da30ef19ed
Architecture Docker (8 services), FastAPI, TimescaleDB, Redis, Streamlit. Stratégies : scalping, intraday, swing. MLEngine + RegimeDetector (HMM). BacktestEngine + WalkForwardAnalyzer + Optuna optimizer. Routes API complètes dont /optimize async. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
474 lines
16 KiB
Python
474 lines
16 KiB
Python
"""
|
|
Tests Unitaires - RegimeDetector.
|
|
|
|
Tests de la détection de régimes de marché avec HMM.
|
|
"""
|
|
|
|
import pytest
|
|
import pandas as pd
|
|
import numpy as np
|
|
from datetime import datetime, timedelta
|
|
|
|
from src.ml.regime_detector import RegimeDetector
|
|
|
|
|
|
class TestRegimeDetectorInitialization:
|
|
"""Tests d'initialisation du RegimeDetector."""
|
|
|
|
def test_initialization_default(self):
|
|
"""Test initialisation avec paramètres par défaut."""
|
|
detector = RegimeDetector()
|
|
|
|
assert detector.n_regimes == 4
|
|
assert detector.random_state == 42
|
|
assert detector.is_fitted is False
|
|
assert len(detector.feature_names) == 0
|
|
|
|
def test_initialization_custom_regimes(self):
|
|
"""Test initialisation avec nombre de régimes personnalisé."""
|
|
detector = RegimeDetector(n_regimes=3)
|
|
|
|
assert detector.n_regimes == 3
|
|
|
|
def test_regime_names_defined(self):
|
|
"""Test que les noms de régimes sont définis."""
|
|
detector = RegimeDetector()
|
|
|
|
assert len(detector.REGIME_NAMES) == 4
|
|
assert 'Trending Up' in detector.REGIME_NAMES.values()
|
|
assert 'Trending Down' in detector.REGIME_NAMES.values()
|
|
assert 'Ranging' in detector.REGIME_NAMES.values()
|
|
assert 'High Volatility' in detector.REGIME_NAMES.values()
|
|
|
|
|
|
class TestRegimeDetectorFitting:
|
|
"""Tests d'entraînement du modèle."""
|
|
|
|
@pytest.fixture
|
|
def sample_data(self):
|
|
"""Génère des données de test."""
|
|
dates = pd.date_range(start='2024-01-01', periods=200, freq='1H')
|
|
|
|
np.random.seed(42)
|
|
returns = np.random.normal(0.0001, 0.01, 200)
|
|
prices = 1.1000 * np.exp(np.cumsum(returns))
|
|
|
|
df = pd.DataFrame(index=dates)
|
|
df['close'] = prices
|
|
df['open'] = df['close'].shift(1).fillna(df['close'].iloc[0])
|
|
df['high'] = df[['open', 'close']].max(axis=1) * 1.001
|
|
df['low'] = df[['open', 'close']].min(axis=1) * 0.999
|
|
df['volume'] = np.random.randint(1000, 10000, 200)
|
|
|
|
return df
|
|
|
|
def test_fit_success(self, sample_data):
|
|
"""Test entraînement réussi."""
|
|
detector = RegimeDetector()
|
|
|
|
detector.fit(sample_data)
|
|
|
|
assert detector.is_fitted is True
|
|
assert len(detector.feature_names) > 0
|
|
|
|
def test_fit_creates_features(self, sample_data):
|
|
"""Test que fit crée les features."""
|
|
detector = RegimeDetector()
|
|
|
|
detector.fit(sample_data)
|
|
|
|
# Vérifier que les features attendues sont créées
|
|
expected_features = ['returns', 'volatility', 'trend', 'range', 'volume_change', 'momentum']
|
|
|
|
for feature in expected_features:
|
|
assert feature in detector.feature_names
|
|
|
|
def test_fit_with_insufficient_data(self):
|
|
"""Test avec données insuffisantes."""
|
|
detector = RegimeDetector()
|
|
|
|
# Données trop courtes
|
|
dates = pd.date_range(start='2024-01-01', periods=10, freq='1H')
|
|
df = pd.DataFrame({
|
|
'close': np.random.randn(10),
|
|
'open': np.random.randn(10),
|
|
'high': np.random.randn(10),
|
|
'low': np.random.randn(10),
|
|
'volume': np.random.randint(1000, 10000, 10)
|
|
}, index=dates)
|
|
|
|
# Devrait lever une erreur ou gérer gracieusement
|
|
try:
|
|
detector.fit(df)
|
|
# Si pas d'erreur, vérifier que le modèle n'est pas fitted
|
|
# ou qu'il y a un warning
|
|
except Exception as e:
|
|
# Acceptable
|
|
pass
|
|
|
|
|
|
class TestRegimeDetectorPrediction:
|
|
"""Tests de prédiction des régimes."""
|
|
|
|
@pytest.fixture
|
|
def fitted_detector(self, sample_data):
|
|
"""Retourne un détecteur entraîné."""
|
|
detector = RegimeDetector()
|
|
detector.fit(sample_data)
|
|
return detector
|
|
|
|
@pytest.fixture
|
|
def sample_data(self):
|
|
"""Génère des données de test."""
|
|
dates = pd.date_range(start='2024-01-01', periods=200, freq='1H')
|
|
|
|
np.random.seed(42)
|
|
returns = np.random.normal(0.0001, 0.01, 200)
|
|
prices = 1.1000 * np.exp(np.cumsum(returns))
|
|
|
|
df = pd.DataFrame(index=dates)
|
|
df['close'] = prices
|
|
df['open'] = df['close'].shift(1).fillna(df['close'].iloc[0])
|
|
df['high'] = df[['open', 'close']].max(axis=1) * 1.001
|
|
df['low'] = df[['open', 'close']].min(axis=1) * 0.999
|
|
df['volume'] = np.random.randint(1000, 10000, 200)
|
|
|
|
return df
|
|
|
|
def test_predict_regime_returns_array(self, fitted_detector, sample_data):
|
|
"""Test que predict_regime retourne un array."""
|
|
regimes = fitted_detector.predict_regime(sample_data)
|
|
|
|
assert isinstance(regimes, np.ndarray)
|
|
assert len(regimes) > 0
|
|
|
|
def test_predict_regime_values_valid(self, fitted_detector, sample_data):
|
|
"""Test que les régimes prédits sont valides."""
|
|
regimes = fitted_detector.predict_regime(sample_data)
|
|
|
|
# Tous les régimes doivent être entre 0 et n_regimes-1
|
|
assert (regimes >= 0).all()
|
|
assert (regimes < fitted_detector.n_regimes).all()
|
|
|
|
def test_predict_current_regime(self, fitted_detector, sample_data):
|
|
"""Test prédiction du régime actuel."""
|
|
current_regime = fitted_detector.predict_current_regime(sample_data)
|
|
|
|
assert isinstance(current_regime, (int, np.integer))
|
|
assert 0 <= current_regime < fitted_detector.n_regimes
|
|
|
|
def test_predict_without_fitting(self, sample_data):
|
|
"""Test prédiction sans entraînement préalable."""
|
|
detector = RegimeDetector()
|
|
|
|
with pytest.raises(ValueError, match="not fitted"):
|
|
detector.predict_regime(sample_data)
|
|
|
|
def test_get_regime_probabilities(self, fitted_detector, sample_data):
|
|
"""Test obtention des probabilités."""
|
|
probabilities = fitted_detector.get_regime_probabilities(sample_data)
|
|
|
|
assert isinstance(probabilities, np.ndarray)
|
|
assert probabilities.shape[1] == fitted_detector.n_regimes
|
|
|
|
# Vérifier que les probabilités somment à 1
|
|
prob_sums = probabilities.sum(axis=1)
|
|
np.testing.assert_array_almost_equal(prob_sums, np.ones(len(prob_sums)), decimal=5)
|
|
|
|
|
|
class TestRegimeDetectorStatistics:
|
|
"""Tests des statistiques de régimes."""
|
|
|
|
@pytest.fixture
|
|
def fitted_detector(self, sample_data):
|
|
"""Retourne un détecteur entraîné."""
|
|
detector = RegimeDetector()
|
|
detector.fit(sample_data)
|
|
return detector
|
|
|
|
@pytest.fixture
|
|
def sample_data(self):
|
|
"""Génère des données de test."""
|
|
dates = pd.date_range(start='2024-01-01', periods=200, freq='1H')
|
|
|
|
np.random.seed(42)
|
|
returns = np.random.normal(0.0001, 0.01, 200)
|
|
prices = 1.1000 * np.exp(np.cumsum(returns))
|
|
|
|
df = pd.DataFrame(index=dates)
|
|
df['close'] = prices
|
|
df['open'] = df['close'].shift(1).fillna(df['close'].iloc[0])
|
|
df['high'] = df[['open', 'close']].max(axis=1) * 1.001
|
|
df['low'] = df[['open', 'close']].min(axis=1) * 0.999
|
|
df['volume'] = np.random.randint(1000, 10000, 200)
|
|
|
|
return df
|
|
|
|
def test_get_regime_name(self, fitted_detector):
|
|
"""Test récupération du nom d'un régime."""
|
|
for regime in range(fitted_detector.n_regimes):
|
|
name = fitted_detector.get_regime_name(regime)
|
|
assert isinstance(name, str)
|
|
assert len(name) > 0
|
|
|
|
def test_get_regime_statistics(self, fitted_detector, sample_data):
|
|
"""Test calcul des statistiques."""
|
|
stats = fitted_detector.get_regime_statistics(sample_data)
|
|
|
|
assert 'regime_counts' in stats
|
|
assert 'regime_percentages' in stats
|
|
assert 'current_regime' in stats
|
|
assert 'current_regime_name' in stats
|
|
|
|
# Vérifier que les pourcentages somment à 1
|
|
total_pct = sum(stats['regime_percentages'].values())
|
|
assert abs(total_pct - 1.0) < 0.01
|
|
|
|
|
|
class TestRegimeDetectorAdaptation:
|
|
"""Tests d'adaptation des paramètres."""
|
|
|
|
def test_adapt_strategy_parameters(self):
|
|
"""Test adaptation des paramètres selon le régime."""
|
|
detector = RegimeDetector()
|
|
|
|
base_params = {
|
|
'min_confidence': 0.6,
|
|
'risk_per_trade': 0.02
|
|
}
|
|
|
|
# Tester pour chaque régime
|
|
for regime in range(4):
|
|
adapted = detector.adapt_strategy_parameters(regime, base_params)
|
|
|
|
assert 'min_confidence' in adapted
|
|
assert 'risk_per_trade' in adapted
|
|
|
|
# Les paramètres doivent être modifiés
|
|
assert adapted != base_params
|
|
|
|
def test_adapt_trending_up(self):
|
|
"""Test adaptation pour régime Trending Up."""
|
|
detector = RegimeDetector()
|
|
|
|
base_params = {
|
|
'min_confidence': 0.6,
|
|
'risk_per_trade': 0.02
|
|
}
|
|
|
|
adapted = detector.adapt_strategy_parameters(0, base_params) # 0 = Trending Up
|
|
|
|
# Devrait être plus agressif
|
|
assert adapted['min_confidence'] < base_params['min_confidence']
|
|
assert adapted['risk_per_trade'] > base_params['risk_per_trade']
|
|
|
|
def test_adapt_high_volatility(self):
|
|
"""Test adaptation pour régime High Volatility."""
|
|
detector = RegimeDetector()
|
|
|
|
base_params = {
|
|
'min_confidence': 0.6,
|
|
'risk_per_trade': 0.02
|
|
}
|
|
|
|
adapted = detector.adapt_strategy_parameters(3, base_params) # 3 = High Volatility
|
|
|
|
# Devrait être plus conservateur
|
|
assert adapted['min_confidence'] > base_params['min_confidence']
|
|
assert adapted['risk_per_trade'] < base_params['risk_per_trade']
|
|
|
|
def test_should_trade_in_regime(self):
|
|
"""Test décision de trading selon régime."""
|
|
detector = RegimeDetector()
|
|
|
|
# Scalping devrait trader en Ranging
|
|
assert detector.should_trade_in_regime(2, 'scalping') is True
|
|
|
|
# Scalping ne devrait pas trader en High Volatility
|
|
assert detector.should_trade_in_regime(3, 'scalping') is False
|
|
|
|
# Intraday devrait trader en Trending
|
|
assert detector.should_trade_in_regime(0, 'intraday') is True
|
|
assert detector.should_trade_in_regime(1, 'intraday') is True
|
|
|
|
# Intraday ne devrait pas trader en Ranging
|
|
assert detector.should_trade_in_regime(2, 'intraday') is False
|
|
|
|
|
|
class TestRegimeDetectorFeatures:
|
|
"""Tests de calcul des features."""
|
|
|
|
@pytest.fixture
|
|
def sample_data(self):
|
|
"""Génère des données de test."""
|
|
dates = pd.date_range(start='2024-01-01', periods=200, freq='1H')
|
|
|
|
np.random.seed(42)
|
|
returns = np.random.normal(0.0001, 0.01, 200)
|
|
prices = 1.1000 * np.exp(np.cumsum(returns))
|
|
|
|
df = pd.DataFrame(index=dates)
|
|
df['close'] = prices
|
|
df['open'] = df['close'].shift(1).fillna(df['close'].iloc[0])
|
|
df['high'] = df[['open', 'close']].max(axis=1) * 1.001
|
|
df['low'] = df[['open', 'close']].min(axis=1) * 0.999
|
|
df['volume'] = np.random.randint(1000, 10000, 200)
|
|
|
|
return df
|
|
|
|
def test_calculate_features(self, sample_data):
|
|
"""Test calcul des features."""
|
|
detector = RegimeDetector()
|
|
|
|
features = detector._calculate_features(sample_data)
|
|
|
|
assert isinstance(features, pd.DataFrame)
|
|
assert len(features) > 0
|
|
|
|
# Vérifier présence des features
|
|
expected_features = ['returns', 'volatility', 'trend', 'range', 'volume_change', 'momentum']
|
|
|
|
for feature in expected_features:
|
|
assert feature in features.columns
|
|
|
|
def test_features_no_nan(self, sample_data):
|
|
"""Test que les features n'ont pas de NaN après nettoyage."""
|
|
detector = RegimeDetector()
|
|
|
|
features = detector._calculate_features(sample_data)
|
|
|
|
# Après dropna, ne devrait pas y avoir de NaN
|
|
assert features.isna().sum().sum() == 0
|
|
|
|
def test_normalize_features(self):
|
|
"""Test normalisation des features."""
|
|
detector = RegimeDetector()
|
|
|
|
# Créer features test
|
|
X = np.random.randn(100, 6)
|
|
|
|
X_normalized = detector._normalize_features(X)
|
|
|
|
# Vérifier que la moyenne est proche de 0 et std proche de 1
|
|
assert abs(X_normalized.mean()) < 0.1
|
|
assert abs(X_normalized.std() - 1.0) < 0.1
|
|
|
|
|
|
class TestRegimeDetectorEdgeCases:
|
|
"""Tests des cas limites."""
|
|
|
|
def test_with_missing_columns(self):
|
|
"""Test avec colonnes manquantes."""
|
|
detector = RegimeDetector()
|
|
|
|
# DataFrame incomplet
|
|
df = pd.DataFrame({
|
|
'close': np.random.randn(100)
|
|
# Manque open, high, low, volume
|
|
})
|
|
|
|
with pytest.raises(KeyError):
|
|
detector.fit(df)
|
|
|
|
def test_with_constant_prices(self):
|
|
"""Test avec prix constants."""
|
|
detector = RegimeDetector()
|
|
|
|
dates = pd.date_range(start='2024-01-01', periods=200, freq='1H')
|
|
|
|
df = pd.DataFrame(index=dates)
|
|
df['close'] = 1.1000 # Prix constant
|
|
df['open'] = 1.1000
|
|
df['high'] = 1.1000
|
|
df['low'] = 1.1000
|
|
df['volume'] = 1000
|
|
|
|
# Devrait gérer gracieusement (ou lever erreur appropriée)
|
|
try:
|
|
detector.fit(df)
|
|
# Si réussit, vérifier que le modèle est fitted
|
|
assert detector.is_fitted
|
|
except Exception:
|
|
# Acceptable si erreur appropriée
|
|
pass
|
|
|
|
def test_regime_name_invalid(self):
|
|
"""Test avec numéro de régime invalide."""
|
|
detector = RegimeDetector()
|
|
|
|
# Régime hors limites
|
|
name = detector.get_regime_name(999)
|
|
|
|
# Devrait retourner un nom par défaut
|
|
assert 'Regime' in name
|
|
|
|
|
|
class TestRegimeDetectorIntegration:
|
|
"""Tests d'intégration."""
|
|
|
|
@pytest.fixture
|
|
def sample_data(self):
|
|
"""Génère des données de test."""
|
|
dates = pd.date_range(start='2024-01-01', periods=500, freq='1H')
|
|
|
|
np.random.seed(42)
|
|
|
|
# Créer différents régimes
|
|
regimes = []
|
|
prices = []
|
|
base_price = 1.1000
|
|
|
|
for i in range(500):
|
|
if i < 125: # Trending Up
|
|
regime = 0
|
|
price = base_price * (1 + i * 0.0001)
|
|
elif i < 250: # Trending Down
|
|
regime = 1
|
|
price = base_price * (1 - (i - 125) * 0.0001)
|
|
elif i < 375: # Ranging
|
|
regime = 2
|
|
price = base_price + 0.001 * np.sin(i / 10)
|
|
else: # High Volatility
|
|
regime = 3
|
|
price = base_price + 0.01 * np.random.randn()
|
|
|
|
regimes.append(regime)
|
|
prices.append(price)
|
|
|
|
df = pd.DataFrame(index=dates)
|
|
df['close'] = prices
|
|
df['open'] = df['close'].shift(1).fillna(df['close'].iloc[0])
|
|
df['high'] = df[['open', 'close']].max(axis=1) * 1.001
|
|
df['low'] = df[['open', 'close']].min(axis=1) * 0.999
|
|
df['volume'] = np.random.randint(1000, 10000, 500)
|
|
|
|
return df
|
|
|
|
def test_full_workflow(self, sample_data):
|
|
"""Test workflow complet."""
|
|
detector = RegimeDetector(n_regimes=4)
|
|
|
|
# 1. Fit
|
|
detector.fit(sample_data)
|
|
assert detector.is_fitted
|
|
|
|
# 2. Predict
|
|
regimes = detector.predict_regime(sample_data)
|
|
assert len(regimes) > 0
|
|
|
|
# 3. Current regime
|
|
current = detector.predict_current_regime(sample_data)
|
|
assert 0 <= current < 4
|
|
|
|
# 4. Statistics
|
|
stats = detector.get_regime_statistics(sample_data)
|
|
assert 'current_regime' in stats
|
|
|
|
# 5. Adaptation
|
|
adapted = detector.adapt_strategy_parameters(current, {'min_confidence': 0.6})
|
|
assert 'min_confidence' in adapted
|
|
|
|
# 6. Should trade
|
|
should_trade = detector.should_trade_in_regime(current, 'intraday')
|
|
assert isinstance(should_trade, bool)
|