def __init__(self, on: Optional[KeyFn] = None, ignore_nulls: bool = True):
self._set_key_fn(on)
on_fn = _to_on_fn(on)
super().__init__(
init=_null_wrap_init(lambda k: 0),
accumulate=_null_wrap_accumulate(ignore_nulls, on_fn, lambda a, r: a + r),
merge=_null_wrap_merge(ignore_nulls, lambda a1, a2: a1 + a2),
finalize=_null_wrap_finalize(lambda a: a),
name=(f"sum({str(on)})"),
)
def __init__(self, on: Optional[KeyFn] = None, ignore_nulls: bool = True):
self._set_key_fn(on)
on_fn = _to_on_fn(on)
super().__init__(
init=_null_wrap_init(lambda k: float("-inf")),
accumulate=_null_wrap_accumulate(ignore_nulls, on_fn, max),
merge=_null_wrap_merge(ignore_nulls, max),
finalize=_null_wrap_finalize(lambda a: a),
name=(f"max({str(on)})"),
)
def __init__(
self,
on: Optional[KeyFn] = None,
ddof: int = 1,
ignore_nulls: bool = True,
):
self._set_key_fn(on)
on_fn = _to_on_fn(on)
def accumulate(a: List[float], r: float):
# Accumulates the current count, the current mean, and the sum of
# squared differences from the current mean (M2).
M2, mean, count = a
count += 1
delta = r - mean
mean += delta / count
delta2 = r - mean
M2 += delta * delta2
return [M2, mean, count]
def merge(a: List[float], b: List[float]):
# Merges two accumulations into one.
# See
# https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Parallel_algorithm
M2_a, mean_a, count_a = a
M2_b, mean_b, count_b = b
delta = mean_b - mean_a
count = count_a + count_b
# NOTE: We use this mean calculation since it's more numerically
# stable than mean_a + delta * count_b / count, which actually
# deviates from Pandas in the ~15th decimal place and causes our
# exact comparison tests to fail.
mean = (mean_a * count_a + mean_b * count_b) / count
# Update the sum of squared differences.
M2 = M2_a + M2_b + (delta ** 2) * count_a * count_b / count
return [M2, mean, count]
def finalize(a: List[float]):
# Compute the final standard deviation from the accumulated
# sum of squared differences from current mean and the count.
M2, mean, count = a
if count < 2:
return 0.0
return math.sqrt(M2 / (count - ddof))
super().__init__(
init=_null_wrap_init(lambda k: [0, 0, 0]),
accumulate=_null_wrap_accumulate(ignore_nulls, on_fn, accumulate),
merge=_null_wrap_merge(ignore_nulls, merge),
finalize=_null_wrap_finalize(finalize),
name=(f"std({str(on)})"),
)
def __init__(self, on: Optional[KeyFn] = None, ignore_nulls: bool = True):
self._set_key_fn(on)
on_fn = _to_on_fn(on)
super().__init__(
init=_null_wrap_init(lambda k: [0, 0]),
accumulate=_null_wrap_accumulate(
ignore_nulls, on_fn, lambda a, r: [a[0] + r, a[1] + 1]),
merge=_null_wrap_merge(
ignore_nulls, lambda a1, a2: [a1[0] + a2[0], a1[1] + a2[1]]),
finalize=_null_wrap_finalize(lambda a: a[0] / a[1]),
name=(f"mean({str(on)})"),
)