def _incremental_mean_and_var(X, epsilon, range, last_mean, last_variance,
last_sample_count):
# old = stats until now
# new = the current increment
# updated = the aggregated stats
last_sum = last_mean * last_sample_count
new_mean = nanmean(X, epsilon=epsilon, axis=0, range=range)
new_sample_count = np.sum(~np.isnan(X), axis=0)
new_sum = new_mean * new_sample_count
updated_sample_count = last_sample_count + new_sample_count
updated_mean = (last_sum + new_sum) / updated_sample_count
if last_variance is None:
updated_variance = None
else:
new_unnormalized_variance = nanvar(
X, epsilon=epsilon, axis=0, range=range) * new_sample_count
last_unnormalized_variance = last_variance * last_sample_count
with np.errstate(divide='ignore', invalid='ignore'):
last_over_new_count = last_sample_count / new_sample_count
updated_unnormalized_variance = (
last_unnormalized_variance + new_unnormalized_variance +
last_over_new_count / updated_sample_count *
(last_sum / last_over_new_count - new_sum)**2)
zeros = last_sample_count == 0
updated_unnormalized_variance[zeros] = new_unnormalized_variance[zeros]
updated_variance = updated_unnormalized_variance / updated_sample_count
return updated_mean, updated_variance, updated_sample_count
def _incremental_mean_and_var(X, epsilon, bounds, last_mean, last_variance,
last_sample_count):
# Initialising new accountant, as budget is tracked in main class. Subject to review in line with GH issue #21
temp_acc = BudgetAccountant()
# old = stats until now
# new = the current increment
# updated = the aggregated stats
last_sum = last_mean * last_sample_count
new_mean = nanmean(X,
epsilon=epsilon,
axis=0,
bounds=bounds,
accountant=temp_acc)
new_sample_count = np.sum(~np.isnan(X), axis=0)
new_sum = new_mean * new_sample_count
updated_sample_count = last_sample_count + new_sample_count
updated_mean = (last_sum + new_sum) / updated_sample_count
if last_variance is None:
updated_variance = None
else:
new_unnormalized_variance = nanvar(
X, epsilon=epsilon, axis=0, bounds=bounds,
accountant=temp_acc) * new_sample_count
last_unnormalized_variance = last_variance * last_sample_count
with np.errstate(divide='ignore', invalid='ignore'):
last_over_new_count = last_sample_count / new_sample_count
updated_unnormalized_variance = (
last_unnormalized_variance + new_unnormalized_variance +
last_over_new_count / updated_sample_count *
(last_sum / last_over_new_count - new_sum)**2)
zeros = last_sample_count == 0
updated_unnormalized_variance[zeros] = new_unnormalized_variance[zeros]
updated_variance = updated_unnormalized_variance / updated_sample_count
return updated_mean, updated_variance, updated_sample_count