import numpy as np
from .baseMetric import BaseMetric
from collections import Counter
__all__ = ['StringCountMetric']
class keylookerupper(object):
"""Helper object to unpack dictionary values as reduceFunction results.
"""
def __init__(self, key='blank', name=None):
self.key = key
self.__name__ = name
def __call__(self, indict):
return np.max(indict[self.key])
[docs]class StringCountMetric(BaseMetric):
"""Count up the number of times each string appears in a column.
Dynamically builds reduce functions for each unique string value, so summary sats can be
named the same as strings in the simData array without knowing the values of those trings ahead of time.
"""
def __init__(self, metricName='stringCountMetric',
col='filter', percent=False, **kwargs):
"""
Parameters
----------
col: str ('filter')
Column name that has strings to look at
percent : bool (False)
Normalize and return results as percents ranther than raw count
"""
if percent:
units = 'percent'
else:
units = 'count'
self.percent = percent
cols = [col]
super(StringCountMetric, self).__init__(cols, metricName, units=units,
metricDtype=object, **kwargs)
self.col = col
[docs] def run(self, dataslice, slicePoint=None):
counter = Counter(dataslice[self.col])
# convert to a numpy array
lables = list(counter.keys())
# Numpy can't handle empty string as a dtype
lables = [x if x != '' else 'blank' for x in lables]
metricValue = np.zeros(1, dtype=list(zip(lables, [float]*len(counter.keys()))))
for key in counter:
if key == '':
metricValue['blank'] = counter[key]
else:
metricValue[key] = counter[key]
if self.percent:
norm = sum(metricValue[0])/100.
# Not sure I really like having to loop here, but the dtype is inflexible
for key in metricValue.dtype.names:
metricValue[key] = metricValue[key]/norm
# Now to dynamically set up the reduce functions
for i, key in enumerate(metricValue.dtype.names):
name = key
self.reduceFuncs[name] = keylookerupper(key=key, name=name)
self.reduceOrder[name] = i
return metricValue