Handy attributes#
gym-dssat provides observations to the agent as any RL environment ; but moreover the user can access additional hidden DSSAT’s state features for the sake of the analysis.
env._state
and env.observation
#
Hint
State and observation variables are defined as indicated in the Decision problems section.
env._state
#
Contains all the raw state variables which is retrieved at every time step from DSSAT-PDI. For example, these variables can be used to inspect the simulations or plot dynamics.
env.observation
#
Observations are subsets of env._state
. This corresponds to the agent perspective for the considered problem.
Histories#
Histories are dictionaries that are filled during each episode and which are emptied with env.reset()
. They are usefull for agent’s performance analysis. Each key contains nested lists of the form:
[[values_episode_1], ..., [values_episode_n]]
.
env.history
#
self.history = {'observation': [...], 'action': [...], 'reward': [...]}
env._history
#
self._history = {'state': [...], 'action': [...], 'reward': [...]}