Question How many parameters my model have

This a silly question. Every machine-learned model likes to brag about their parameter count. And I started to wonder what is this number for ML agents. Is it a fixed amount or I can train it long enough to reach billions of them? My guess is that the number of parameters in my model will be the number of observations that I feed to the system. Curious If I am right or wrong

 

You decide it in your config. It's a function of the number of layers, and the size of each layer.

The parameters doesn't change as you train it.

Think of the number of parameters as the size of your brain. It doesn't change much as you study mlagents, Unity etc. But the brain that you have becomes smarter and more knowledgeable over time, as you read more, and try more things.

Inside your brain are lots of cells called neurons, connected together somehow. As you learn, the connections grow and change.

This is exactly what happens in a neural network that you train using mlagents. The number of neurons stays the same, but the connections between the neurons change.

 

Interesting, thank you!