File: //usr/lib/python3.9/site-packages/redis/commands/search/reducers.py
from .aggregation import Reducer, SortDirection
class FieldOnlyReducer(Reducer):
def __init__(self, field):
super().__init__(field)
self._field = field
class count(Reducer):
"""
Counts the number of results in the group
"""
NAME = "COUNT"
def __init__(self):
super().__init__()
class sum(FieldOnlyReducer):
"""
Calculates the sum of all the values in the given fields within the group
"""
NAME = "SUM"
def __init__(self, field):
super().__init__(field)
class min(FieldOnlyReducer):
"""
Calculates the smallest value in the given field within the group
"""
NAME = "MIN"
def __init__(self, field):
super().__init__(field)
class max(FieldOnlyReducer):
"""
Calculates the largest value in the given field within the group
"""
NAME = "MAX"
def __init__(self, field):
super().__init__(field)
class avg(FieldOnlyReducer):
"""
Calculates the mean value in the given field within the group
"""
NAME = "AVG"
def __init__(self, field):
super().__init__(field)
class tolist(FieldOnlyReducer):
"""
Returns all the matched properties in a list
"""
NAME = "TOLIST"
def __init__(self, field):
super().__init__(field)
class count_distinct(FieldOnlyReducer):
"""
Calculate the number of distinct values contained in all the results in
the group for the given field
"""
NAME = "COUNT_DISTINCT"
def __init__(self, field):
super().__init__(field)
class count_distinctish(FieldOnlyReducer):
"""
Calculate the number of distinct values contained in all the results in the
group for the given field. This uses a faster algorithm than
`count_distinct` but is less accurate
"""
NAME = "COUNT_DISTINCTISH"
class quantile(Reducer):
"""
Return the value for the nth percentile within the range of values for the
field within the group.
"""
NAME = "QUANTILE"
def __init__(self, field, pct):
super().__init__(field, str(pct))
self._field = field
class stddev(FieldOnlyReducer):
"""
Return the standard deviation for the values within the group
"""
NAME = "STDDEV"
def __init__(self, field):
super().__init__(field)
class first_value(Reducer):
"""
Selects the first value within the group according to sorting parameters
"""
NAME = "FIRST_VALUE"
def __init__(self, field, *byfields):
"""
Selects the first value of the given field within the group.
### Parameter
- **field**: Source field used for the value
- **byfields**: How to sort the results. This can be either the
*class* of `aggregation.Asc` or `aggregation.Desc` in which
case the field `field` is also used as the sort input.
`byfields` can also be one or more *instances* of `Asc` or `Desc`
indicating the sort order for these fields
"""
fieldstrs = []
if (
len(byfields) == 1
and isinstance(byfields[0], type)
and issubclass(byfields[0], SortDirection)
):
byfields = [byfields[0](field)]
for f in byfields:
fieldstrs += [f.field, f.DIRSTRING]
args = [field]
if fieldstrs:
args += ["BY"] + fieldstrs
super().__init__(*args)
self._field = field
class random_sample(Reducer):
"""
Returns a random sample of items from the dataset, from the given property
"""
NAME = "RANDOM_SAMPLE"
def __init__(self, field, size):
"""
### Parameter
**field**: Field to sample from
**size**: Return this many items (can be less)
"""
args = [field, str(size)]
super().__init__(*args)
self._field = field