Collatz Conjecture Machine Learning
9 September, 2020
0
0
0
Contributors
The github import option is not working for me so I am attaching the link.
I have always been fascinated by the collatz conjecture. I can definitely not explain it as well as resources on the internet so I'm just going to assume familiarity with it. I wanted to try and see if Machine learning could sneak the secrets of it out. I created a python program that created a json dataset 1000 numbers at a time, another python program to transfer this dataset to a csv file to make it easier for the learning models. I then tried to train a model to use the number as an input and number of steps away from 1 as the output. However, the model had a very hard time learning this. When I looked into it, I found that machine learning models often have trouble learning periodic functions like if a function is even or odd and the collatz conjecture turn number is fairly periodic.
Some ideas on how to pursue it further: create a classification model instead of regression, use sine as a activation function, represent the data differently.
Please contact me with any questions.
numpy
python
keras
collatz conjecture
math