tests/test_rl_engine_extra.py

# tests/test_rl_engine_extra.py
import pytest
from datetime import datetime, date, timedelta, timezone
from unittest.mock import MagicMock, patch

from app.services.rl_engine import (
    _compute_circadian_bonus,
    _apply_memory_decay,
    CIRCADIAN_WINDOWS,
    MEMORY_DECAY_GAMMA
)

def make_user(user_id="testuser", rl_bandit_params=None):
    user = MagicMock()
    user.user_id = user_id
    user.rl_bandit_params = rl_bandit_params
    user.rl_history_count = 10
    user.rl_last_decay_date = None
    user.days_since_last_login = 0
    user.current_streak = 0
    return user

def test_compute_circadian_bonus_no_db():
    user = make_user()
    bonus = _compute_circadian_bonus(user, None, "Deep Work")
    assert bonus == 0.0

def test_compute_circadian_bonus_wrong_action():
    user = make_user()
    db = MagicMock()
    bonus = _compute_circadian_bonus(user, db, "Normal")
    assert bonus == 0.0

@patch("app.services.rl_engine.datetime")
def test_compute_circadian_bonus_with_db(mock_dt):
    # Mock datetime to a specific hour, say 14 (afternoon)
    mock_now = datetime(2023, 1, 1, 14, 0, 0)
    mock_dt.now.return_value = mock_now
    
    class FakeProgress:
        def __init__(self, hour, satisfaction):
            self.completed_at = datetime(2023, 1, 1, hour, 0, 0, tzinfo=timezone.utc)
            self.satisfaction = satisfaction
            
    rows = [
        FakeProgress(14, 1.0),
        FakeProgress(15, 0.9),
        FakeProgress(16, 1.0),
        FakeProgress(8, 0.2),
        FakeProgress(9, 0.3),
        FakeProgress(22, 0.1),
        FakeProgress(23, 0.2),
    ]
    
    db = MagicMock()
    db.query.return_value.filter.return_value.all.return_value = rows
    
    user = make_user()
    bonus = _compute_circadian_bonus(user, db, "Deep Work")
    assert bonus > 0
    assert bonus <= 3.0

def test_apply_memory_decay_no_decay_needed():
    user = make_user()
    user.last_login_at = datetime.now(timezone.utc) - timedelta(days=1)
    user.rl_last_decay_date = (datetime.now(timezone.utc) - timedelta(days=1)).isoformat()
    params = {
        "Deep Work": {"alpha": 10.0, "beta": 10.0},
        "Learning": {"alpha": 5.0, "beta": 5.0},
        "Maintenance": {"alpha": 3.0, "beta": 3.0},
        "Recovery": {"alpha": 1.0, "beta": 1.0},
    }
    db = MagicMock()
    
    new_params = _apply_memory_decay(user, params, db)
    assert new_params["Deep Work"]["alpha"] == 10.0

def test_apply_memory_decay_performs_decay():
    user = make_user()
    user.last_login_at = datetime.now(timezone.utc) - timedelta(days=10)
    params = {
        "Deep Work": {"alpha": 10.0, "beta": 10.0},
        "Learning": {"alpha": 5.0, "beta": 5.0},
        "Maintenance": {"alpha": 3.0, "beta": 3.0},
        "Recovery": {"alpha": 1.0, "beta": 1.0},
    }
    db = MagicMock()
    
    new_params = _apply_memory_decay(user, params, db)
    if new_params["Deep Work"]["alpha"] == 10.0:
        print("Decay failed to apply. Params returned unchanged.")
        
    expected_alpha = 10.0 * MEMORY_DECAY_GAMMA + 2.0 * (1 - MEMORY_DECAY_GAMMA)
    expected_beta = 10.0 * MEMORY_DECAY_GAMMA + 2.0 * (1 - MEMORY_DECAY_GAMMA)
    assert abs(new_params["Deep Work"]["alpha"] - expected_alpha) < 0.001
    assert abs(new_params["Deep Work"]["beta"] - expected_beta) < 0.001


def test_get_next_day_signal_does_not_crash():
    from app.services.rl_engine import get_next_day_signal
    user = make_user()
    user.competence = 0.5
    user.mood = 0.5
    user.fatigue = 0.5
    user.sleep_quality = 0.5
    user.shock = 0.0
    user.sick = 0.0
    user.stress = 0.0
    
    # Should not raise TypeError
    signal = get_next_day_signal(user)
    assert signal["preview"] is True
    assert "action_label" in signal


@patch("app.services.rl_engine._compute_historical_features")
def test_reward_hacking_discount(mock_hist):
    from app.services.rl_engine import update_reward
    user = make_user("hacker", "{}")
    db = MagicMock()
    
    # Simulate high fatigue to trigger reward hacking squash
    mock_hist.return_value = {"ema_fatigue": 0.9}
    
    update_reward(db, user, "Normal", 1.0)
    
    # 1.0 should be squashed toward 0.5 -> 0.75
    # Default alpha is 3.0 for Normal. alpha += 0.75 -> 3.75
    import json
    params = json.loads(user.rl_bandit_params)
    assert params["Normal"]["alpha"] < 4.0
    assert params["Normal"]["alpha"] >= 3.70


@patch("app.services.rl_engine._compute_historical_features")
def test_circuit_breaker_debt(mock_hist):
    from app.services.rl_engine import get_behavior_signal
    user = make_user("burnout", '{"_recovery_debt": 16.0}')
    db = MagicMock()
    mock_hist.return_value = {
        "consecutive_hard_days": 5.0,
        "recent_completion_rate": 0.5,
        "avg_satisfaction_7d": 0.5,
        "fatigue_trend": 0.0,
        "overwork_signal": 0.0,
        "deep_work_streak": 0.0,
        "ema_fatigue": 0.0,
        "ema_mood": 0.5,
        "last_action": "Deep Work"
    }
    
    signal = get_behavior_signal(user, state_override=[0.5, 0.5, 0.9, 0.5, 0.0, 0.0, 0.0], db=db)
    # Debt is > 15, so Deep Work should be circuit broken into Recovery.
    # Also fatigue is 0.9, triggering ceiling.
    assert signal["action_label"] in ("Recovery", "Light Review")