replay_buffer¶
- class replay_buffer.ReplayBuffer(buffer_size: int)¶
The ReplayBuffer class. Models a fixed size replay buffer. The buffer is represented by using a deque from Python’s built-in collections library. This is basically a list that we can set a maximum size. If we try to add a new element whilst the list is already full, it will remove the first item in the list and add the new item to the end of the list. Hence new experiences replace the oldest experiences. The experiences themselves are tuples of (state1, reward, action, state2, done) that we append to the replay deque and they are represented via the named tuple ExperienceTuple
- __getitem__(name_attr: str) List¶
Return the full batch of the name_attr attribute
- Parameters
name_attr (The name of the attribute to collect the) –
values (batch) –
- Return type
A list
- __init__(buffer_size: int)¶
Constructor
- Parameters
buffer_size (The maximum capacity of the buffer) –
- __len__() int¶
Return the current size of the internal memory.
- add(state: Any, action: Any, reward: Any, next_state: Any, done: Any, info: dict = {}) None¶
Add a new experience tuple in the buffer
- Parameters
state (The current state) –
action (The action taken) –
reward (The reward observed) –
next_state (The next state observed) –
done (Whether the episode is done) –
info (Any other info needed) –
- Return type
None
- get_item_as_torch_tensor(name_attr: str) Tensor¶
Returns a torch.Tensor representation of the the named item
- Parameters
name_attr (The name of the attribute) –
- Return type
An instance of torch.Tensor
- reinitialize() None¶
Reinitialize the internal buffer
- Return type
None
- sample(batch_size: int) List[ExperienceTuple]¶
Randomly sample a batch of experiences from memory.
- Parameters
batch_size (The batch size we want to sample) –
- Return type
A list of ExperienceTuple