def test_alive_probs(self, bbgb_params):
sim_data = beta_geometric_beta_binom_model(N=6, size=50000, **bbgb_params)
assert (np.abs(sim_data.loc[(sim_data['frequency'] == 0) & (sim_data['recency'] == 0),
'alive'].mean() - 0.11) < 0.01)
assert (np.abs(sim_data.loc[(sim_data['frequency'] == 2) & (sim_data['recency'] == 4),
'alive'].mean() - 0.59) < 0.01)
assert (np.abs(sim_data.loc[(sim_data['frequency'] == 6) & (sim_data['recency'] == 6),
'alive'].mean() - 0.93) < 0.01)
def test_params_same_from_sim_data(self, bbgb_params):
sim_data = beta_geometric_beta_binom_model(N=6, size=100000, **bbgb_params)
bbtf = estimation.BetaGeoBetaBinomFitter()
grouped_data = sim_data.groupby(['frequency', 'recency', 'n_periods'])['customer_id'].count()
grouped_data = grouped_data.reset_index().rename(columns={'customer_id': 'weights'})
bbtf.fit(grouped_data['frequency'],
grouped_data['recency'],
grouped_data['n_periods'],
grouped_data['weights'])
npt.assert_allclose(
np.asarray(list(bbgb_params.values())).astype(float),
np.asarray(bbtf._unload_params('alpha', 'beta', 'gamma', 'delta')).astype(float),
atol=0.1, rtol=1e-2)
def test_params_same_from_sim_data(self):
sim_data = beta_geometric_beta_binom_model(N=6,
size=100000,
**bbgb_params)
bbtf = estimation.BetaGeoBetaBinomFitter()
grouped_data = sim_data.groupby(['frequency', 'recency',
'n'])['customer_id'].count()
grouped_data = grouped_data.reset_index().rename(
columns={'customer_id': 'n_custs'})
bbtf.fit(grouped_data['frequency'], grouped_data['recency'],
grouped_data['n'], grouped_data['n_custs'])
assert (
(np.array(list(bbgb_params.values())) -
np.array(bbtf._unload_params('alpha', 'beta', 'gamma', 'delta')))
< 0.1).all()
def test_params_same_from_sim_data(self, bbgb_params):
sim_data = beta_geometric_beta_binom_model(N=6,
size=100000,
**bbgb_params)
bbtf = BetaGeoBetaBinomFitter()
grouped_data = sim_data.groupby(["frequency", "recency",
"n_periods"])["customer_id"].count()
grouped_data = grouped_data.reset_index().rename(
columns={"customer_id": "weights"})
bbtf.fit(grouped_data["frequency"], grouped_data["recency"],
grouped_data["n_periods"], grouped_data["weights"])
npt.assert_allclose(
np.asarray(list(bbgb_params.values())).astype(float),
np.asarray(bbtf._unload_params("alpha", "beta", "gamma",
"delta")).astype(float),
atol=0.1,
rtol=1e-2,
)
def test_hitting_max(self, bbgb_params):
sim_data = beta_geometric_beta_binom_model(N=6, size=5000, **bbgb_params)
assert sim_data['frequency'].max() == 6
assert sim_data['recency'].max() == 6
def test_positivity(self, bbgb_params):
sim_data = beta_geometric_beta_binom_model(N=6, size=5000, **bbgb_params)
assert (sim_data['frequency'] >= 0).all()
assert (sim_data['recency'] >= 0).all()
