def get_histogram(data_instances):
static_obj = statics.MultivariateStatisticalSummary(data_instances, cols_index=-1)
label_historgram = static_obj.get_label_histogram()
event_total = label_historgram.get(1, 0)
non_event_total = label_historgram.get(0, 0)
if event_total == 0 or non_event_total == 0:
LOGGER.warning(f"event_total or non_event_total might have errors, event_total: {event_total},"
f" non_event_total: {non_event_total}")
return event_total, non_event_total
def zeros(self, data_shape, fit_intercept, data_instances):
"""
If fit intercept, use the following formula to initialize b can get a faster converge rate
b = log(P(1)/P(0))
"""
inits = np.zeros(data_shape)
if fit_intercept and data_instances is not None:
static_obj = statics.MultivariateStatisticalSummary(data_instances, cols_index=-1)
label_historgram = static_obj.get_label_histogram()
LOGGER.debug("label_histogram is : {}".format(label_historgram))
one_count = label_historgram.get(1)
zero_count = label_historgram.get(0, 0) + label_historgram.get(-1, 0)
init_intercept = np.log((one_count / zero_count))
inits[-1] = init_intercept
return inits