def get_z_sum(interval_length,
wet_or_dry,
zindex_values,
periods_per_year,
calibration_start_year,
calibration_end_year,
input_start_year):
tempZ = Stack()
values_to_sum = Stack()
summed_values = Stack()
for zindex in zindex_values:
# we need to skip Z-index values from the list if they don't exist, this can result from empty months in the final year of the data set
if not np.isnan(zindex):
tempZ.push(zindex)
# remove periods before the start of the calibration interval
initial_calibration_period_index = (calibration_start_year - input_start_year) * periods_per_year
for i in range(initial_calibration_period_index):
if tempZ.size > 0:
tempZ.pop()
else:
break
calibration_periods_left = (calibration_end_year - calibration_start_year + 1) * periods_per_year;
# get the first interval length of values from the end of the calibration period working backwards, creating the first sum_value of interval periods
sum_value = 0
for i in range(interval_length):
if tempZ.isEmpty():
i = interval_length
else:
# pull a value off the end of the list
z = tempZ.pop()
# reduce the remaining number of calibration interval periods we have left to process
calibration_periods_left -= 1
'''
/* assumes that nCalibrationPeriods is >= length, reasonable
** This is a reasonable assumption and does not hurt if
** anything if false--just calibrate over a slightly longer
** interval, which is already way too short in that case */
'''
if (not np.isnan(z) and (MISSING_VALUE != z)):
# add to the sum_value
sum_value += z
# add to the array of values we've used for the initial sum_value
values_to_sum.add_first(z)
else:
# reduce the loop counter so we don't skip a calibration interval period
i -= 1
# if we're dealing with wet conditions then we want to be using positive numbers, and if dry conditions
# then we need to be using negative numbers, so we introduce a sign variable to help with this
if wet_or_dry == 'wet':
sign = 1
else: # dry
sign =-1
# now for each remaining Z value, recalculate the sum_value based on last value in the list to sum_value and the next Z value
max_sum = sum_value
summed_values.add_first(sum_value)
while (tempZ.size > 0) and (calibration_periods_left > 0):
# take the next Z-index value off the end of the list
z = tempZ.pop()
# reduce by one period for each removal
calibration_periods_left -= 1
if not np.isnan(z) and (MISSING_VALUE != z) and not values_to_sum.isEmpty():
# come up with a new sum_value for this new group of values to sum_value
# remove the last value from both the sum_value and the values to sum_value array
sum_value -= values_to_sum.peek()
values_to_sum.pop()
# add to the sum_value
sum_value += z
# update the values to sum_value and summed values lists
values_to_sum.add_first(z);
summed_values.add_first(sum_value);
# update the maximum sum_value value if we have a new max
if (sign * sum_value) > (sign * max_sum):
max_sum = sum_value
# highest reasonable is the highest (or lowest) value that is not due to some freak anomaly in the data.
# "freak anomaly" is defined as a value that is either
# 1) 25% higher than the 98th percentile
# 2) 25% lower than the 2nd percentile
if wet_or_dry == 'wet':
safe_percentile_index = int(summed_values.size * 0.98)
else: # DRY
safe_percentile_index = int(summed_values.size * 0.02)
# sort the list of sums into ascending order and get the sum_value value referenced by the safe percentile index
summed_values_list = summed_values.sorted_list()
sum_at_safe_percentile = summed_values_list[safe_percentile_index]
# find the highest reasonable value out of the summed values
highest_reasonable_value = 0.0
reasonable_tolerance_ratio = 1.25
while (summed_values.size > 0):
sum_value = summed_values.pop()
if (sign * sum_value > 0):
if ((sum_value / sum_at_safe_percentile) < reasonable_tolerance_ratio):
if (sign * sum_value > sign * highest_reasonable_value):
highest_reasonable_value = sum_value
if wet_or_dry == 'wet':
return highest_reasonable_value
else: # DRY
return max_sum
