def test_bet():
"""Test the bet method."""
bettor = ClassifierBettor(
DummyClassifier(strategy='constant', constant=[False, True,
True])).fit(X, Y)
np.testing.assert_array_equal(bettor.bet(X, O),
np.array([False, True, True]) * (O > 1))
def test_backtest_params(n_splits):
"""Test the backtest parameters."""
bettor = ClassifierBettor(DummyClassifier()).backtest(
X, Y, O, tscv=TimeSeriesSplit(n_splits=n_splits), init_cash=1e5)
assert isinstance(bettor.tscv_, TimeSeriesSplit)
assert bettor.tscv_.n_splits == n_splits
assert bettor.init_cash_ == 1e5
def test_backtest_params_non_df_raise_value_error():
"""Test raising an error on the wrong backtest input data."""
with pytest.raises(
ValueError,
match='Input data `X` should be pandas dataframe with a date index.'
):
ClassifierBettor(DummyClassifier()).backtest(X.values, Y, O)
def test_backtest_params_init_cash_raise_value_error(init_cash):
"""Test raising an error on the wrong backtest params."""
with pytest.raises(ValueError,
match=f"init_cash == {init_cash}, must be > 0.0."):
ClassifierBettor(DummyClassifier()).backtest(X,
Y,
O,
init_cash=init_cash)
def test_backtest_params_tscv_raise_error(tscv):
"""Test raising an error on the wrong backtest tscv param."""
with pytest.raises(
TypeError,
match=
'Parameter `tscv` should be a TimeSeriesSplit cross-validator object.',
):
ClassifierBettor(DummyClassifier()).backtest(X, Y, O, tscv=tscv)
def test_fit_raise_error(classifier):
"""Test raising an error on the wrong classifier type."""
with pytest.raises(
TypeError,
match='`ClassifierBettor` requires a classifier. '
f'Instead {type(classifier)} is given.',
):
ClassifierBettor(classifier).fit(X, Y)
def test_bet_raise_not_fitted_error():
"""Test raising of not fitted error."""
with pytest.raises(
NotFittedError,
match="This ClassifierBettor instance is"
" not fitted yet. Call 'fit' with appropriate arguments before"
" using this estimator",
):
ClassifierBettor(DummyClassifier()).bet(X, O)
def test_backtest_params_init_cash_raise_type_error(init_cash):
"""Test raising an error on the wrong backtest params."""
with pytest.raises(
TypeError,
match=re.escape(
"init_cash must be an instance of (<class 'float'>, <class 'int'>), "
f"not {type(init_cash)}."),
):
ClassifierBettor(DummyClassifier()).backtest(X,
Y,
O,
init_cash=init_cash)
def test_backtest_default_params():
"""Test the backtest default parameters."""
bettor = ClassifierBettor(DummyClassifier()).backtest(X, Y, O)
assert isinstance(bettor.tscv_, TimeSeriesSplit)
assert bettor.init_cash_ == 1e3
def test_fit_check_classifier():
"""Test the cloned classifier."""
clf = DummyClassifier()
bettor = ClassifierBettor(clf).fit(X, Y)
check_is_fitted(bettor.classifier_)
assert isinstance(bettor.classifier_, DummyClassifier)
def test_backtest():
"""Test the outcome of backtest."""
n_splits = 2
tscv = TimeSeriesSplit(n_splits)
clf = DummyClassifier(strategy='constant', constant=[True, False, True])
# Backtesting results
bettor = ClassifierBettor(clf).backtest(X, Y, O, tscv=tscv)
bettor.backtest(X, Y, O, tscv)
# Assertions
results = []
for train_ind, test_ind in tscv.split(X):
clf.fit(X.iloc[train_ind], Y.iloc[train_ind])
Y_pred_prob = np.concatenate(
[
prob[:, -1].reshape(-1, 1)
for prob in clf.predict_proba(X.iloc[test_ind])
],
axis=1,
)
value_bets = Y_pred_prob * O.values[test_ind] > 1
returns = np.nan_to_num(
(Y.values[test_ind] * O.values[test_ind] - 1) * value_bets)
dates = X.iloc[test_ind].index
returns_dates = pd.Series(returns.sum(axis=1),
index=dates).reindex(pd.date_range(
dates.min(), dates.max()),
fill_value=0)
drawdowns = qs.stats.drawdown_details(returns_dates)
max_drawdown = np.nan
max_drawdown_period = pd.NaT
if (drawdowns['max drawdown'] < 0).sum() > 0:
dind = drawdowns['max drawdown'].idxmin()
max_drawdown = -drawdowns['max drawdown'][dind] / bettor.init_cash_
mask = returns_dates[drawdowns['end'][dind]:] > 0
max_drawdown_period = mask.index[mask][0] - pd.Timestamp(
drawdowns['start'][dind])
results.append([
X.index[train_ind[0]],
X.index[train_ind[-1]],
X.index[train_ind[-1]] - X.index[train_ind[0]],
dates.min(),
dates.max(),
dates.max() - dates.min() + pd.Timedelta('1d'),
bettor.init_cash_,
bettor.init_cash_ + returns.sum(),
100 * returns.sum() / bettor.init_cash_,
max_drawdown,
max_drawdown_period,
(returns != 0).sum(),
100 * (returns > 0).sum() / (returns != 0).sum(),
returns.max() * 100,
returns.min() * 100,
returns[returns > 0].mean() * 100,
returns[returns < 0].mean() * 100,
-returns[returns > 0].sum() / returns[returns < 0].sum(),
qs.stats.sharpe(returns_dates,
periods=len(returns_dates),
rf=0.1,
smart=True),
returns[returns != 0].mean() * 100,
returns[returns != 0].std() * 100,
])
assert len(bettor.backtest_results_) == n_splits
pd.testing.assert_frame_equal(
bettor.backtest_results_,
pd.DataFrame(results, columns=bettor.backtest_results_.columns),
check_exact=False,
atol=0.02,
check_dtype=False,
)