Hyper Parameters

Q-learning Keras Hyperparameters

A model always has a certain amount of input, output and a way to transform this input to output. The little knobs we can turn to change that transformation are called hyperparameters, our environment has the following:

An overview of the options in our model

Math Symbol	Variable	Description	Default	Possible values	Type
-	inputs	The input size	2	> 0	int
-	outputs	The output size	4	> 0	int
-	learning_rate	The initial step size of our model	1e-3	> 0	float
-	memory	How much our network can remember	2k	> 0	int
γ	gamma	How important an action is to us in the future	0.95	0 - 1	float
𝜖	epsilon	The chance of our network taking a random actions instead of a prediction	1.0	0 - 1	float
𝜖	epsilon_low	(EPSILON lower bound) How liberal our network is once it's learned patterns	1e-2	0 - <1	float
𝜖-greedy	epsilon_decay	How much EPSILON decreases by (EPSILON *= EPSILON_DECAY)	0.95	0 - <1	float
-	batch_size	How much memory it trains on	32	> 0	int