Update the hyperparameters for the brax example

Author: engintoklu

examples/notebooks/Brax_Experiments_with_PGPE.ipynb

@@ -7,7 +7,7 @@
    "source": [
     "# Solving a Brax environment using EvoTorch\n",
     "\n",
-    "This notebook demonstrates how the Brax environment named `humanoid` can be solved using EvoTorch.\n",
+    "This notebook demonstrates how the Brax environment named `humanoid` can be solved using EvoTorch. The hyperparameters here are tuned for brax version 0.10.5.\n",
     "\n",
     "EvoTorch provides `VecGymNE`, a neuroevolution problem type that focuses on solving vectorized environments. If GPU is available, `VecGymNE` can utilize it to boost performance. In this notebook, we use `VecGymNE` to solve the `humanoid` task.\n",
     "\n",
@@ -170,11 +170,11 @@
    "outputs": [],
    "source": [
     "# --- A simple linear policy ---\n",
-    "policy = \"Linear(obs_length, act_length)\"\n",
+    "# policy = \"Linear(obs_length, act_length)\"\n",
     "\n",
     "\n",
     "# --- A feed-forward network ---\n",
-    "# policy = \"Linear(obs_length, 64) >> Tanh() >> Linear(64, act_length)\"\n",
+    "policy = \"Linear(obs_length, 64) >> Tanh() >> Linear(64, act_length)\"\n",
     "\n",
     "\n",
     "# --- A feed-forward network with layer normalization ---\n",
@@ -308,17 +308,18 @@
    "outputs": [],
    "source": [
     "RADIUS = 2.25\n",
-    "MAX_SPEED = RADIUS / 6\n",
-    "CENTER_LR = MAX_SPEED / 2\n",
+    "MAX_SPEED = RADIUS / 15\n",
+    "CENTER_LR = MAX_SPEED * 0.75\n",
     "\n",
     "POPSIZE = 4000\n",
-    "NUM_GENERATIONS = 140\n",
+    "NUM_GENERATIONS = 1000\n",
     "SAVE_INTERVAL = 20\n",
     "\n",
     "# Instantiate a PGPE using the hyperparameters prepared above\n",
     "searcher = PGPE(\n",
     "    problem,\n",
     "    popsize=POPSIZE,\n",
+    "    num_interactions=(POPSIZE * 1000 * 0.75),\n",
     "    radius_init=RADIUS,\n",
     "    center_learning_rate=CENTER_LR,\n",
     "    optimizer=\"clipup\",\n",