Skip to content

Optimize natural_break on large inputs #562

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 3 commits into from
Oct 4, 2021
Merged

Optimize natural_break on large inputs #562

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
cfd6d5e
Optimize natural_break on large inputs
Sep 29, 2021
b5b094d
Split a long line to adhere to linter rules
alex-soklev Sep 29, 2021
93771ca
Fix a bug in natural_breaks
Sep 30, 2021
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
23 changes: 13 additions & 10 deletions xrspatial/classify.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ class cupy(object):
from datashader.colors import rgb
from xarray import DataArray

from numba import cuda
import numba as nb
import dask.array as da

from numpy.random import RandomState
Expand Down Expand Up @@ -112,7 +112,7 @@ def _run_dask_numpy_bin(data, bins, new_values):
return out


@cuda.jit(device=True)
@nb.cuda.jit(device=True)
def _gpu_bin(data, bins, new_values):
nbins = len(bins)
val = data[0, 0]
Expand All @@ -139,9 +139,9 @@ def _gpu_bin(data, bins, new_values):
return out


@cuda.jit
@nb.cuda.jit
def _run_gpu_bin(data, bins, new_values, out):
i, j = cuda.grid(2)
i, j = nb.cuda.grid(2)
if (i >= 0 and i < out.shape[0] and j >= 0 and j < out.shape[1]):
out[i, j] = _gpu_bin(data[i:i+1, j:j+1], bins, new_values)

Expand Down Expand Up @@ -440,7 +440,7 @@ def quantile(agg: xr.DataArray,
attrs=agg.attrs)


@ngjit
@nb.jit(nopython=True, parallel=True)
def _run_numpy_jenks_matrices(data, n_classes):
n_data = data.shape[0]
lower_class_limits = np.zeros((n_data + 1, n_classes + 1),
Expand All @@ -453,7 +453,7 @@ def _run_numpy_jenks_matrices(data, n_classes):
nl = data.shape[0] + 1
variance = 0.0

for l in range(2, nl): # noqa
for l in nb.prange(2, nl): # noqa
sum = 0.0
sum_squares = 0.0
w = 0.0
Expand Down Expand Up @@ -494,7 +494,7 @@ def _run_numpy_jenks_matrices(data, n_classes):
return lower_class_limits, var_combinations


@ngjit
@nb.jit(nopython=True)
def _run_numpy_jenks(data, n_classes):
# ported from existing cython implementation:
# https://github.com/perrygeo/jenks/blob/master/jenks.pyx
Expand Down Expand Up @@ -526,7 +526,9 @@ def _run_numpy_natural_break(data, num_sample, k):
# randomly select sample from the whole dataset
# create a pseudo random number generator
generator = RandomState(1234567890)
idx = [i for i in range(0, data.size)]
idx = np.linspace(
0, data.size, data.size, endpoint=False, dtype=np.uint32
)
generator.shuffle(idx)
sample_idx = idx[:num_sample]
sample_data = data.flatten()[sample_idx]
Expand All @@ -544,7 +546,8 @@ def _run_numpy_natural_break(data, num_sample, k):
Warning)

# only include non-nan values
sample_data = np.asarray([i for i in sample_data if np.isfinite(i)])
if not isinstance(sample_data, np.ndarray):
sample_data = np.asarray(sample_data)

uv = np.unique(sample_data[np.isfinite(sample_data)])
uvk = len(uv)
Expand All @@ -560,7 +563,7 @@ def _run_numpy_natural_break(data, num_sample, k):
uv.sort()
bins = uv
else:
centroids = _run_numpy_jenks(sample_data, k)
centroids = _run_numpy_jenks(uv, k)
bins = np.array(centroids[1:])

out = _bin(data, bins, np.arange(uvk))
Expand Down