added training against checkpoint

tests/test_opponents.py

@@ -1,7 +1,7 @@
 import pytest
 from generals.agents import ExpanderAgent, HunterAgent, RandomAgent
 
-from general_bots_training.opponents import SelfPlayOpponent, make_opponent
+from general_bots_training.opponents import ModelOpponent, SelfPlayOpponent, make_opponent
 
 
 @pytest.mark.parametrize(
@@ -28,3 +28,28 @@ def test_make_opponent_supports_self_play_aliases():
 def test_make_opponent_rejects_unknown_name():
     with pytest.raises(ValueError, match="unknown opponent"):
         make_opponent("turtle")
+
+
+def test_make_opponent_model_requires_checkpoint():
+    with pytest.raises(ValueError, match="requires a checkpoint path"):
+        make_opponent("model")
+
+
+def test_make_opponent_model_loads_checkpoint(tmp_path):
+    import equinox as eqx
+    import jax.random as jrandom
+
+    from general_bots_training.network import PolicyValueNetwork
+
+    network = PolicyValueNetwork(jrandom.PRNGKey(0), in_channels=14)
+    checkpoint_path = tmp_path / "model.eqx"
+    eqx.tree_serialise_leaves(checkpoint_path, network)
+
+    opponent = make_opponent(f"model:{checkpoint_path}")
+
+    assert isinstance(opponent, ModelOpponent)
+
+
+def test_make_opponent_rejects_checkpoint_on_non_model():
+    with pytest.raises(ValueError, match="does not take a checkpoint"):
+        make_opponent("random:some.eqx")