nature-of-code / noc-s17-2-intelligence-learning Goto Github PK

Should I bother with this or just start the first week with GA or KNN or something like that? The graph stuff is interesting with steering agents probably and also somewhat helps us think about network graph data structures as foundation for neural networks?

Better to do earlier somehow due to connection to steering behaviors?

I love When I see YouTubers collaborate on scientific issues to bring the best explanation but what happen that made you remove Siraj's video references.

https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

Also these visuals are helpful!

https://docs.gimp.org/en/plug-in-convmatrix.html

Also:

http://scs.ryerson.ca/~aharley/vis/conv/flat.html

Thanks @shekit!

You mentioned the genetic image program in the genetic TSP problem, I wrote a Processing.org sketch that does that :) You only need to download the G4P library for the basic GUI. I occasionally improve the code, but it runs fine just as it is.
https://github.com/obackhoff/Genetic_Image

As a suggestion, for genetic algorithms, I would add different levels of mutation that could occur.. sometimes drastic change, sometimes subtle. Also, I would keep the n most fit objects of the population, so the population does not diverge from the current best fit and turns into all random. Cheers for the great content.

DNA line 15:
function DNA(total, order) { ... }

Sketch line 30:
population[i] = new DNA(totalCities);

The constructor function has 2 parameters and, when you create the object, only pass one.

For the week on images!

http://cs.stanford.edu/people/karpathy/densecap/

https://www.floydhub.com/

Just watched your video and found interesting to improve your CrossOver function.
https://github.com/lebionick/CrossOver
here's the link for my git, it is in Java and there's some excess code, but I just want to show my idea:)
Thanks for your videos, they are really fun and cognitive!

1. Intro and Search
2. GA 1
3. GA 2
4. Classification and Regression w/o ANN
5. ANN for Classification and Regression
6. Intro to Python, Keras/Tensorflow
7. Midterm Project
8. CNN
9. RNN
10. GANs, auto-encoders, and more. . .
11. Reinforcement Learning
12. Neuro-Evolution
13. User Testing
14. Final Projects

• Binary Tree Example
• Breadth-first Search
• Dijkstra's algorithm
• More creative versions of above with data / animation
• A*
• A* solve maze
• TSP Brute Force
• Add steering agents

RNN and Convolutional in one week?

would this improve any speed/performance in particular with the RNN example of suggestions while typing?

(Probably just moving to keras.js or other non-webserver based system would be best?)

1. Binary Tree is I could just use text and sort alphabetically as a demonstration.
2. What are some good data sources to build graphs to demonstrate BFS, Dijkstra and/or TSP? Might be more interesting than just random graphs.
3. A star I have my maze solving example already which will do.

Based on a combination of Coding Challenges, I created the attached sketch with two sets of evolving steering agents: prey that steer towards food and avoid predators, and predators that steer towards prey. It need more tweaking of the parameters to get a stable system though.

PredatorPrey.zip

Thank you @rhacking for the new FDG layout for BFS! I'd like to simplify the code a bit to make it more readable for beginners as well as use p5.Vector to tie into other Nature of Code physics examples.

Can I fit tSNE into KNN, Regression week?

To be investigated

https://hackernoon.com/how-do-gans-intuitively-work-2dda07f247a1

• Add comments to code.
• Add links to all the code examples running in the browser via github pages to week 1 README
• Write up week 1 homework wiki

https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/
https://github.com/Kiri23/flask/blob/master/label_image.py
http://cs.stanford.edu/people/karpathy/densecap/
http://cs.stanford.edu/people/karpathy/convnetjs/
http://www.patrickhebron.com/learning-machines/week8.html

Edit: See #10

• Decision Trees / Random Forests
• Sort of related to above: my "animal guessing game" example: https://github.com/shiffman/AnimalGuessingGame
• tSNE

I've added a glossary list for machine learning and statistics in the wiki. The aim is to make it easier to get an overview than a bunch of wikipedia links. Also, I hope this will inspire others to add similar pages for other topics :-)

Hi,

Just in case anyone is interested, I have few reference projects which i have slowly been putting together for the past few months that I am using as reference material for myself and anyone else who might find them useful. The projects are written in JavaScript (specifically TypeScript) and are well documented, you can find them here.

• neuron pure javascript implementation of a multi layer perceptron network small demo
• tesseract webgl 2.0 GPGPU compute library.
• neuron-gpgpu neuron ported to the GPU (using tesseract) which looks solving networks using compute parallelism.

This is in reference to video 10.4: Neural Networks: Multilayer Perceptron Part 1 - The Nature of Code. Happy to discuss these implementations in this thread if anyone finds them useful, or has general questions. I hope they serve as a good reference for anyone wanting to learn some of the pure programming aspect of ML.

Thanks again for the videos on machine learning !

First week? Second week?

At some point, I would like to make a version of the recommendation system example using the large MovieLens dataset and pre-computing item relationships for fast recommendations. (See Programming Collective Intelligence). The data is in the week 2 examples, but I didn't have a chance to build it yet.

QLearning JS demos

http://cs.stanford.edu/people/karpathy/reinforcejs/waterworld.html

Siraj's policy gradient video:
https://www.youtube.com/watch?v=PDbXPBwOavc

Another topic not on the syllabus that could fit with week 3 and/or be part of a 14 week version of this course.

https://generalabstractnonsense.com/2017/03/A-quick-look-at-Support-Vector-Machines/