pyRDDLGym-jax 2.7__py3-none-any.whl → 3.0__py3-none-any.whl

pyRDDLGym_jax/examples/configs/HVAC_ippc2023_slp.cfg

@@ -1,18 +1,17 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 5}
-rounding_kwargs={'weight': 5}
-control_kwargs={'weight': 5}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+bernoulli_sigmoid_weight=5
+sigmoid_weight=5
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
-optimizer_kwargs={'learning_rate': 0.02}
+optimizer_kwargs={'learning_rate': 0.1}
 batch_size_train=1
 batch_size_test=1
 
-[Training]
+[Optimize]
 key=42
 epochs=30000
-train_seconds=30
+train_seconds=60

pyRDDLGym_jax/examples/configs/MountainCar_Continuous_gym_slp.cfg

@@ -1,10 +1,8 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,7 +11,7 @@ batch_size_train=1
 batch_size_test=1
 clip_grad=1.0
 
-[Training]
+[Optimize]
 key=42
 epochs=1000
 train_seconds=30

pyRDDLGym_jax/examples/configs/MountainCar_ippc2023_slp.cfg

@@ -1,10 +1,8 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,7 +11,7 @@ batch_size_train=1
 batch_size_test=1
 clip_grad=1.0
 
-[Training]
+[Optimize]
 key=42
 epochs=1000
 train_seconds=30

pyRDDLGym_jax/examples/configs/PowerGen_Continuous_drp.cfg

@@ -1,18 +1,17 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxDeepReactivePolicy'
-method_kwargs={'topology': [256, 128]}
+method_kwargs={'topology': [128, 128]}
 optimizer='rmsprop'
 optimizer_kwargs={'learning_rate': 0.0001}
 batch_size_train=32
 batch_size_test=32
+pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=4000
 train_seconds=30

pyRDDLGym_jax/examples/configs/PowerGen_Continuous_replan.cfg

@@ -1,10 +1,8 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -12,8 +10,9 @@ optimizer_kwargs={'learning_rate': 0.1}
 batch_size_train=32
 batch_size_test=32
 rollout_horizon=5
+pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=2000
 train_seconds=1

pyRDDLGym_jax/examples/configs/PowerGen_Continuous_slp.cfg

@@ -1,18 +1,17 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
 optimizer_kwargs={'learning_rate': 0.05}
 batch_size_train=32
 batch_size_test=32
+pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=10000
 train_seconds=30

pyRDDLGym_jax/examples/configs/Quadcopter_drp.cfg

@@ -1,19 +1,17 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxDeepReactivePolicy'
-method_kwargs={'topology': [256, 128], 'activation': 'tanh'}
+method_kwargs={'topology': [128, 128], 'activation': 'tanh'}
 optimizer='rmsprop'
 optimizer_kwargs={'learning_rate': 0.001}
 batch_size_train=1
 batch_size_test=1
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=100000
 train_seconds=360

pyRDDLGym_jax/examples/configs/Quadcopter_physics_drp.cfg

@@ -0,0 +1,17 @@
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
+
+[Planner]
+method='JaxDeepReactivePolicy'
+method_kwargs={'topology': [128, 128], 'activation': 'tanh'}
+optimizer='rmsprop'
+optimizer_kwargs={'learning_rate': 0.001}
+batch_size_train=1
+batch_size_test=1
+pgpe=None
+
+[Optimize]
+key=42
+epochs=100000
+train_seconds=360

pyRDDLGym_jax/examples/configs/Quadcopter_physics_slp.cfg

@@ -0,0 +1,17 @@
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
+
+[Planner]
+method='JaxStraightLinePlan'
+method_kwargs={}
+optimizer='rmsprop'
+optimizer_kwargs={'learning_rate': 0.03}
+batch_size_train=1
+batch_size_test=1
+pgpe=None
+
+[Optimize]
+key=42
+epochs=100000
+train_seconds=360

pyRDDLGym_jax/examples/configs/Quadcopter_slp.cfg

@@ -1,10 +1,8 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=10
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,7 +11,7 @@ batch_size_train=1
 batch_size_test=1
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=100000
 train_seconds=360

pyRDDLGym_jax/examples/configs/Reservoir_Continuous_drp.cfg

@@ -1,10 +1,7 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
 
-[Optimizer]
+[Planner]
 method='JaxDeepReactivePolicy'
 method_kwargs={'topology': [64, 32]}
 optimizer='rmsprop'
@@ -13,7 +10,7 @@ batch_size_train=32
 batch_size_test=32
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=5000
 train_seconds=60

pyRDDLGym_jax/examples/configs/Reservoir_Continuous_replan.cfg

@@ -1,10 +1,7 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -12,8 +9,9 @@ optimizer_kwargs={'learning_rate': 0.1}
 batch_size_train=32
 batch_size_test=32
 rollout_horizon=5
+pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=500
 train_seconds=1

pyRDDLGym_jax/examples/configs/Reservoir_Continuous_slp.cfg

@@ -1,10 +1,7 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 10}
-rounding_kwargs={'weight': 10}
-control_kwargs={'weight': 10}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,7 +10,7 @@ batch_size_train=32
 batch_size_test=32
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=2000
 train_seconds=30

pyRDDLGym_jax/examples/configs/UAV_Continuous_slp.cfg

@@ -1,10 +1,8 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 1}
-rounding_kwargs={'weight': 1}
-control_kwargs={'weight': 1}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=1
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,7 +11,7 @@ batch_size_train=1
 batch_size_test=1
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=30000
 train_seconds=120

pyRDDLGym_jax/examples/configs/Wildfire_MDP_ippc2014_drp.cfg

@@ -1,10 +1,9 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 100}
-rounding_kwargs={'weight': 100}
-control_kwargs={'weight': 100}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=100
+bernoulli_sigmoid_weight=100
 
-[Optimizer]
+[Planner]
 method='JaxDeepReactivePolicy'
 method_kwargs={'topology': [128, 64]}
 optimizer='rmsprop'
@@ -13,7 +12,7 @@ batch_size_train=32
 batch_size_test=32
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=1000
 train_seconds=30

pyRDDLGym_jax/examples/configs/Wildfire_MDP_ippc2014_replan.cfg

@@ -1,10 +1,9 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 100}
-rounding_kwargs={'weight': 100}
-control_kwargs={'weight': 100}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=100
+bernoulli_sigmoid_weight=100
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -14,7 +13,7 @@ batch_size_test=32
 rollout_horizon=5
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=1000
 train_seconds=1

pyRDDLGym_jax/examples/configs/Wildfire_MDP_ippc2014_slp.cfg

@@ -1,10 +1,9 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 100}
-rounding_kwargs={'weight': 100}
-control_kwargs={'weight': 100}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=100
+bernoulli_sigmoid_weight=100
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,7 +12,7 @@ batch_size_train=32
 batch_size_test=32
 pgpe=None
 
-[Training]
+[Optimize]
 key=42
 epochs=1000
 train_seconds=30

pyRDDLGym_jax/examples/configs/default_drp.cfg

@@ -1,10 +1,7 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 20}
-rounding_kwargs={'weight': 20}
-control_kwargs={'weight': 20}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
 
-[Optimizer]
+[Planner]
 method='JaxDeepReactivePolicy'
 method_kwargs={}
 optimizer='rmsprop'
@@ -12,8 +9,8 @@ optimizer_kwargs={'learning_rate': 0.0001}
 batch_size_train=32
 batch_size_test=32
 
-[Training]
+[Optimize]
 key=42
 epochs=30000
 train_seconds=60
-policy_hyperparams=2.0
+policy_hyperparams=1.0

pyRDDLGym_jax/examples/configs/default_replan.cfg

@@ -1,10 +1,7 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 20}
-rounding_kwargs={'weight': 20}
-control_kwargs={'weight': 20}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -13,9 +10,9 @@ batch_size_train=32
 batch_size_test=32
 rollout_horizon=5
 
-[Training]
+[Optimize]
 key=42
 epochs=2000
 train_seconds=1
-policy_hyperparams=2.0
+policy_hyperparams=1.0
 print_summary=False

pyRDDLGym_jax/examples/configs/default_slp.cfg

@@ -1,10 +1,7 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': 20}
-rounding_kwargs={'weight': 20}
-control_kwargs={'weight': 20}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -12,8 +9,8 @@ optimizer_kwargs={'learning_rate': 0.01}
 batch_size_train=32
 batch_size_test=32
 
-[Training]
+[Optimize]
 key=42
 epochs=30000
 train_seconds=60
-policy_hyperparams=2.0
+policy_hyperparams=1.0

pyRDDLGym_jax/examples/configs/tuning_drp.cfg

@@ -1,19 +1,17 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': MODEL_WEIGHT_TUNE}
-rounding_kwargs={'weight': MODEL_WEIGHT_TUNE}
-control_kwargs={'weight': MODEL_WEIGHT_TUNE}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=MODEL_WEIGHT_TUNE
+print_warnings=False
 
-[Optimizer]
+[Planner]
 method='JaxDeepReactivePolicy'
 method_kwargs={'topology': [LAYER1_TUNE, LAYER2_TUNE]}
 optimizer='rmsprop'
 optimizer_kwargs={'learning_rate': LEARNING_RATE_TUNE}
 batch_size_train=32
 batch_size_test=32
-print_warnings=False
 
-[Training]
+[Optimize]
 train_seconds=30
 policy_hyperparams=POLICY_WEIGHT_TUNE
 print_summary=False

pyRDDLGym_jax/examples/configs/tuning_replan.cfg

@@ -1,10 +1,9 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': MODEL_WEIGHT_TUNE}
-rounding_kwargs={'weight': MODEL_WEIGHT_TUNE}
-control_kwargs={'weight': MODEL_WEIGHT_TUNE}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=MODEL_WEIGHT_TUNE
+print_warnings=False
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
@@ -12,9 +11,8 @@ optimizer_kwargs={'learning_rate': LEARNING_RATE_TUNE}
 batch_size_train=32
 batch_size_test=32
 rollout_horizon=ROLLOUT_HORIZON_TUNE
-print_warnings=False
 
-[Training]
+[Optimize]
 train_seconds=1
 policy_hyperparams=POLICY_WEIGHT_TUNE
 print_summary=False

pyRDDLGym_jax/examples/configs/tuning_slp.cfg

@@ -1,19 +1,17 @@
-[Model]
-logic='FuzzyLogic'
-comparison_kwargs={'weight': MODEL_WEIGHT_TUNE}
-rounding_kwargs={'weight': MODEL_WEIGHT_TUNE}
-control_kwargs={'weight': MODEL_WEIGHT_TUNE}
+[Compiler]
+method='DefaultJaxRDDLCompilerWithGrad'
+sigmoid_weight=MODEL_WEIGHT_TUNE
+print_warnings=False
 
-[Optimizer]
+[Planner]
 method='JaxStraightLinePlan'
 method_kwargs={}
 optimizer='rmsprop'
 optimizer_kwargs={'learning_rate': LEARNING_RATE_TUNE}
 batch_size_train=32
 batch_size_test=32
-print_warnings=False
 
-[Training]
+[Optimize]
 train_seconds=30
 policy_hyperparams=POLICY_WEIGHT_TUNE
 print_summary=False

pyRDDLGym_jax/examples/run_plan.py

@@ -25,36 +25,6 @@ from pyRDDLGym_jax.core.planner import (
     load_config, JaxBackpropPlanner, JaxOfflineController, JaxOnlineController
 )
 
-
-def run_cnn1d():
-    import haiku as hk
-    import jax
-    import jax.numpy as jnp
-
-    class CNN(hk.Module):
-        def __init__(self, name=None):
-            super().__init__(name=name)
-            self.conv1d_layer = hk.Conv1D(
-                output_channels=4,
-                kernel_shape=6,  # Kernel size for 1D convolution
-                padding="SAME",
-                name="conv"
-            )
-
-        def __call__(self, x):
-            return self.conv1d_layer(x)
-
-    # Example usage:
-    key = jax.random.PRNGKey(42)
-    input_data = jnp.ones([1, 4])  # Batch size 1, sequence length 10, 1 input channel
-
-    # Transform the Haiku module into a pure function
-    f = hk.transform(lambda x: CNN()(x))
-    params = f.init(key, input_data)
-    print(params['cnn/~/conv']['w'].shape)
-    print(params['cnn/~/conv']['b'].shape)
-    print(f.apply(params, key, input_data).shape)
-
     
 def main(domain: str, instance: str, method: str, episodes: int=1) -> None:
     
@@ -76,8 +46,8 @@ def main(domain: str, instance: str, method: str, episodes: int=1) -> None:
         exit(1)
     
     planner_args, _, train_args = load_config(config_path)
-    if 'dashboard' in train_args: 
-        train_args['dashboard'].launch()
+    if 'dashboard' in planner_args: 
+        planner_args['dashboard'].launch()
     
     # create the planning algorithm
     planner = JaxBackpropPlanner(
@@ -93,7 +63,6 @@ def main(domain: str, instance: str, method: str, episodes: int=1) -> None:
         
 
 def run_from_args(args):
-    run_cnn1d()
     if len(args) < 3:
         print('python run_plan.py <domain> <instance> <method> [<episodes>]')
         exit(1)

pyRDDLGym_jax/examples/run_tune.py

@@ -77,8 +77,8 @@ def main(domain: str, instance: str, method: str,
     # evaluate the agent on the best parameters
     planner_args, _, train_args = load_config_from_string(tuning.best_config)
     planner = JaxBackpropPlanner(rddl=env.model, **planner_args)
-    klass = JaxOnlineController if method == 'replan' else JaxOfflineController
-    controller = klass(planner, **train_args)
+    class_ = JaxOnlineController if method == 'replan' else JaxOfflineController
+    controller = class_(planner, **train_args)
     controller.evaluate(env, episodes=1, verbose=True, render=True)
     env.close()