upgrade pyright and fix new errors

.github/workflows/ci.yml

@@ -1,4 +1,7 @@
 on:
+  pull_request:
+    branches:
+      - master
   push:
     branches:
       - main
@@ -38,7 +41,7 @@ jobs:
       - name: Pull pip dependencies
         run: ./install-pythonpackages.sh
       - name: Install pyright
-        run: npm install -g pyright
+        run: npm install -g pyright@1.1.151
       - name: Pull typing stubs from cache
         uses: actions/cache@v2
         with:

createstubs.sh

@@ -7,6 +7,8 @@ set -e
 
 . ./getpythonpath.sh
 
+pushd "$(dirname "$0")"
+
 echo "$(tput setaf 2)Creating type stubs$(tput sgr0)"
 createstub() {
     local name=$1
@@ -42,12 +44,16 @@ fi
 
 if [ ! -d "boolean" ]; then
     pyright --createstub boolean
-    sed -i '/class BooleanAlgebra(object):/a\    TRUE = ...\n    FALSE = ...' typings/boolean/boolean.pyi
+    sed -i '/class BooleanAlgebra:/a\    TRUE = ...\n    FALSE = ...' typings/boolean/boolean.pyi
 else
     echo stub 'boolean' already created
 fi
 
+echo 'Typing stub generation completed'
+
 # Stubs that needed manual patching and that
 # were instead checked-in in git
 #   pyright --createstub boolean
 #   pyright --createstub gym
+
+popd

cyberbattle/_env/graph_wrapper.py

@@ -142,7 +142,7 @@ class CyberBattleGraph(gym.Wrapper):
     def __init__(self, env, maximum_total_credentials=22, maximum_node_count=22):
         super().__init__(env)
         self._bounds = self.env._bounds
-        self.__graph = None
+        self.__graph = nx.DiGraph()
         self.observation_space = DiGraph(self._bounds.maximum_node_count)
 
     def reset(self):

cyberbattle/agents/baseline/agent_dql.py

@@ -318,7 +318,8 @@ def optimize_model(self, norm_clipping=False):
             torch.nn.utils.clip_grad_norm_(self.policy_net.parameters(), 1.0)
         else:
             for param in self.policy_net.parameters():
-                param.grad.data.clamp_(-1, 1)
+                if param.grad is not None:
+                    param.grad.data.clamp_(-1, 1)
         self.optimizer.step()
 
     def get_actor_state_vector(self, global_state: ndarray, actor_features: ndarray) -> ndarray:

cyberbattle/agents/baseline/agent_wrapper.py

@@ -18,7 +18,7 @@ class StateAugmentation:
     """Default agent state augmentation, consisting of the gym environment
     observation itself and nothing more."""
 
-    def __init__(self, observation: Optional[cyberbattle_env.Observation] = None):
+    def __init__(self, observation: cyberbattle_env.Observation):
         self.observation = observation
 
     def on_step(self, action: cyberbattle_env.Action, reward: float, done: bool, observation: cyberbattle_env.Observation):
@@ -424,7 +424,7 @@ class ActionTrackingStateAugmentation(StateAugmentation):
        - failed_action_count: count of action taken and failed at the current node
      """
 
-    def __init__(self, p: EnvironmentBounds, observation: Optional[cyberbattle_env.Observation] = None):
+    def __init__(self, p: EnvironmentBounds, observation: cyberbattle_env.Observation):
         self.aa = AbstractAction(p)
         self.success_action_count = np.zeros(shape=(p.maximum_node_count, self.aa.n_actions), dtype=np.int32)
         self.failed_action_count = np.zeros(shape=(p.maximum_node_count, self.aa.n_actions), dtype=np.int32)

cyberbattle/agents/baseline/learner.py

@@ -198,7 +198,7 @@ def epsilon_greedy_search(
     all_episodes_availability = []
 
     wrapped_env = AgentWrapper(cyberbattle_gym_env,
-                               ActionTrackingStateAugmentation(environment_properties))
+                               ActionTrackingStateAugmentation(environment_properties, cyberbattle_gym_env.reset()))
     steps_done = 0
     plot_title = f"{title} (epochs={episode_count}, ϵ={initial_epsilon}, ϵ_min={epsilon_minimum}," \
         + (f"ϵ_multdecay={epsilon_multdecay}," if epsilon_multdecay else '') \

requirements.txt

@@ -1,6 +1,6 @@
 gym~=0.17.3
 numpy==1.19.4
-boolean.py~=3.7
+boolean.py==3.8
 networkx==2.4
 pyyaml~=5.4.1
 setuptools~=49.2.1