Fix Sampling inside gradient loop issue

CHANGELOG.md

@@ -14,6 +14,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 - Make flattened tensor storage in memory the default option (revert changed introduced in version 1.3.0)
 - Drop support for PyTorch versions prior to 1.10 (the previous supported version was 1.9).
 
+### Fixed
+- Moved the batch sampling inside gradient step loop for DQN, DDQN, DDPG (RNN), TD3 (RNN), SAC and SAC (RNN)
+
 ### Removed
 - Remove OpenAI Gym (`gym`) from dependencies and source code. **skrl** continues to support gym environments,
   it is just not installed as part of the library. If it is needed, it needs to be installed manually.

docs/source/api/agents/ddqn.rst

@@ -40,10 +40,10 @@ Learning algorithm
 
 |
 | :literal:`_update(...)`
-| :green:`# sample a batch from memory`
-| [:math:`s, a, r, s', d`] :math:`\leftarrow` states, actions, rewards, next_states, dones of size :guilabel:`batch_size`
 | :green:`# gradient steps`
 | **FOR** each gradient step up to :guilabel:`gradient_steps` **DO**
+|     :green:`# sample a batch from memory`
+|     [:math:`s, a, r, s', d`] :math:`\leftarrow` states, actions, rewards, next_states, dones of size :guilabel:`batch_size`
 |     :green:`# compute target values`
 |     :math:`Q' \leftarrow Q_{\phi_{target}}(s')`
 |     :math:`Q_{_{target}} \leftarrow Q'[\underset{a}{\arg\max} \; Q_\phi(s')] \qquad` :gray:`# the only difference with DQN`

docs/source/api/agents/dqn.rst

@@ -40,10 +40,10 @@ Learning algorithm
 
 |
 | :literal:`_update(...)`
-| :green:`# sample a batch from memory`
-| [:math:`s, a, r, s', d`] :math:`\leftarrow` states, actions, rewards, next_states, dones of size :guilabel:`batch_size`
 | :green:`# gradient steps`
 | **FOR** each gradient step up to :guilabel:`gradient_steps` **DO**
+|     :green:`# sample a batch from memory`
+|     [:math:`s, a, r, s', d`] :math:`\leftarrow` states, actions, rewards, next_states, dones of size :guilabel:`batch_size`
 |     :green:`# compute target values`
 |     :math:`Q' \leftarrow Q_{\phi_{target}}(s')`
 |     :math:`Q_{_{target}} \leftarrow \underset{a}{\max} \; Q' \qquad` :gray:`# the only difference with DDQN`

docs/source/api/agents/sac.rst

@@ -34,10 +34,10 @@ Learning algorithm
 
 |
 | :literal:`_update(...)`
-| :green:`# sample a batch from memory`
-| [:math:`s, a, r, s', d`] :math:`\leftarrow` states, actions, rewards, next_states, dones of size :guilabel:`batch_size`
 | :green:`# gradient steps`
 | **FOR** each gradient step up to :guilabel:`gradient_steps` **DO**
+|     :green:`# sample a batch from memory`
+|     [:math:`s, a, r, s', d`] :math:`\leftarrow` states, actions, rewards, next_states, dones of size :guilabel:`batch_size`
 |     :green:`# compute target values`
 |     :math:`a',\; logp' \leftarrow \pi_\theta(s')`
 |     :math:`Q_{1_{target}} \leftarrow Q_{{\phi 1}_{target}}(s', a')`

skrl/agents/jax/dqn/ddqn.py

@@ -336,13 +336,14 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/jax/dqn/dqn.py

@@ -333,13 +333,14 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/jax/sac/sac.py

@@ -396,13 +396,14 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self._tensors_names, batch_size=self._batch_size)[0]
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self._tensors_names, batch_size=self._batch_size)[0]
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/jax/td3/td3.py

@@ -435,6 +435,7 @@ def _update(self, timestep: int, timesteps: int) -> None:
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
+
             # sample a batch from memory
             sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
                 self.memory.sample(names=self._tensors_names, batch_size=self._batch_size)[0]

skrl/agents/torch/ddpg/ddpg_rnn.py

@@ -360,18 +360,19 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self._tensors_names, batch_size=self._batch_size, sequence_length=self._rnn_sequence_length)[0]
-
-        rnn_policy = {}
-        if self._rnn:
-            sampled_rnn = self.memory.sample_by_index(names=self._rnn_tensors_names, indexes=self.memory.get_sampling_indexes())[0]
-            rnn_policy = {"rnn": [s.transpose(0, 1) for s in sampled_rnn], "terminated": sampled_dones}
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self._tensors_names, batch_size=self._batch_size, sequence_length=self._rnn_sequence_length)[0]
+
+            rnn_policy = {}
+            if self._rnn:
+                sampled_rnn = self.memory.sample_by_index(names=self._rnn_tensors_names, indexes=self.memory.get_sampling_indexes())[0]
+                rnn_policy = {"rnn": [s.transpose(0, 1) for s in sampled_rnn], "terminated": sampled_dones}
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/torch/dqn/ddqn.py

@@ -283,13 +283,14 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/torch/dqn/dqn.py

@@ -283,13 +283,14 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self.tensors_names, batch_size=self._batch_size)[0]
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/torch/sac/sac.py

@@ -307,12 +307,13 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
+
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
             # sample a batch from memory
             sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self._tensors_names, batch_size=self._batch_size)[0]
+                self.memory.sample(names=self._tensors_names, batch_size=self._batch_size)[0]
 
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)

skrl/agents/torch/sac/sac_rnn.py

@@ -344,18 +344,19 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self._tensors_names, batch_size=self._batch_size, sequence_length=self._rnn_sequence_length)[0]
-
-        rnn_policy = {}
-        if self._rnn:
-            sampled_rnn = self.memory.sample_by_index(names=self._rnn_tensors_names, indexes=self.memory.get_sampling_indexes())[0]
-            rnn_policy = {"rnn": [s.transpose(0, 1) for s in sampled_rnn], "terminated": sampled_dones}
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self._tensors_names, batch_size=self._batch_size, sequence_length=self._rnn_sequence_length)[0]
+
+            rnn_policy = {}
+            if self._rnn:
+                sampled_rnn = self.memory.sample_by_index(names=self._rnn_tensors_names, indexes=self.memory.get_sampling_indexes())[0]
+                rnn_policy = {"rnn": [s.transpose(0, 1) for s in sampled_rnn], "terminated": sampled_dones}
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)
 

skrl/agents/torch/td3/td3_rnn.py

@@ -382,18 +382,19 @@ def _update(self, timestep: int, timesteps: int) -> None:
         :param timesteps: Number of timesteps
         :type timesteps: int
         """
-        # sample a batch from memory
-        sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
-            self.memory.sample(names=self._tensors_names, batch_size=self._batch_size, sequence_length=self._rnn_sequence_length)[0]
-
-        rnn_policy = {}
-        if self._rnn:
-            sampled_rnn = self.memory.sample_by_index(names=self._rnn_tensors_names, indexes=self.memory.get_sampling_indexes())[0]
-            rnn_policy = {"rnn": [s.transpose(0, 1) for s in sampled_rnn], "terminated": sampled_dones}
 
         # gradient steps
         for gradient_step in range(self._gradient_steps):
 
+            # sample a batch from memory
+            sampled_states, sampled_actions, sampled_rewards, sampled_next_states, sampled_dones = \
+                self.memory.sample(names=self._tensors_names, batch_size=self._batch_size, sequence_length=self._rnn_sequence_length)[0]
+
+            rnn_policy = {}
+            if self._rnn:
+                sampled_rnn = self.memory.sample_by_index(names=self._rnn_tensors_names, indexes=self.memory.get_sampling_indexes())[0]
+                rnn_policy = {"rnn": [s.transpose(0, 1) for s in sampled_rnn], "terminated": sampled_dones}
+
             sampled_states = self._state_preprocessor(sampled_states, train=True)
             sampled_next_states = self._state_preprocessor(sampled_next_states, train=True)