Skip to content

ENH: Make group_var_ use Welford's algorithm. #10472

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 8, 2015
Merged

ENH: Make group_var_ use Welford's algorithm. #10472

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions doc/source/whatsnew/v0.17.0.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,8 +135,11 @@ Bug Fixes
- Bug in ``DatetimeIndex`` and ``PeriodIndex.value_counts`` resets name from its result, but retains in result's ``Index``. (:issue:`10150`)

- Bug in `pandas.concat` with ``axis=0`` when column is of dtype ``category`` (:issue:`10177`)

- Bug in ``read_msgpack`` where input type is not always checked (:issue:`10369`)

- Bug in `pandas.read_csv` with ``index_col=False`` or with ``index_col=['a', 'b']`` (:issue:`10413`, :issue:`10467`)

- Bug in `Series.from_csv` with ``header`` kwarg not setting the ``Series.name`` or the ``Series.index.name`` (:issue:`10483`)

- Bug in `groupby.var` which caused variance to be inaccurate for small float values (:issue:`10448`)
53 changes: 19 additions & 34 deletions pandas/src/generate_code.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1147,67 +1147,52 @@ def group_prod_bin_%(name)s(ndarray[%(dest_type2)s, ndim=2] out,

group_var_template = """@cython.wraparound(False)
@cython.boundscheck(False)
@cython.cdivision(True)
def group_var_%(name)s(ndarray[%(dest_type2)s, ndim=2] out,
ndarray[int64_t] counts,
ndarray[%(dest_type2)s, ndim=2] values,
ndarray[int64_t] labels):
cdef:
Py_ssize_t i, j, N, K, lab, ncounts = len(counts)
%(dest_type2)s val, ct
ndarray[%(dest_type2)s, ndim=2] nobs, sumx, sumxx
%(dest_type2)s val, ct, oldmean
ndarray[%(dest_type2)s, ndim=2] nobs, mean

if not len(values) == len(labels):
raise AssertionError("len(index) != len(labels)")

nobs = np.zeros_like(out)
sumx = np.zeros_like(out)
sumxx = np.zeros_like(out)
mean = np.zeros_like(out)

N, K = (<object> values).shape

with nogil:
if K > 1:
for i in range(N):

lab = labels[i]
if lab < 0:
continue
out[:, :] = 0.0

counts[lab] += 1

for j in range(K):
val = values[i, j]

# not nan
if val == val:
nobs[lab, j] += 1
sumx[lab, j] += val
sumxx[lab, j] += val * val
else:
for i in range(N):
with nogil:
for i in range(N):
lab = labels[i]
if lab < 0:
continue

lab = labels[i]
if lab < 0:
continue
counts[lab] += 1

counts[lab] += 1
val = values[i, 0]
for j in range(K):
val = values[i, j]

# not nan
if val == val:
nobs[lab, 0] += 1
sumx[lab, 0] += val
sumxx[lab, 0] += val * val

nobs[lab, j] += 1
oldmean = mean[lab, j]
mean[lab, j] += (val - oldmean) / nobs[lab, j]
out[lab, j] += (val - mean[lab, j]) * (val - oldmean)

for i in range(ncounts):
for j in range(K):
ct = nobs[i, j]
if ct < 2:
out[i, j] = NAN
else:
out[i, j] = ((ct * sumxx[i, j] - sumx[i, j] * sumx[i, j]) /
(ct * ct - ct))
out[i, j] /= (ct - 1)

"""

group_var_bin_template = """@cython.wraparound(False)
Expand Down
106 changes: 38 additions & 68 deletions pandas/src/generated.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7232,131 +7232,101 @@ def group_prod_bin_float32(ndarray[float32_t, ndim=2] out,

@cython.wraparound(False)
@cython.boundscheck(False)
@cython.cdivision(True)
def group_var_float64(ndarray[float64_t, ndim=2] out,
ndarray[int64_t] counts,
ndarray[float64_t, ndim=2] values,
ndarray[int64_t] labels):
cdef:
Py_ssize_t i, j, N, K, lab, ncounts = len(counts)
float64_t val, ct
ndarray[float64_t, ndim=2] nobs, sumx, sumxx
float64_t val, ct, oldmean
ndarray[float64_t, ndim=2] nobs, mean

if not len(values) == len(labels):
raise AssertionError("len(index) != len(labels)")

nobs = np.zeros_like(out)
sumx = np.zeros_like(out)
sumxx = np.zeros_like(out)
mean = np.zeros_like(out)

N, K = (<object> values).shape

with nogil:
if K > 1:
for i in range(N):

lab = labels[i]
if lab < 0:
continue
out[:, :] = 0.0

counts[lab] += 1

for j in range(K):
val = values[i, j]

# not nan
if val == val:
nobs[lab, j] += 1
sumx[lab, j] += val
sumxx[lab, j] += val * val
else:
for i in range(N):
with nogil:
for i in range(N):
lab = labels[i]
if lab < 0:
continue

lab = labels[i]
if lab < 0:
continue
counts[lab] += 1

counts[lab] += 1
val = values[i, 0]
for j in range(K):
val = values[i, j]

# not nan
if val == val:
nobs[lab, 0] += 1
sumx[lab, 0] += val
sumxx[lab, 0] += val * val

nobs[lab, j] += 1
oldmean = mean[lab, j]
mean[lab, j] += (val - oldmean) / nobs[lab, j]
out[lab, j] += (val - mean[lab, j]) * (val - oldmean)

for i in range(ncounts):
for j in range(K):
ct = nobs[i, j]
if ct < 2:
out[i, j] = NAN
else:
out[i, j] = ((ct * sumxx[i, j] - sumx[i, j] * sumx[i, j]) /
(ct * ct - ct))
out[i, j] /= (ct - 1)


@cython.wraparound(False)
@cython.boundscheck(False)
@cython.cdivision(True)
def group_var_float32(ndarray[float32_t, ndim=2] out,
ndarray[int64_t] counts,
ndarray[float32_t, ndim=2] values,
ndarray[int64_t] labels):
cdef:
Py_ssize_t i, j, N, K, lab, ncounts = len(counts)
float32_t val, ct
ndarray[float32_t, ndim=2] nobs, sumx, sumxx
float32_t val, ct, oldmean
ndarray[float32_t, ndim=2] nobs, mean

if not len(values) == len(labels):
raise AssertionError("len(index) != len(labels)")

nobs = np.zeros_like(out)
sumx = np.zeros_like(out)
sumxx = np.zeros_like(out)
mean = np.zeros_like(out)

N, K = (<object> values).shape

with nogil:
if K > 1:
for i in range(N):

lab = labels[i]
if lab < 0:
continue
out[:, :] = 0.0

counts[lab] += 1

for j in range(K):
val = values[i, j]

# not nan
if val == val:
nobs[lab, j] += 1
sumx[lab, j] += val
sumxx[lab, j] += val * val
else:
for i in range(N):
with nogil:
for i in range(N):
lab = labels[i]
if lab < 0:
continue

lab = labels[i]
if lab < 0:
continue
counts[lab] += 1

counts[lab] += 1
val = values[i, 0]
for j in range(K):
val = values[i, j]

# not nan
if val == val:
nobs[lab, 0] += 1
sumx[lab, 0] += val
sumxx[lab, 0] += val * val

nobs[lab, j] += 1
oldmean = mean[lab, j]
mean[lab, j] += (val - oldmean) / nobs[lab, j]
out[lab, j] += (val - mean[lab, j]) * (val - oldmean)

for i in range(ncounts):
for j in range(K):
ct = nobs[i, j]
if ct < 2:
out[i, j] = NAN
else:
out[i, j] = ((ct * sumxx[i, j] - sumx[i, j] * sumx[i, j]) /
(ct * ct - ct))
out[i, j] /= (ct - 1)



@cython.wraparound(False)
Expand Down
Loading