switch to a harder problem for example

EthanJamesLew · Apr 13, 2024 · b08476e · b08476e
1 parent 5e2254c
commit b08476e
Showing 1 changed file with 13 additions and 10 deletions.
diff --git a/notebooks/weighted-cost-func.ipynb b/notebooks/weighted-cost-func.ipynb
@@ -29,7 +29,9 @@
     "sys.path.append(\"..\")\n",
     "# this is the convenience function\n",
     "from autokoopman import auto_koopman\n",
-    "import autokoopman as ak"
+    "import autokoopman as ak\n",
+    "\n",
+    "np.random.seed(1234)"
    ]
   },
   {
@@ -50,7 +52,7 @@
     "\n",
     "# add garbage states -- we will weight these values to zero\n",
     "training_data = ak.TrajectoriesData({\n",
-    "    key: ak.Trajectory(t.times, np.hstack([t.states, np.random.rand(len(t.states), 3)]), t.inputs) for key, t in training_data._trajs.items()\n",
+    "    key: ak.Trajectory(t.times, np.hstack([t.states, np.random.rand() * np.ones((len(t.states), 3))]), t.inputs) for key, t in training_data._trajs.items()\n",
     "})"
    ]
   },
@@ -88,8 +90,8 @@
     "    #w = np.sum(traj.abs().states, axis=1)\n",
     "    #w = 1/(traj.abs().states+1.0)\n",
     "    w = np.ones(traj.states.shape)\n",
-    "    w[:, -3:] = 0.0\n",
-    "    w[:, 0] = 1.0\n",
+    "    w[:, -3:] = 1.0\n",
+    "    w[:, :2] = 1.0\n",
     "    weights.append(w)\n",
     "\n",
     "    # weight garbage trajectory to zero\n",
@@ -131,11 +133,11 @@
     "    learning_weights=weights, # weight the eDMD algorithm objectives\n",
     "    scoring_weights=weights, # pass weights as required for cost_func=\"weighted\"\n",
     "    opt=\"grid\",             # grid search to find best hyperparameters\n",
-    "    n_obs=20,              # maximum number of observables to try\n",
+    "    n_obs=40,              # maximum number of observables to try\n",
     "    max_opt_iter=200,       # maximum number of optimization iterations\n",
     "    grid_param_slices=5,   # for grid search, number of slices for each parameter\n",
-    "    n_splits=5,             # k-folds validation for tuning, helps stabilize the scoring\n",
-    "    rank=(1, 200, 40)       # rank range (start, stop, step) DMD hyperparameter\n",
+    "    n_splits=None,             # k-folds validation for tuning, helps stabilize the scoring\n",
+    "    rank=(20, 21, 1)       # rank range (start, stop, step) DMD hyperparameter\n",
     ")"
    ]
   },
@@ -163,11 +165,12 @@
     "    learning_weights=None,  # don't use eDMD weighting\n",
     "    scoring_weights=weights, # pass weights as required for cost_func=\"weighted\"\n",
     "    opt=\"grid\",             # grid search to find best hyperparameters\n",
-    "    n_obs=20,              # maximum number of observables to try\n",
+    "    n_obs=40,              # maximum number of observables to try\n",
     "    max_opt_iter=200,       # maximum number of optimization iterations\n",
     "    grid_param_slices=5,   # for grid search, number of slices for each parameter\n",
     "    n_splits=5,             # k-folds validation for tuning, helps stabilize the scoring\n",
-    "    rank=(1, 200, 40)       # rank range (start, stop, step) DMD hyperparameter\n",
+    "    lengthscale=(0.1, 1.0),\n",
+    "    rank=(20, 21, 1)       # rank range (start, stop, step) DMD hyperparameter\n",
     ")"
    ]
   },
@@ -195,7 +198,7 @@
     "tend = 7.0\n",
     "\n",
     "# simulate using the learned model\n",
-    "iv = [0.5, 0.5, 0.0, 0.0, 0.0]\n",
+    "iv = [0.5, 0.5, 0.5, 0.5, 0.5]\n",
     "trajectory = model.solve_ivp(\n",
     "    initial_state=iv,\n",
     "    tspan=(0.0, tend, 0.0),\n",