courses/machine-learning-for-musicians-and-artists-v/sessions/developing-a-practice-with-machine-learning-wrap-up

 
Share this