Merge pull request #13 from boettiger-lab/prune-old-envs

pruned off old envs

README.md

@@ -2,22 +2,31 @@
 
 Models:
 
-- asm.py: provides `Asm()`.  Observes 1 dimension (total N).  Action is harvest
-- asm_2o.py: provides `Asm2o()`. Observes 2 dimensions: total N and mean biomass (wt).
-- ams_esc.py: escapement `AsmEsc()` escapement-encoded.
+- `asm_env.py`: provides `AsmEnv()`. This encodes our population dynamics model, coupled with an observation process, and a harvest process with a corresponding utility model. These processes can all be modified using the `config` argument. Their defaults are defined in `asm_fns.py`. By default, observations are stock biomass and mean weight.
+- `asm_esc.py`: provides `AsmEscEnv()` which inherits from `AsmEnv` and has one difference to it: actions in `AsmEscEnv()` represent escapement levels rather than fishing intensities. 
+- `ams_cr_like.py`: provides `AsmCRLike()`. In this environment, mean weight is observed and the action is to set parameters `(x1, x2, y2)` for a biomass-based harvest control rule of the type `CautionaryRule` (specified below).
 
-Methods: 
+Strategies evaluated with MSE: 
 
-(For both 1d and 2d observations)
-- MSE piecewise linear rule (in mortality space) 
-- Constant Mortality
-- Constant escapement
+- `agents.cautionary_rule.CautionaryRule`: piece-wise linear harvest-control rule specified by three parameters `(x1, x2, y2)`. Example plot (TBD).
+- `agents.msy.Msy`: constant mortality harvest control rule. Specified by one parameter `mortality`.
+- `agents.const_esc.ConstEsc`: constant escapement harvest control rule. Specified by one parameter `escapement`.
 
+## Installation
 
-## RL training:
+Clone this repo, then:
+
+```bash
+cd rl4fisheries
+pip install .
+```
 
-requires `rl4eco` package. Simply run `scripts/train.py` pointing at the chosen configuration. The trained model is automatically pushed to Huggingface (requires a HF token).
+## RL training:
 
+Simply run 
 ```bash
-python scripts/train.py -f hyperpars/ppo-asm2o-v0-1.yml
+python scripts/train.py -f path/to/config/file.yml
 ```
+The trained model is automatically pushed to Huggingface (requires a HF token). 
+The config files used for our results are found in `hyperpars/for_results/`
+

src/rl4fisheries/__init__.py

@@ -1,8 +1,5 @@
 # Importing from sub-directories here makes these available as 'top-level' imports
-from rl4fisheries.envs.asm import Asm
-from rl4fisheries.envs.asm_2o import Asm2o
 from rl4fisheries.envs.asm_env import AsmEnv
-
 from rl4fisheries.envs.asm_esc import AsmEnvEsc
 from rl4fisheries.envs.asm_cr_like import AsmCRLike
 
@@ -13,14 +10,10 @@
 
 
 from gymnasium.envs.registration import register
-# action is 'harvest'
-register(id="Asm-v0", entry_point="rl4fisheries.envs.asm:Asm")
+# action is fishing intensity
+register(id="AsmEnv", entry_point="rl4fisheries.envs.asm_env:AsmEnv")
 # action is 'escapement'
 register(id="AsmEnvEsc", entry_point="rl4fisheries.envs.asm_esc:AsmEnvEsc")
-# action is harvest, but observes both total count and mean biomass
-register(id="Asm2o-v0", entry_point="rl4fisheries.envs.asm_2o:Asm2o")
-# action is harvest, but observes both total count and mean biomass
-register(id="AsmEnv", entry_point="rl4fisheries.envs.asm_env:AsmEnv")
 # CR-like actions
 register(id="AsmCRLike", entry_point="rl4fisheries.envs.asm_cr_like:AsmCRLike")