@elizaos/training 2.0.0-alpha.10

package/python/tests/test_scenario_pool.py

@@ -0,0 +1,735 @@
+"""
+Tests for ScenarioPool and CurriculumManager
+
+Tests cover:
+- Scenario generation (synthetic and structure)
+- Curriculum learning (tracking, priorities, reset)
+- Pool management (sampling, refresh, persistence)
+"""
+
+import json
+import tempfile
+from pathlib import Path
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from src.training.scenario_pool import (
+    CurriculumManager,
+    CurriculumState,
+    MarketState,
+    NewsItem,
+    PerpetualState,
+    PortfolioState,
+    Scenario,
+    ScenarioPool,
+    ScenarioPoolConfig,
+    SocialPost,
+)
+
+
+# =============================================================================
+# Data Structure Tests
+# =============================================================================
+
+
+class TestMarketState:
+    """Tests for MarketState dataclass"""
+
+    def test_creation(self):
+        market = MarketState(
+            market_id="test-1",
+            question="Will BTC exceed $100K?",
+            yes_price=0.65,
+            no_price=0.35,
+            volume_24h=100000.0,
+            liquidity=500000.0,
+            expires_at=1735689600000,
+            category="crypto",
+        )
+        
+        assert market.market_id == "test-1"
+        assert market.yes_price == 0.65
+        assert market.no_price == 0.35
+
+    def test_to_dict(self):
+        market = MarketState(
+            market_id="test-1",
+            question="Will BTC exceed $100K?",
+            yes_price=0.65,
+            no_price=0.35,
+            volume_24h=100000.0,
+            liquidity=500000.0,
+            expires_at=1735689600000,
+        )
+        
+        result = market.to_dict()
+        
+        assert result["id"] == "test-1"
+        assert result["yesPrice"] == 0.65
+        assert result["noPrice"] == 0.35
+        assert result["volume24h"] == 100000.0
+        assert "question" in result
+
+
+class TestPerpetualState:
+    """Tests for PerpetualState dataclass"""
+
+    def test_creation(self):
+        perp = PerpetualState(
+            ticker="BTC",
+            mark_price=100000.0,
+            index_price=99990.0,
+            funding_rate=0.0001,
+            open_interest=50000000.0,
+            volume_24h=100000000.0,
+            change_24h=0.02,
+            high_24h=102000.0,
+            low_24h=98000.0,
+        )
+        
+        assert perp.ticker == "BTC"
+        assert perp.mark_price == 100000.0
+
+    def test_to_dict(self):
+        perp = PerpetualState(
+            ticker="ETH",
+            mark_price=3500.0,
+            index_price=3495.0,
+            funding_rate=-0.0002,
+            open_interest=25000000.0,
+            volume_24h=50000000.0,
+            change_24h=-0.01,
+            high_24h=3600.0,
+            low_24h=3400.0,
+        )
+        
+        result = perp.to_dict()
+        
+        assert result["ticker"] == "ETH"
+        assert result["markPrice"] == 3500.0
+        assert result["fundingRate"] == -0.0002
+
+
+class TestScenario:
+    """Tests for Scenario dataclass"""
+
+    def test_creation_minimal(self):
+        scenario = Scenario(
+            id="test-scenario",
+            source="synthetic",
+        )
+        
+        assert scenario.id == "test-scenario"
+        assert scenario.source == "synthetic"
+        assert len(scenario.markets) == 0
+        assert scenario.portfolio.balance == 10000.0
+
+    def test_creation_full(self):
+        market = MarketState(
+            market_id="m1",
+            question="Test?",
+            yes_price=0.5,
+            no_price=0.5,
+            volume_24h=1000.0,
+            liquidity=5000.0,
+            expires_at=1735689600000,
+        )
+        
+        perp = PerpetualState(
+            ticker="BTC",
+            mark_price=100000.0,
+            index_price=100000.0,
+            funding_rate=0.0,
+            open_interest=1000000.0,
+            volume_24h=5000000.0,
+            change_24h=0.0,
+            high_24h=100000.0,
+            low_24h=100000.0,
+        )
+        
+        news = NewsItem(
+            headline="Test headline",
+            sentiment="bullish",
+            impact="high",
+            source="Test",
+            timestamp=1735689600000,
+        )
+        
+        post = SocialPost(
+            author="test_user",
+            content="Test post",
+            sentiment="neutral",
+            likes=10,
+            replies=2,
+            timestamp=1735689600000,
+        )
+        
+        scenario = Scenario(
+            id="full-scenario",
+            source="production",
+            markets=[market],
+            perpetuals=[perp],
+            news=[news],
+            social_posts=[post],
+            portfolio=PortfolioState(balance=15000.0),
+            archetype_focus="trader",
+            difficulty="hard",
+        )
+        
+        assert scenario.id == "full-scenario"
+        assert len(scenario.markets) == 1
+        assert len(scenario.perpetuals) == 1
+        assert len(scenario.news) == 1
+        assert len(scenario.social_posts) == 1
+        assert scenario.portfolio.balance == 15000.0
+        assert scenario.archetype_focus == "trader"
+        assert scenario.difficulty == "hard"
+
+    def test_to_dict(self):
+        scenario = Scenario(
+            id="dict-test",
+            source="synthetic",
+            markets=[MarketState(
+                market_id="m1",
+                question="Test?",
+                yes_price=0.6,
+                no_price=0.4,
+                volume_24h=1000.0,
+                liquidity=5000.0,
+                expires_at=1735689600000,
+            )],
+            difficulty="easy",
+        )
+        
+        result = scenario.to_dict()
+        
+        assert result["id"] == "dict-test"
+        assert result["source"] == "synthetic"
+        assert len(result["markets"]) == 1
+        assert result["difficulty"] == "easy"
+        assert "portfolio" in result
+
+    def test_to_observation(self):
+        scenario = Scenario(
+            id="obs-test",
+            source="synthetic",
+            markets=[MarketState(
+                market_id="m1",
+                question="Will BTC moon?",
+                yes_price=0.7,
+                no_price=0.3,
+                volume_24h=100000.0,
+                liquidity=500000.0,
+                expires_at=1735689600000,
+            )],
+            news=[NewsItem(
+                headline="Bullish news",
+                sentiment="bullish",
+                impact="high",
+                source="Test",
+                timestamp=1735689600000,
+            )],
+        )
+        
+        obs = scenario.to_observation()
+        
+        assert "markets" in obs
+        assert "perpetuals" in obs
+        assert "news" in obs
+        assert "portfolio" in obs
+        assert "marketSummary" in obs
+        assert obs["marketSummary"]["totalMarkets"] == 1
+
+    def test_sentiment_calculation(self):
+        # Mostly bullish
+        scenario = Scenario(
+            id="bullish-test",
+            source="synthetic",
+            news=[
+                NewsItem(headline="Bull1", sentiment="bullish", impact="high", source="X", timestamp=0),
+                NewsItem(headline="Bull2", sentiment="bullish", impact="high", source="X", timestamp=0),
+                NewsItem(headline="Neutral", sentiment="neutral", impact="low", source="X", timestamp=0),
+            ],
+        )
+        
+        obs = scenario.to_observation()
+        assert obs["marketSummary"]["avgSentiment"] == "bullish"
+        
+        # Mostly bearish
+        scenario2 = Scenario(
+            id="bearish-test",
+            source="synthetic",
+            news=[
+                NewsItem(headline="Bear1", sentiment="bearish", impact="high", source="X", timestamp=0),
+                NewsItem(headline="Bear2", sentiment="bearish", impact="high", source="X", timestamp=0),
+            ],
+        )
+        
+        obs2 = scenario2.to_observation()
+        assert obs2["marketSummary"]["avgSentiment"] == "bearish"
+
+
+# =============================================================================
+# CurriculumManager Tests
+# =============================================================================
+
+
+class TestCurriculumManager:
+    """Tests for CurriculumManager"""
+
+    def test_creation(self):
+        manager = CurriculumManager()
+        
+        assert len(manager.attempts) == 0
+        assert len(manager.scores) == 0
+        assert len(manager.solved) == 0
+
+    def test_record_attempt(self):
+        manager = CurriculumManager()
+        
+        manager.record_attempt("scenario-1", 0.5)
+        
+        assert manager.attempts["scenario-1"] == 1
+        assert len(manager.scores["scenario-1"]) == 1
+        assert manager.scores["scenario-1"][0] == 0.5
+
+    def test_multiple_attempts(self):
+        manager = CurriculumManager()
+        
+        manager.record_attempt("scenario-1", 0.3)
+        manager.record_attempt("scenario-1", 0.5)
+        manager.record_attempt("scenario-1", 0.7)
+        
+        assert manager.attempts["scenario-1"] == 3
+        assert len(manager.scores["scenario-1"]) == 3
+
+    def test_solved_detection(self):
+        manager = CurriculumManager(
+            solve_threshold=0.8,
+            min_attempts_for_solved=3,
+        )
+        
+        # Not enough attempts
+        manager.record_attempt("scenario-1", 0.9)
+        manager.record_attempt("scenario-1", 0.9)
+        assert "scenario-1" not in manager.solved
+        
+        # Third attempt triggers solved check
+        manager.record_attempt("scenario-1", 0.9)
+        assert "scenario-1" in manager.solved
+
+    def test_solved_requires_high_scores(self):
+        manager = CurriculumManager(
+            solve_threshold=0.8,
+            min_attempts_for_solved=3,
+        )
+        
+        # Scores below threshold
+        manager.record_attempt("scenario-1", 0.5)
+        manager.record_attempt("scenario-1", 0.6)
+        manager.record_attempt("scenario-1", 0.7)
+        
+        assert "scenario-1" not in manager.solved
+
+    def test_should_skip(self):
+        manager = CurriculumManager(
+            max_avg_for_skip=0.85,
+        )
+        
+        # Solved scenarios should be skipped
+        manager.solved.add("solved-scenario")
+        assert manager.should_skip("solved-scenario") is True
+        
+        # High-scoring scenarios should be skipped
+        manager.scores["easy-scenario"] = [0.9, 0.9, 0.9]
+        assert manager.should_skip("easy-scenario") is True
+        
+        # Low-scoring scenarios should not be skipped
+        manager.scores["hard-scenario"] = [0.3, 0.4, 0.5]
+        assert manager.should_skip("hard-scenario") is False
+        
+        # New scenarios should not be skipped
+        assert manager.should_skip("new-scenario") is False
+
+    def test_get_priority(self):
+        manager = CurriculumManager()
+        
+        # Solved scenarios have zero priority
+        manager.solved.add("solved")
+        assert manager.get_priority("solved") == 0.0
+        
+        # New scenarios have high priority
+        priority_new = manager.get_priority("new-scenario")
+        assert priority_new > 0.5
+        
+        # Difficult scenarios have higher priority
+        manager.scores["hard"] = [0.2, 0.3, 0.2]
+        manager.attempts["hard"] = 3
+        
+        manager.scores["easy"] = [0.8, 0.9, 0.85]
+        manager.attempts["easy"] = 3
+        
+        assert manager.get_priority("hard") > manager.get_priority("easy")
+
+    def test_reset(self):
+        manager = CurriculumManager()
+        
+        manager.solved.add("scenario-1")
+        manager.solved.add("scenario-2")
+        
+        manager.reset()
+        
+        assert len(manager.solved) == 0
+
+    def test_get_stats(self):
+        manager = CurriculumManager()
+        
+        manager.record_attempt("s1", 0.5)
+        manager.record_attempt("s1", 0.6)
+        manager.record_attempt("s2", 0.8)
+        
+        stats = manager.get_stats()
+        
+        assert stats["total_scenarios"] == 2
+        assert stats["total_attempts"] == 3
+        assert stats["avg_score"] == pytest.approx((0.5 + 0.6 + 0.8) / 3, rel=1e-3)
+
+    def test_checkpoint_save_load(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            checkpoint_path = Path(tmpdir) / "curriculum.json"
+            
+            # Create and populate manager
+            manager1 = CurriculumManager(checkpoint_path=str(checkpoint_path))
+            manager1.record_attempt("s1", 0.5)
+            manager1.record_attempt("s1", 0.6)
+            manager1.solved.add("s2")
+            manager1._save_checkpoint()
+            
+            # Load in new manager
+            manager2 = CurriculumManager(checkpoint_path=str(checkpoint_path))
+            
+            assert manager2.attempts["s1"] == 2
+            assert manager2.scores["s1"] == [0.5, 0.6]
+            assert "s2" in manager2.solved
+
+    def test_history_trimming(self):
+        manager = CurriculumManager(max_history_per_scenario=5)
+        
+        for i in range(10):
+            manager.record_attempt("scenario", float(i) / 10)
+        
+        assert len(manager.scores["scenario"]) == 5
+        # Should keep the most recent
+        assert manager.scores["scenario"][-1] == 0.9
+
+
+# =============================================================================
+# ScenarioPool Tests
+# =============================================================================
+
+
+class TestScenarioPool:
+    """Tests for ScenarioPool"""
+
+    def test_creation(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        assert len(pool.scenarios) == 0
+        assert pool._sample_counter == 0
+
+    def test_generate_synthetic_batch(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        scenarios = pool.generate_synthetic_batch(count=10)
+        
+        assert len(scenarios) == 10
+        for scenario in scenarios:
+            assert scenario.source == "synthetic"
+            assert len(scenario.markets) > 0
+            assert len(scenario.perpetuals) > 0
+            assert scenario.id.startswith("synth-")
+
+    def test_generate_with_archetype_focus(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        scenarios = pool.generate_synthetic_batch(count=5, archetype_focus="degen")
+        
+        for scenario in scenarios:
+            assert scenario.archetype_focus == "degen"
+
+    def test_difficulty_distribution(self):
+        config = ScenarioPoolConfig(
+            synthetic_difficulty_distribution={"easy": 0.5, "medium": 0.3, "hard": 0.2}
+        )
+        pool = ScenarioPool(config)
+        
+        scenarios = pool.generate_synthetic_batch(count=100)
+        
+        easy_count = sum(1 for s in scenarios if s.difficulty == "easy")
+        medium_count = sum(1 for s in scenarios if s.difficulty == "medium")
+        hard_count = sum(1 for s in scenarios if s.difficulty == "hard")
+        
+        # Allow some variance
+        assert 40 <= easy_count <= 60
+        assert 20 <= medium_count <= 40
+        assert 10 <= hard_count <= 30
+
+    def test_sample_without_curriculum(self):
+        config = ScenarioPoolConfig(use_curriculum=False)
+        pool = ScenarioPool(config)
+        pool.scenarios = pool.generate_synthetic_batch(count=20)
+        
+        sampled = pool.sample(count=5)
+        
+        assert len(sampled) == 5
+        for s in sampled:
+            assert s in pool.scenarios
+
+    def test_sample_with_curriculum(self):
+        config = ScenarioPoolConfig(use_curriculum=True)
+        pool = ScenarioPool(config)
+        pool.scenarios = pool.generate_synthetic_batch(count=10)
+        
+        # Initially all scenarios should be available
+        sampled = pool.sample(count=3)
+        assert len(sampled) == 3
+        
+        # Record good scores for some scenarios
+        for s in sampled[:2]:
+            pool.curriculum.record_attempt(s.id, 0.9)
+            pool.curriculum.record_attempt(s.id, 0.9)
+            pool.curriculum.record_attempt(s.id, 0.9)
+        
+        # Those should now be skipped
+        for _ in range(10):
+            new_sampled = pool.sample(count=5)
+            # Solved scenarios should have lower probability
+            solved_ids = {sampled[0].id, sampled[1].id}
+            sampled_ids = {s.id for s in new_sampled}
+            # They might still appear due to probability, but should be less frequent
+
+    def test_record_results(self):
+        config = ScenarioPoolConfig(use_curriculum=True)
+        pool = ScenarioPool(config)
+        pool.scenarios = pool.generate_synthetic_batch(count=5)
+        
+        ids = [s.id for s in pool.scenarios[:3]]
+        scores = [0.5, 0.7, 0.9]
+        
+        pool.record_results(ids, scores)
+        
+        assert pool.curriculum.attempts[ids[0]] == 1
+        assert pool.curriculum.scores[ids[0]][0] == 0.5
+
+    def test_get_stats(self):
+        config = ScenarioPoolConfig(use_curriculum=True)
+        pool = ScenarioPool(config)
+        pool.scenarios = pool.generate_synthetic_batch(count=10)
+        
+        stats = pool.get_stats()
+        
+        assert stats["total_scenarios"] == 10
+        assert stats["synthetic_scenarios"] == 10
+        assert stats["production_scenarios"] == 0
+        assert "curriculum" in stats
+
+    def test_save_and_load_scenarios(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            save_path = Path(tmpdir) / "scenarios.json"
+            
+            config = ScenarioPoolConfig()
+            pool1 = ScenarioPool(config)
+            pool1.scenarios = pool1.generate_synthetic_batch(count=5)
+            original_ids = [s.id for s in pool1.scenarios]
+            
+            pool1.save_scenarios(str(save_path))
+            
+            pool2 = ScenarioPool(config)
+            pool2.load_scenarios(str(save_path))
+            
+            loaded_ids = [s.id for s in pool2.scenarios]
+            
+            assert original_ids == loaded_ids
+            assert len(pool2.scenarios) == 5
+
+    def test_refresh_mechanism(self):
+        config = ScenarioPoolConfig(refresh_interval=5)
+        pool = ScenarioPool(config)
+        pool.scenarios = pool.generate_synthetic_batch(count=10)
+        
+        original_ids = {s.id for s in pool.scenarios}
+        
+        # Sample 5 times (reaches refresh interval)
+        for _ in range(5):
+            pool.sample(count=1)
+        
+        # Synthetic scenarios should be regenerated
+        new_ids = {s.id for s in pool.scenarios}
+        
+        # At least some should be different
+        assert original_ids != new_ids
+
+    @pytest.mark.asyncio
+    async def test_initialize_without_database(self):
+        config = ScenarioPoolConfig(max_scenarios=10)
+        pool = ScenarioPool(config)
+        
+        await pool.initialize()
+        
+        # Should fill with synthetic scenarios
+        assert len(pool.scenarios) == 10
+        for s in pool.scenarios:
+            assert s.source == "synthetic"
+
+
+class TestScenarioGeneration:
+    """Tests for scenario content generation"""
+
+    def test_random_market_generation(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        market = pool._generate_random_market(0)
+        
+        assert market.market_id == "market-1"
+        assert 0.2 <= market.yes_price <= 0.8
+        assert market.yes_price + market.no_price == pytest.approx(1.0, rel=1e-3)
+        assert market.volume_24h > 0
+        assert market.liquidity > 0
+
+    def test_default_perpetuals_generation(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        perps = pool._generate_default_perpetuals()
+        
+        assert len(perps) == 5  # BTC, ETH, SOL, DOGE, AVAX
+        tickers = {p.ticker for p in perps}
+        assert "BTC" in tickers
+        assert "ETH" in tickers
+
+    def test_news_generation(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        news = pool._generate_random_news(5, "medium")
+        
+        assert len(news) == 5
+        for item in news:
+            assert item.headline
+            assert item.sentiment in ["bullish", "bearish", "neutral"]
+            assert item.impact in ["high", "medium", "low"]
+            assert item.source
+
+    def test_posts_generation(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        posts = pool._generate_random_posts(6)
+        
+        assert len(posts) == 6
+        for post in posts:
+            assert post.author
+            assert post.content
+            assert post.sentiment in ["bullish", "bearish", "neutral"]
+            assert post.likes >= 0
+            assert post.replies >= 0
+
+    def test_contextual_news_generation(self):
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        
+        markets = [
+            MarketState(
+                market_id="m1",
+                question="Will BTC exceed $100K?",
+                yes_price=0.5,
+                no_price=0.5,
+                volume_24h=100000.0,
+                liquidity=500000.0,
+                expires_at=1735689600000,
+            )
+        ]
+        
+        news = pool._generate_contextual_news(markets)
+        
+        assert len(news) > 0
+        # Should have at least one BTC-related news item
+        btc_news = [n for n in news if "bitcoin" in n.headline.lower() or "btc" in n.headline.lower()]
+        assert len(btc_news) >= 1
+
+
+# =============================================================================
+# Integration Tests
+# =============================================================================
+
+
+class TestScenarioPoolIntegration:
+    """Integration tests for ScenarioPool"""
+
+    @pytest.mark.asyncio
+    async def test_full_workflow(self):
+        """Test complete workflow: init, sample, record, refresh"""
+        with tempfile.TemporaryDirectory() as tmpdir:
+            config = ScenarioPoolConfig(
+                max_scenarios=20,
+                refresh_interval=10,
+                use_curriculum=True,
+                curriculum_checkpoint_path=str(Path(tmpdir) / "curriculum.json"),
+            )
+            pool = ScenarioPool(config)
+            
+            await pool.initialize()
+            assert len(pool.scenarios) == 20
+            
+            # Sample and record results
+            for _ in range(5):
+                scenarios = pool.sample(count=2)
+                ids = [s.id for s in scenarios]
+                scores = [0.6, 0.8]
+                pool.record_results(ids, scores)
+            
+            stats = pool.get_stats()
+            assert stats["curriculum"]["total_attempts"] == 10
+            
+            # Save and reload
+            save_path = str(Path(tmpdir) / "scenarios.json")
+            pool.save_scenarios(save_path)
+            
+            pool2 = ScenarioPool(config)
+            pool2.load_scenarios(save_path)
+            assert len(pool2.scenarios) == 20
+
+    def test_scenario_observation_format(self):
+        """Test that observations have correct format for agent consumption"""
+        config = ScenarioPoolConfig()
+        pool = ScenarioPool(config)
+        scenarios = pool.generate_synthetic_batch(count=1)
+        scenario = scenarios[0]
+        
+        obs = scenario.to_observation()
+        
+        # Verify required fields
+        required_fields = ["timestamp", "markets", "perpetuals", "news", "socialFeed", "portfolio", "marketSummary"]
+        for field in required_fields:
+            assert field in obs, f"Missing required field: {field}"
+        
+        # Verify nested structure
+        if obs["markets"]:
+            market = obs["markets"][0]
+            assert "id" in market
+            assert "yesPrice" in market
+            assert "question" in market
+        
+        if obs["perpetuals"]:
+            perp = obs["perpetuals"][0]
+            assert "ticker" in perp
+            assert "markPrice" in perp
+        
+        assert "balance" in obs["portfolio"]
+
+