Skip to content

Mighty replay buffer

mighty.mighty_replay.mighty_replay_buffer #

Mighty replay buffer.

MightyReplay #

MightyReplay(
    capacity, keep_infos=False, flatten_infos=False
)

Bases: MightyBuffer

Simple replay buffer.

:param capacity: Buffer size :param random_seed: Seed for sampling :param keep_infos: Keep the extra info dict. Required for some algorithms. :param flatten_infos: Make flat list from infos. Might be necessary, depending on info content. :return:

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
def __init__(self, capacity, keep_infos=False, flatten_infos=False):
    """Initialize Buffer.

    :param capacity: Buffer size
    :param random_seed: Seed for sampling
    :param keep_infos: Keep the extra info dict. Required for some algorithms.
    :param flatten_infos: Make flat list from infos.
        Might be necessary, depending on info content.
    :return:
    """
    self.capacity = capacity
    self.keep_infos = keep_infos
    self.flatten_infos = flatten_infos
    self.rng = np.random.default_rng()
    self.reset()

full property #

full

Check if the buffer is full.

add #

add(transition_batch, _)

Add transition(s).

:param transition_batch: Transition(s) to add :param metrics: Current metrics dict :return:

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
def add(self, transition_batch, _):
    """Add transition(s).

    :param transition_batch: Transition(s) to add
    :param metrics: Current metrics dict
    :return:
    """
    if not self.keep_infos:
        transition_batch.extra_info = []
    elif self.flatten_infos:
        transition_batch.extra_info = [
            list(flatten_infos(transition_batch.extra_info))
        ]

    self.index += transition_batch.size
    if len(self.obs) == 0:
        self.obs = transition_batch.observations
        self.next_obs = transition_batch.next_obs
        self.actions = transition_batch.actions
        self.rewards = transition_batch.rewards
        self.dones = transition_batch.dones
    else:
        self.obs = torch.cat((self.obs, transition_batch.observations))
        self.next_obs = torch.cat((self.next_obs, transition_batch.next_obs))
        self.actions = torch.cat((self.actions, transition_batch.actions))
        self.rewards = torch.cat((self.rewards, transition_batch.rewards))
        self.dones = torch.cat((self.dones, transition_batch.dones))
    if len(self) > self.capacity:
        self.obs = self.obs[len(self) - self.capacity :]
        self.next_obs = self.next_obs[len(self) - self.capacity :]
        self.actions = self.actions[len(self) - self.capacity :]
        self.rewards = self.rewards[len(self) - self.capacity :]
        self.dones = self.dones[len(self) - self.capacity :]
        self.index = self.capacity

reset #

reset()

Reset the buffer.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
146
147
148
149
150
151
152
153
def reset(self):
    """Reset the buffer."""
    self.obs = []
    self.next_obs = []
    self.actions = []
    self.rewards = []
    self.dones = []
    self.index = 0

sample #

sample(batch_size=32)

Sample transitions.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
135
136
137
138
139
140
141
142
143
144
def sample(self, batch_size=32):
    """Sample transitions."""
    batch_indices = self.rng.choice(np.arange(len(self)), size=batch_size)
    return TransitionBatch(
        self.obs[batch_indices],
        self.actions[batch_indices],
        self.rewards[batch_indices],
        self.next_obs[batch_indices],
        self.dones[batch_indices],
    )

save #

save(filename='buffer.pkl')

Save the buffer to a file.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
161
162
163
164
def save(self, filename="buffer.pkl"):
    """Save the buffer to a file."""
    with open(filename, "wb") as f:
        pickle.dump(self, f)

PrioritizedReplay #

PrioritizedReplay(
    capacity,
    alpha=1.0,
    beta=1.0,
    epsilon=0.0001,
    keep_infos=False,
    flatten_infos=False,
)

Bases: MightyReplay

Prioritized Replay Buffer.

:param capacity: Buffer size :param alpha: Priorization exponent :param beta: Bias exponent :param epsilon: Step size :param random_seed: Seed for sampling :param keep_infos: Keep the extra info dict. Required for some algorithms. :param flatten_infos: Make flat list from infos. Might be necessary, depending on info content. :return:

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
def __init__(
    self,
    capacity,
    alpha=1.0,
    beta=1.0,
    epsilon=1e-4,
    keep_infos=False,
    flatten_infos=False,
):
    """Initialize Buffer.

    :param capacity: Buffer size
    :param alpha: Priorization exponent
    :param beta: Bias exponent
    :param epsilon: Step size
    :param random_seed: Seed for sampling
    :param keep_infos: Keep the extra info dict. Required for some algorithms.
    :param flatten_infos: Make flat list from infos.
        Might be necessary, depending on info content.
    :return:
    """
    super().__init__(capacity, keep_infos, flatten_infos)
    self.alpha = alpha
    self.beta = beta
    self.epsilon = epsilon

full property #

full

Check if the buffer is full.

add #

add(transition_batch, metrics)

Add transition(s).

:param transition_batch: Transition(s) to add :param metrics: Current metrics dict :return:

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
def add(self, transition_batch, metrics):
    """Add transition(s).

    :param transition_batch: Transition(s) to add
    :param metrics: Current metrics dict
    :return:
    """
    super().add(transition_batch, metrics)
    advantage = metrics["td_error"]
    advantage = np.power(np.abs(advantage) + self.epsilon, self.alpha)
    if len(self.advantages) == 0:
        self.advantages = torch.from_numpy(advantage)
    else:
        self.advantages = torch.cat((self.advantages, torch.from_numpy(advantage)))
    while len(self.advantages) > self.capacity:
        self.advantages.pop(0)

reset #

reset()

Reset the buffer.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
214
215
216
217
def reset(self):
    """Reset the buffer."""
    super().reset()
    self.advantages = []

sample #

sample(batch_size=32)

Sample transitions.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
def sample(self, batch_size=32):
    """Sample transitions."""
    probabilities = np.array(self.advantages) / sum(self.advantages)
    sample_weights = np.power(probabilities * len(self), -self.beta)
    sample_weights /= sample_weights.max()
    normalizer = 1 / sum(sample_weights)
    sample_weights = np.array([x * normalizer for x in sample_weights])
    # Get rid of rounding errors
    sample_weights[-1] = max(0, 1 - np.sum(sample_weights[0:-1]))

    batch_indices = self.rng.choice(
        np.arange(len(self)), size=batch_size, p=sample_weights
    )
    return TransitionBatch(
        self.obs[batch_indices],
        self.actions[batch_indices],
        self.rewards[batch_indices],
        self.next_obs[batch_indices],
        self.dones[batch_indices],
    )

save #

save(filename='buffer.pkl')

Save the buffer to a file.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
161
162
163
164
def save(self, filename="buffer.pkl"):
    """Save the buffer to a file."""
    with open(filename, "wb") as f:
        pickle.dump(self, f)

TransitionBatch #

TransitionBatch(
    observations, actions, rewards, next_observations, dones
)

Transition batch.

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
def __init__(
    self, observations, actions, rewards, next_observations, dones
) -> None:
    """Initialize TransitionBatch."""
    if isinstance(rewards, float | int):
        observations = np.array([observations], dtype=np.float32)
        actions = np.array([actions], dtype=np.float32)
        rewards = np.array([rewards], dtype=np.float32)
        next_observations = np.array([next_observations], dtype=np.float32)
        dones = np.array([dones], dtype=np.float32)
    if isinstance(rewards, np.ndarray):
        self.observations = torch.from_numpy(observations.astype(np.float32))
        self.actions = torch.from_numpy(actions.astype(np.float32))
        self.rewards = torch.from_numpy(rewards.astype(np.float32))
        self.next_obs = torch.from_numpy(next_observations.astype(np.float32))
        self.dones = torch.from_numpy(dones.astype(np.int64))
    else:
        self.observations = observations
        self.actions = actions
        self.rewards = rewards
        self.next_obs = next_observations
        self.dones = dones

size property #

size

Current buffer size.

flatten_infos #

flatten_infos(xs)

Transform info dict to flat list.

:param xs: info dict :return: flattened infos

Source code in mighty/mighty_replay/mighty_replay_buffer.py 
14
15
16
17
18
19
20
21
22
23
24
25
26
def flatten_infos(xs):
    """Transform info dict to flat list.

    :param xs: info dict
    :return: flattened infos
    """
    if isinstance(xs, dict):
        xs = list(xs.values())
    for x in xs:
        if isinstance(x, Iterable) and not isinstance(x, str | bytes):
            yield from flatten_infos(x)
        else:
            yield x