The Challenge to understand in depth the concept of Machine Learning and Neural Networks was the next thing presented to me I had to make sure my understanding was enough that I could fully implement the use of it within my idea. I started researching into these topics, I found that YouTube was a good place to start as many people have created tutorials and educational videos on the subject.
3Blue1Brown the YouTube channel provided the best basic understanding in the ins and outs of how Neural Networks specifically work. The video below is the first part in his series on the explanation of Neural Networking.
I found this a good starting point for a clear understanding of what was required and the higher level understanding of a neural network. I did find that it didn’t provide a clear understand on how to actually code a neural network, for this I turned to a book.
“Make your own neural network” by Tariq Rashid. This book covers the finer points of the concept as well as a clear way on how to implement a neural network in Python. I found this incredibly helpful. The only issue with Python is the lack of connection with web mediums it has.
I mentioned a Google project in a previous post called TensorFlow I decided now that I understood neural networks it would be worth taking a look at this to see if I could use it in my project. TensorFlow is an api develop in C++ by Google which has interfaces with Python by more importantly Java and JavaScript. This meant that I could look at easily deploying my idea to the web with an implementation either in Java or JavaScript. As my current understand of Java is better I decided to look into using TensorFlow to solve a simple problem to prove my understanding of how it works.
I put together a quick game of snake in P3 this allowed to to quickly prototype a game that I could use to demonstrate an implementation of TensorFlow.
This is the simple snake game that I created to demonstrate machine learning. I started to develop the inputs required for the snake game. I calculated it would need 8 inputs for the directions it could look, the score and time. so all together 10 inputs these would then come together to 4 outputs through a number of hidden layers. these outputs would be the change in direction that the snake has to keep it alive.
The brain that wanted to develop would have to know where the fruit was the edge of the game board and that colliding with itself were all bad. Now that I know how to implement machine learning I think it is time to complete the main project.
Comments