def test_fit_with_and_without_weights(self, donations):
exploded_dataset = pd.DataFrame(
columns=["frequency", "recency", "periods"])
for _, row in donations.iterrows():
exploded_dataset = exploded_dataset.append(
pd.DataFrame(
[[row["frequency"], row["recency"], row["periods"]]] *
row["weights"],
columns=["frequency", "recency", "periods"],
))
exploded_dataset = exploded_dataset.astype(np.int64)
exploded_dataset.to_csv("exploded.csv")
assert exploded_dataset.shape[0] == donations["weights"].sum()
bbtf_noweights = lt.BetaGeoBetaBinomFitter()
bbtf_noweights.fit(exploded_dataset["frequency"],
exploded_dataset["recency"],
exploded_dataset["periods"])
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"], donations["recency"],
donations["periods"], donations["weights"])
npt.assert_array_almost_equal(
np.array(
bbtf_noweights._unload_params("alpha", "beta", "gamma",
"delta")),
np.array(bbtf._unload_params("alpha", "beta", "gamma", "delta")),
decimal=4,
)
def test_model_has_standard_error_variance_matrix_and_confidence_intervals_(
self, donations):
donations = donations
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"], donations["recency"],
donations["periods"], donations["weights"])
assert hasattr(bbtf, "standard_errors_")
assert hasattr(bbtf, "variance_matrix_")
assert hasattr(bbtf, "confidence_intervals_")
def test_fit_with_index(self, donations):
bbtf = lt.BetaGeoBetaBinomFitter()
index = range(len(donations), 0, -1)
bbtf.fit(donations["frequency"],
donations["recency"],
donations["periods"],
donations["weights"],
index=index)
assert (bbtf.data.index == index).all() == True
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"],
donations["recency"],
donations["periods"],
donations["weights"],
index=None)
assert (bbtf.data.index == index).all() == False
def test_params_out_is_close_to_Hardie_paper(self, donations):
donations = donations
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"], donations["recency"],
donations["periods"], donations["weights"])
expected = np.array([1.204, 0.750, 0.657, 2.783])
npt.assert_array_almost_equal(
expected,
np.array(bbtf._unload_params("alpha", "beta", "gamma", "delta")),
decimal=2)
def test_expected_purchases_in_n_periods_returns_same_value_as_Hardie_excel_sheet(
self, donations):
"""Total expected from Hardie's In-Sample Fit sheet."""
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"], donations["recency"],
donations["periods"], donations["weights"])
expected = np.array([3454.9, 1253.1]) # Cells C18 and C24
estimated = bbtf.expected_number_of_transactions_in_first_n_periods(
6).loc[[0, 6]].values.flatten()
npt.assert_almost_equal(expected, estimated, decimal=0)
def test_conditional_expectation_returns_same_value_as_Hardie_excel_sheet(
self, donations):
"""
Total from Hardie's Conditional Expectations (II) sheet.
http://brucehardie.com/notes/010/BGBB_2011-01-20_XLSX.zip
"""
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"], donations["recency"],
donations["periods"], donations["weights"])
pred_purchases = (
bbtf.conditional_expected_number_of_purchases_up_to_time(
5, donations["frequency"], donations["recency"],
donations["periods"]) * donations["weights"])
expected = 12884.2 # Sum of column F Exp Tot
npt.assert_almost_equal(expected, pred_purchases.sum(), decimal=0)
def test_prob_alive_is_close_to_Hardie_paper_table_6(self, donations):
"""Table 6: P(Alive in 2002) as a Function of Recency and Frequency"""
bbtf = lt.BetaGeoBetaBinomFitter()
bbtf.fit(donations["frequency"], donations["recency"],
donations["periods"], donations["weights"])
bbtf.data["prob_alive"] = bbtf.conditional_probability_alive(
1, donations["frequency"], donations["recency"],
donations["periods"])
# Expected probabilities for last year 1995-0 repeat, 1999-2 repeat, 2001-6 repeat
expected = np.array([0.11, 0.59, 0.93])
prob_list = np.zeros(3)
prob_list[0] = bbtf.data[(bbtf.data["frequency"] == 0)
& (bbtf.data["recency"] == 0)]["prob_alive"]
prob_list[1] = bbtf.data[(bbtf.data["frequency"] == 2)
& (bbtf.data["recency"] == 4)]["prob_alive"]
prob_list[2] = bbtf.data[(bbtf.data["frequency"] == 6)
& (bbtf.data["recency"] == 6)]["prob_alive"]
npt.assert_array_almost_equal(expected, prob_list, decimal=2)