How exactly should target_cost be set?

[hanhanlixianji]

In the README, you mentioned:
"Set target cost by using env.set_target_cost(target_cost) function, where target_cost is the undiscounted sum of costs of an episode."
Here, target_cost is the "undiscounted sum of costs of an episode."

Then you also said:
"You can use the env.get_normalized_score(return, cost_return) function to compute a normalized reward and cost for an episode, where returns and cost_returns are the undiscounted sum of rewards and costs respectively of an episode."
Here, cost_returns is also the "undiscounted sum of costs of an episode."

However, I found in the source code dsrl.offline_env.py at line 185:

DSRL/dsrl/offline_env.py

Line 185 in 8775fac

normalized_cost = (cost + self.epsilon) / (self.target_cost + self.epsilon)

Here, cost is cost_return, and according to your description in the README, the result of this should be 1. However, this is clearly not correct.

[hanhanlixianji]

In the README, you mentioned: "Set target cost by using env.set_target_cost(target_cost) function, where target_cost is the undiscounted sum of costs of an episode." Here, target_cost is the "undiscounted sum of costs of an episode."

Then you also said: "You can use the env.get_normalized_score(return, cost_return) function to compute a normalized reward and cost for an episode, where returns and cost_returns are the undiscounted sum of rewards and costs respectively of an episode." Here, cost_returns is also the "undiscounted sum of costs of an episode."

However, I found in the source code dsrl.offline_env.py at line 185:

DSRL/dsrl/offline_env.py

Line 185 in 8775fac

normalized_cost = (cost + self.epsilon) / (self.target_cost + self.epsilon)

Here, cost is cost_return, and according to your description in the README, the result of this should be 1. However, this is clearly not correct.

Also, could you please explain the detailed calculation process of the normalized cost? Thank you so much!