def convert_action2one_hot(self, a_counts):
'''
用于在训练前将buffer中的离散动作的索引转换为one_hot类型
'''
if 'a' in self.buffer.keys():
self.buffer['a'] = [
sth.int2one_hot(a.astype(np.int32), a_counts)
for a in self.buffer['a']
]
def get_transitions(self, databuffer, data_name_list=['s', 'a', 'r', 's_', 'done']):
'''
TODO: Annotation
'''
data = databuffer.sample() # 经验池取数据
if not self.is_continuous and 'a' in data_name_list:
a_idx = data_name_list.index('a')
a = data[a_idx].astype(np.int32)
pre_shape = a.shape
a = a.reshape(-1)
a = sth.int2one_hot(a, self.a_counts)
a = a.reshape(pre_shape+(-1,))
data[a_idx] = a
return dict([
[n, d] for n, d in zip(data_name_list, list(map(self.data_convert, data)))
])
def get_transitions(self,
data_name_list=[
's', 'visual_s', 'a', 'r', 's_', 'visual_s_',
'done'
]):
'''
TODO: Annotation
'''
data = self.data.sample() # 经验池取数据
if not self.is_continuous and 'a' in data_name_list:
a_idx = data_name_list.index('a')
data[a_idx] = sth.int2one_hot(data[a_idx].astype(np.int32),
self.a_dim)
return dict([[n, d] for n, d in zip(
data_name_list, list(map(self.data_convert, data)))])