def test_bet_size_reserve_return_params(self, mock_likely_parameters):
"""
Tests for successful execution of 'bet_size_reserve' using return_parameters=True.
Function 'most_likely_parameters' needs to be patched because the 'M2N.mp_fit' method makes use of
random numbers.
"""
# Setup the test DataFrame.
np.random.seed(0)
sample_size = 500
start_date = dt.datetime(2000, 1, 1)
date_step = dt.timedelta(days=1)
dates = np.array(
[start_date + i * date_step for i in range(sample_size)])
shift_dt = np.array([
dt.timedelta(days=d)
for d in np.random.uniform(1., 20., sample_size)
])
dates_shifted = dates + shift_dt
time_1 = pd.Series(data=dates_shifted, index=dates)
df_events = time_1.to_frame()
df_events = df_events.rename(columns={0: 't1'})
df_events['p'] = np.random.uniform(0.0, 1.0, sample_size)
df_events = df_events[['t1', 'p']]
df_events['side'] = df_events['p'].apply(lambda x: 1
if x >= 0.5 else -1)
# Calculate the correct results.
events_active = get_concurrent_sides(df_events['t1'],
df_events['side'])
events_active['c_t'] = events_active['active_long'] - events_active[
'active_short']
central_moments = [
moment(events_active['c_t'].to_numpy(), moment=i)
for i in range(1, 6)
]
raw_moments = raw_moment(central_moments=central_moments,
dist_mean=events_active['c_t'].mean())
m2n_test = M2N(raw_moments,
epsilon=1e-5,
factor=5,
n_runs=25,
variant=2,
max_iter=10_000,
num_workers=1)
test_results = m2n_test.mp_fit()
test_params = most_likely_parameters(test_results)
mock_likely_parameters.return_value = test_params
test_fit = [
test_params[key]
for key in ['mu_1', 'mu_2', 'sigma_1', 'sigma_2', 'p_1']
]
events_active['bet_size'] = events_active['c_t'].apply(
lambda c: single_bet_size_mixed(c, test_fit))
# Evaluate.
eval_events, eval_params = bet_size_reserve(df_events['t1'],
df_events['side'],
fit_runs=25,
return_parameters=True)
self.assertEqual(test_params, eval_params)
self.assertTrue(events_active.equals(eval_events))
def test_most_likely_parameters_list_arg(self):
"""
Tests the helper function 'most_likely_parameters' when passing a list to 'ignore_columns'.
"""
mu_1_list = [
-2.074149682208028, -2.1464760973734522, -1.7318027625411423,
-1.7799163398785354, -1.9766582333677596
]
mu_2_list = [
0.9958122958772418, 0.9927128514876395, 1.013574632526087,
1.0065707257309104, 1.009533655971151
]
sigma_1_list = [
1.9764097851543956, 1.9516780127056625, 2.080573657129795,
2.071328499049906, 1.9988591140726848
]
sigma_2_list = [
1.002964090440232, 1.0054392587806025, 0.9872577865302316,
0.9909001363163131, 0.9971327048101786
]
p_1_list = [
0.09668610445835334, 0.09379917992315062, 0.11351960785118335,
0.10993400151299484, 0.10264463363929438
]
df_test = pd.DataFrame.from_dict({
'mu_1': mu_1_list,
'mu_2': mu_2_list,
'sigma_1': sigma_1_list,
'sigma_2': sigma_2_list,
'p_1': p_1_list
})
most_likely_correct = {
'mu_1': -2.03765,
'mu_2': 1.00863,
'sigma_1': 1.9832,
'sigma_2': 1.00012,
'p_1': 0.09942
}
d_results = most_likely_parameters(data=df_test,
ignore_columns=['error'])
self.assertTrue(
np.allclose(list(d_results.values()),
list(most_likely_correct.values()), 1e-7))
# Calculate the first 5 centered and raw moments from the c_t distribution.
central_mmnts = [
moment(events_active['c_t'].to_numpy(), moment=i) for i in range(1, 6)
]
raw_mmnts = raw_moment(central_moments=central_mmnts,
dist_mean=events_active['c_t'].mean())
# Fit the mixture of distributions.
m2n = M2N(raw_mmnts,
epsilon=epsilon,
factor=factor,
n_runs=fit_runs,
variant=variant,
max_iter=max_iter,
num_workers=num_workers)
df_fit_results = m2n.mp_fit()
fit_params = most_likely_parameters(df_fit_results)
params_list = [
fit_params[key]
for key in ['mu_1', 'mu_2', 'sigma_1', 'sigma_2', 'p_1']
]
# Calculate the bet size.
events_active['bet_size'] = events_active['c_t'].apply(
lambda c: single_bet_size_mixed(c, params_list))
if return_parameters:
return events_active, fit_params
return events_active
def confirm_and_cast_to_df(d_vars):
"""