Reward not called on diverging flow error

[bwitherspoon]

Environment

• Grid2op version: 1.2.2
• System: Fedora 32

Bug description

The reward is not called on diverging flow exceptions and possibly other errors. The minimum reward is returned instead

Code snippet

import grid2op
from grid2op.Reward import BaseReward
from grid2op.dtypes import dt_float

class Reward(BaseReward):
    def __init__(self):
        super().__init__()
        self.reward_min = dt_float(100.0) # Note difference from below
        self.reward_max = dt_float(0.0)

    def __call__(self, action, env, has_error, is_done, is_illegal, is_ambiguous):
        if has_error:
            return dt_float(-10.0)
        else:
            return dt_float(1.0)

env = grid2op.make("l2rpn_case14_sandbox", reward_class=Reward)

obs = env.reset()
while True:
    act = env.action_space.sample()
    obs, reward, done, info = env.step(act)

    if len(info["exception"]) > 0:
        for e in info["exception"]:
            if type(e) is grid2op.Exceptions.DivergingPowerFlow:
                print("reward -", reward)
    if done:
        break

Current output

reward - 100.0

Expected output

reward - -10.0

[BDonnot]

Hello,
Thanks for reporting this bug :-)
I'll try to have a fix ready for the next version (hopefully released next week, but no promises)

As it is required that "min_reward" is acutally the minimum reward, i moved this issue from "bug" to "enhancement": if the requirement is met, then the behaviour is normal.
Though I totally agree this is quite surprising (and not really explicit) behaviour.

Thanks for the issue :-)

Benjamin

[BDonnot]

Fixed in version 1.2.3 :-) [release in progress at this very moment]