Changes to _adaptive.py

deluca/agents/__init__.py

@@ -16,5 +16,9 @@
 from deluca.agents._pid import PID
 from deluca.agents._random import Random
 from deluca.agents._zero import Zero
+from deluca.agents._adaptive import Adaptive
+from deluca.agents._gpc import GPC
+from deluca.agents._lqr import LQR
 
-__all__ = ["BangBang", "PID", "Random", "Zero"]
+
+__all__ = ["BangBang", "PID", "Random", "Zero", "Adaptive", "GPC", "LQR"]

deluca/agents/_adaptive.py

@@ -20,7 +20,7 @@
 import jax.numpy as jnp
 import numpy as np
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 
 
 def quad_loss(x: jnp.ndarray, u: jnp.ndarray) -> Real:
@@ -66,13 +66,13 @@ def __init__(
         self.n, self.m = B.shape
 
         cost_fn = cost_fn or quad_loss
+        self.cost_fn = cost_fn
 
         self.base_controller = base_controller
 
         # Start From Uniform Distribution
         self.T = T
         self.weights = np.zeros(T)
-        self.weights[0] = 1.0
 
         # Track current timestep
         self.t, self.expert_density = 0, expert_density
@@ -111,10 +111,14 @@ def evolve(x, h):
     def __call__(self, x, A, B):
 
         play_i = np.argmax(self.weights)
+        if not self.alive[play_i]:
+            play_i = 0
+
         self.u = self.learners[play_i].get_action(x)
 
         # Update alive models
         for i in jnp.nonzero(self.alive)[0]:
+            i = int(i)
             loss_i = self.policy_loss(self.learners[i], A, B, x, self.w)
             self.weights[i] *= np.exp(-self.eta * loss_i)
             self.weights[i] = min(max(self.weights[i], self.eps), self.inf)
@@ -138,13 +142,17 @@ def __call__(self, x, A, B):
                 del self.learners[kill]
                 self.weights[kill] = 0
 
-        # Rescale
-        max_w = np.max(self.weights)
-        if max_w < 1:
-            self.weights /= max_w
+        # # Rescale
+        # max_w = np.max(self.weights)
+        # if max_w < 1:
+        #     self.weights /= max_w
+        max_weight_i = np.argmax(self.weights)
+        max_w = self.weights[max_weight_i]
+        if self.alive[max_weight_i] and max_w < 1:
+          self.weights /= max_w
 
         # Get new noise (will be located at w[-1])
-        self.w = self.w.at[0].set(x - self.A @ self.x + self.B @ self.u)
+        self.w = self.w.at[0].set(x - self.A @ self.x - self.B @ self.u)
         self.w = jnp.roll(self.w, -1, axis=0)
 
         # Update System

deluca/agents/_bpc.py

@@ -20,7 +20,7 @@
 import numpy.random as random
 
 from deluca.agents._lqr import LQR
-from deluca.agents.core import Agent
+from deluca.core import Agent
 
 
 def generate_uniform(shape, norm=1.00):

deluca/agents/_deep.py

@@ -18,7 +18,7 @@
 import jax
 import jax.numpy as jnp
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 from deluca.utils import Random
 
 

deluca/agents/_drc.py

@@ -21,7 +21,7 @@
 from jax import grad
 from jax import jit
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 from deluca.utils import Random
 
 

deluca/agents/_gpc.py

@@ -23,7 +23,7 @@
 from jax import jit
 
 from deluca.agents._lqr import LQR
-from deluca.agents.core import Agent
+from deluca.core import Agent
 
 
 def quad_loss(x: jnp.ndarray, u: jnp.ndarray) -> Real:

deluca/agents/_hinf.py

@@ -19,7 +19,7 @@
 import jax.numpy as jnp
 from jax.numpy.linalg import inv
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 
 
 class Hinf(Agent):

deluca/agents/_igpc.py

@@ -15,7 +15,7 @@
 import jax
 import jax.numpy as jnp
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 from deluca.utils.planning import f_at_x
 from deluca.utils.planning import LQR
 from deluca.utils.planning import rollout

deluca/agents/_ilc.py

@@ -14,7 +14,7 @@
 
 import jax.numpy as jnp
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 from deluca.utils.planning import f_at_x
 from deluca.utils.planning import LQR
 from deluca.utils.planning import rollout

deluca/agents/_ilqr.py

@@ -22,7 +22,7 @@
 
 import jax.numpy as jnp
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 
 
 def iLQR_loop(env, U_initial, T, alpha=1.0, log=None):

deluca/agents/_lqr.py

@@ -16,7 +16,7 @@
 import jax.numpy as jnp
 from scipy.linalg import solve_discrete_are as dare
 
-from deluca.agents.core import Agent
+from deluca.core import Agent
 
 
 class LQR(Agent):