Homework 1

Note: The code can be found in this repository.

Problem 2.2

For both experiments:

1. Hidden layer = 128 units
2. Training Data Size = 100 rollouts = 100,000 timesteps
3. Epochs = 500
4. Test Data Size = 20 rollouts

Hopper

	Expert Policy	Trained Model
**Mean**	3778.79126	3779.27708
**Standard Deviation**	3.03886	3.07484

Humanoid

	Expert Policy	Trained Model
**Mean**	10306.80848	407.63263
**Standard Deviation**	979.44124	23.56048

Problem 2.3

The data size was varied and the rewards observed:

As the data size increases, the mean reward initially increases up to a certain point and then remains constant. Rationale: The size of data is one of the most important parameter of a model. If the data is not enough, the model will never be able to generalize well to the entire distribution from which the data is being drawn.