Finalize strict grader discovery and stabilize RL checkpoint loading

.github/workflows/deploy-readiness.yml

@@ -1,30 +0,0 @@
-name: Deploy Readiness
-
-on:
-  push:
-    branches: [ "main", "master" ]
-  pull_request:
-
-jobs:
-  test-and-build:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install -r requirements.txt
-          pip install -e ./triage_env
-
-      - name: Run tests
-        run: python -m pytest -q
-
-      - name: Build Docker image
-        run: docker build -t medicaltriage:ci .

.gitignore

@@ -41,6 +41,7 @@ dist/
 #Environment variables
 triage_env/training/*.pkl
 triage_env/training/*.meta.json
+triage_env/training/triage_rl_qtable*.json
 !triage_env/training/q_agent.pkl
 !triage_env/training/q_agent_task1.pkl
 !triage_env/training/q_agent_task2.pkl

DEPLOYMENT.md

@@ -55,8 +55,9 @@ Default manifests:
 - `deployment/k8s/deployment.yaml`
 - `deployment/k8s/service.yaml`
 
-## 6) CI Readiness Workflow
-A baseline CI workflow exists at:
-- `.github/workflows/deploy-readiness.yml`
-
-It runs tests and Docker build on push/PR.
+## 6) Manual Readiness Check
+Run local tests and image build before release:
+```bash
+python -m pytest -q
+docker build -t medicaltriage:ci .
+```

openenv.yaml

@@ -23,7 +23,7 @@ tasks:
     grader_command: python graders/task1_grader.py
     grader_path: graders/task1_grader.py
     grader_file: graders/task1_grader.py
-    grader_legacy_path: graders/task1_grader.py
+    grader_legacy_path: graders/task1.py
     graders:
       - type: python
         name: default
@@ -40,7 +40,7 @@ tasks:
     grader_command: python graders/task2_grader.py
     grader_path: graders/task2_grader.py
     grader_file: graders/task2_grader.py
-    grader_legacy_path: graders/task2_grader.py
+    grader_legacy_path: graders/task2.py
     graders:
       - type: python
         name: default
@@ -57,7 +57,7 @@ tasks:
     grader_command: python graders/task3_grader.py
     grader_path: graders/task3_grader.py
     grader_file: graders/task3_grader.py
-    grader_legacy_path: graders/task3_grader.py
+    grader_legacy_path: graders/task3.py
     graders:
       - type: python
         name: default

triage_env/agents/rl_agents.py

@@ -34,6 +34,13 @@ class RLAgent(BaseAgent):
     def _state_key(self, observation: TriageObservation):
         return encode_observation(observation)
 
+    def _freeze_json_value(self, value):
+        if isinstance(value, list):
+            return tuple(self._freeze_json_value(item) for item in value)
+        if isinstance(value, dict):
+            return tuple(sorted((k, self._freeze_json_value(v)) for k, v in value.items()))
+        return value
+
     def _valid_actions(self, observation: TriageObservation):
         alive = [p for p in observation.patients if p.alive]
         actions = [("wait", -1)]
@@ -143,8 +150,8 @@ class RLAgent(BaseAgent):
 
         self.q_table = {}
         for state_str, actions in data["q_table"].items():
-            state = tuple(json.loads(state_str))
+            state = self._freeze_json_value(json.loads(state_str))
             self.q_table[state] = {}
             for action_str, value in actions.items():
-                action = tuple(json.loads(action_str))
+                action = self._freeze_json_value(json.loads(action_str))
                 self.q_table[state][action] = float(value)

triage_env/openenv.yaml

@@ -23,7 +23,7 @@ tasks:
     grader_command: python graders/task1_grader.py
     grader_path: graders/task1_grader.py
     grader_file: graders/task1_grader.py
-    grader_legacy_path: graders/task1_grader.py
+    grader_legacy_path: graders/task1.py
     graders:
       - type: python
         name: default
@@ -40,7 +40,7 @@ tasks:
     grader_command: python graders/task2_grader.py
     grader_path: graders/task2_grader.py
     grader_file: graders/task2_grader.py
-    grader_legacy_path: graders/task2_grader.py
+    grader_legacy_path: graders/task2.py
     graders:
       - type: python
         name: default
@@ -57,7 +57,7 @@ tasks:
     grader_command: python graders/task3_grader.py
     grader_path: graders/task3_grader.py
     grader_file: graders/task3_grader.py
-    grader_legacy_path: graders/task3_grader.py
+    grader_legacy_path: graders/task3.py
     graders:
       - type: python
         name: default