bacow/easy-rl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

hot update A2C

Browse Source
This commit is contained in:
johnjim0816
2022-08-29 15:12:33 +08:00
parent 99a3c1afec
commit 0b0f7e857d
109 changed files with 8213 additions and 1658 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
projects/codes/Sarsa/sarsa.py
View File

@@ -5,7 +5,7 @@ Author: John
Email: johnjim0816@gmail.com
Date: 2021-03-12 16:58:16
LastEditor: John
LastEditTime: 2022-08-25 00:23:22
LastEditTime: 2022-08-25 21:26:08
Discription:
Environment:
'''
@@ -30,7 +30,7 @@ class Sarsa(object):
self.sample_count += 1
self.epsilon = self.epsilon_end + (self.epsilon_start - self.epsilon_end) * \
math.exp(-1. * self.sample_count / self.epsilon_decay) # The probability to select a random action, is is log decayed
best_action = np.argmax(self.Q_table[state])
best_action = np.argmax(self.Q_table[str(state)]) # array cannot be hashtable, thus convert to str
action_probs = np.ones(self.n_actions, dtype=float) * self.epsilon / self.n_actions
action_probs[best_action] += (1.0 - self.epsilon)
action = np.random.choice(np.arange(len(action_probs)), p=action_probs)
@@ -38,27 +38,27 @@ class Sarsa(object):
def predict_action(self,state):
''' predict action while testing
'''
action = np.argmax(self.Q_table[state])
action = np.argmax(self.Q_table[str(state)])
return action
def update(self, state, action, reward, next_state, next_action,done):
Q_predict = self.Q_table[state][action]
Q_predict = self.Q_table[str(state)][action]
if done:
Q_target = reward # terminal state
else:
Q_target = reward + self.gamma * self.Q_table[next_state][next_action] # the only difference from Q learning
self.Q_table[state][action] += self.lr * (Q_target - Q_predict)
Q_target = reward + self.gamma * self.Q_table[str(next_state)][next_action] # the only difference from Q learning
self.Q_table[str(state)][action] += self.lr * (Q_target - Q_predict)
def save_model(self,path):
import dill
from pathlib import Path
# create path
Path(path).mkdir(parents=True, exist_ok=True)
torch.save(
obj=self.Q_table_table,
obj=self.Q_table,
f=path+"checkpoint.pkl",
pickle_module=dill
)
print("Model saved!")
def load_model(self, path):
import dill
self.Q_table_table =torch.load(f=path+'checkpoint.pkl',pickle_module=dill)
self.Q_table=torch.load(f=path+'checkpoint.pkl',pickle_module=dill)
print("Mode loaded!")