Merge pull request #1 from jacarvalho/profiling

Profiling
jacarvalho · Feb 1, 2020 · 29ea00c · 29ea00c
2 parents 8059263 + 0263f96
commit 29ea00c
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 7 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1,4 +1,5 @@
 venv/
 .idea/
 __pycache__/
-*.egg-info/
+*.egg-info/
+quanser/
diff --git a/examples/pendulum_nopg_d_behavioral_sampleP.py b/examples/pendulum_nopg_d_behavioral_sampleP.py
@@ -65,7 +65,7 @@
 
 nopg = NOPG(dataset, policy, **nopg_params)
 
-n_policy_updates = 1000
+n_policy_updates = 50
 def optimizer(x): return optim.Adam(x, lr=1e-2)
 evaluation_params = {'eval_mdp': mdp,
                      'eval_every_n': 200,

diff --git a/examples/pendulum_nopg_d_uniform.py b/examples/pendulum_nopg_d_uniform.py
@@ -65,7 +65,7 @@
 
 nopg = NOPG(dataset, policy, **nopg_params)
 
-n_policy_updates = 1000
+n_policy_updates = 50
 def optimizer(x): return optim.Adam(x, lr=1e-2)
 evaluation_params = {'eval_mdp': mdp,
                      'eval_every_n': 200,

diff --git a/src/nopg/nopg.py b/src/nopg/nopg.py
@@ -226,17 +226,15 @@ def kernel(X, Y, bandwidth):
     @staticmethod
     def normalize(X, dim=1):
         """
-        Normalize X to sum up to 1 along dim.
+        Normalize X to sum up to 1 along dim, with l1-norm.
 
         :param X: the tensor to normalize
         :type X: torch.tensor
         :param dim: dimension
         :type dim: int
         :return: a normalized tensor, summing to 1 along dim
         """
-        tmp = X / X.sum(dim=dim, keepdim=True)
-        tmp[torch.isnan(tmp)] = 0.
-        return tmp
+        return F.normalize(X, p=1, dim=dim)
 
     def compute_q_mu(self):
         """