Observation space#
As introduced in Section Decision problems, gym-DSSAT problems are POMDPs. In Section Observation spaces, we shortly showed the successive subsets of DSSAT’s internal variables to define the observations provided to the agent. By modifying the list of variables constituting the observations, you can modulate “the degree of partial observability” of the POMDP.
${GYM_DSSAT_PDI_PATH}/envs/configs/env_config.yml
If we take the example of a new foo
fertilization problem, we can define it following:
#################
# RAW STATE SPACE
#################
state: # <- list of all DSSAT's internal features made available in gym-DSSAT
# please keep it untouched !
vstage:
type: float
low: 0
high: .inf
info: vegetative growth stage (number of leaves)
cleach:
type: float
low: 0
high: .inf
info: cumulative nitrate leaching (kg/ha)
dtt:
type: float
low: 0
high: 100
info: growing degree days for current day (C/d)
dul:
type: array
subtype: float
low: 0
high: 1
size: 9
info: volumetric soil water content in soil layers at drained upper limit (cm3 [water] / cm3 [soil])
xlai:
type: float
low: 0
high: .inf
info: plant population leaf area index (m2_leaf/m2_soil)
...
#######################################################
# ACTION AND STATE SPACES RESTRICTIONS FOR ENV SETTINGS
#######################################################
setting:
foo: # <- the new setting
action:
- anfer
state:
- cleach
- dtt
- vstage
context: # additional contextual information ; corresponds to gym's "info".
- dul
experiment_number: 1 # number of the experiment in DSSAT's fileX.
Warning
You cannot import new state features from DSSAT-PDI by simply adding variables to the state
YAML key of the raw state space without modifying more deeply DSSAT-pdi. For the straighforward way, you can only modify the state
key inside the setting
key, for the considered setting as shown above.
Note
For understanding what DSSAT’s fileX is, please check DSSAT’s user guide.