feat(evaluate): evaluate_policy base

configs/evaluation/poster.yaml

@@ -0,0 +1,20 @@
+# @package evaluation
+# Configuration for the models used in the poster comparison.
+
+# Standard evaluation settings
+evaluate_checkpoints: false
+eval_max_steps: 5000
+eval_seed: 0
+
+# Cross-model comparison settings
+# We use 10 episodes to get a more robust average for the final poster results.
+comparison_base_seed: 0
+comparison_num_episodes: 10
+comparison_output_csv: "metrics/poster_comparison.csv"
+
+# Paths to the .cleanrl_model files to be compared (relative to workspace root).
+# These are placeholders; replace with actual trained model paths for the poster.
+comparison_models:
+  - "experiments/poster/centralized.cleanrl_model"
+  - "experiments/poster/decentralized.cleanrl_model"
+  - "experiments/poster/decentralized_amputated.cleanrl_model"