ReplayBuffer¶
python_motion_planning.local_planner.ddpg.ReplayBuffer
¶
Bases: object
Experience replay buffer to store the transitions.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
state_dim
|
int
|
state dimension |
required |
action_dim
|
int
|
action dimension |
required |
max_size
|
int
|
maximum replay buffer size |
required |
device
|
device
|
device to store the data |
required |
sample(batch_size)
¶
Sample a batch of transitions from the replay buffer.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
batch_size
|
int
|
batch size |
required |
Returns:
| Name | Type | Description |
|---|---|---|
batch_s |
Tensor
|
batch of states |
batch_a |
Tensor
|
batch of actions |
batch_r |
Tensor
|
batch of rewards |
batch_s_ |
Tensor
|
batch of next states |
batch_win |
Tensor
|
batch of win or otherwise, True: win (reached the goal), False: otherwise. |
store(s, a, r, s_, win)
¶
Store a new transition in the replay buffer.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
s
|
Tensor
|
state |
required |
a
|
Tensor
|
action |
required |
r
|
Tensor
|
reward |
required |
s_
|
Tensor
|
next state |
required |
win
|
bool
|
win or otherwise, True: win (reached the goal), False: otherwise. |
required |