def _colorize(agg, color_key, how, min_alpha, name):
if cupy and isinstance(agg.data, cupy.ndarray):
from ._cuda_utils import interp
array = cupy.array
else:
interp = np.interp
array = np.array
if not agg.ndim == 3:
raise ValueError("agg must be 3D")
cats = agg.indexes[agg.dims[-1]]
if color_key is None:
raise ValueError("Color key must be provided, with at least as many " +
"colors as there are categorical fields")
if not isinstance(color_key, dict):
color_key = dict(zip(cats, color_key))
if len(color_key) < len(cats):
raise ValueError(
"Insufficient colors provided ({}) for the categorical fields available ({})"
.format(len(color_key), len(cats)))
if not (0 <= min_alpha <= 255):
raise ValueError(
"min_alpha ({}) must be between 0 and 255".format(min_alpha))
colors = [rgb(color_key[c]) for c in cats]
rs, gs, bs = map(array, zip(*colors))
# Reorient array (transposing the category dimension first)
agg_t = agg.transpose(*((agg.dims[-1], ) + agg.dims[:2]))
data = orient_array(agg_t).transpose([1, 2, 0])
total = data.sum(axis=2)
# zero-count pixels will be 0/0, but it's safe to ignore that when dividing
with np.errstate(divide='ignore', invalid='ignore'):
r = (data.dot(rs) / total).astype(np.uint8)
g = (data.dot(gs) / total).astype(np.uint8)
b = (data.dot(bs) / total).astype(np.uint8)
offset = total.min()
mask = np.isnan(total)
if offset == 0:
mask = mask | (total <= 0)
offset = total[total > 0].min()
a = _normalize_interpolate_how(how)(total - offset, mask)
a = interp(a,
array([np.nanmin(a).item(),
np.nanmax(a).item()]),
array([min_alpha, 255]),
left=0,
right=255).astype(np.uint8)
r[mask] = g[mask] = b[mask] = 255
values = np.dstack([r, g, b, a]).view(np.uint32).reshape(a.shape)
if cupy and isinstance(values, cupy.ndarray):
# Convert cupy array to numpy for final image
values = cupy.asnumpy(values)
return Image(values,
dims=agg.dims[:-1],
coords=OrderedDict([
(agg.dims[1], agg.coords[agg.dims[1]]),
(agg.dims[0], agg.coords[agg.dims[0]]),
]),
name=name)
def _colorize(agg, color_key, how, alpha, span, min_alpha, name,
color_baseline):
if cupy and isinstance(agg.data, cupy.ndarray):
from ._cuda_utils import interp, masked_clip_2d
array = cupy.array
else:
from ._cpu_utils import masked_clip_2d
interp = np.interp
array = np.array
if not agg.ndim == 3:
raise ValueError("agg must be 3D")
cats = agg.indexes[agg.dims[-1]]
if not len(cats
): # No categories and therefore no data; return an empty image
return Image(np.zeros(agg.shape[0:2], dtype=np.uint32),
dims=agg.dims[:-1],
coords=OrderedDict([
(agg.dims[1], agg.coords[agg.dims[1]]),
(agg.dims[0], agg.coords[agg.dims[0]])
]),
name=name)
if color_key is None:
raise ValueError("Color key must be provided, with at least as many " +
"colors as there are categorical fields")
if not isinstance(color_key, dict):
color_key = dict(zip(cats, color_key))
if len(color_key) < len(cats):
raise ValueError(
"Insufficient colors provided ({}) for the categorical fields available ({})"
.format(len(color_key), len(cats)))
colors = [rgb(color_key[c]) for c in cats]
rs, gs, bs = map(array, zip(*colors))
# Reorient array (transposing the category dimension first)
agg_t = agg.transpose(*((agg.dims[-1], ) + agg.dims[:2]))
data = orient_array(agg_t).transpose([1, 2, 0])
if isinstance(data, da.Array):
data = data.compute()
color_data = data.copy()
# subtract color_baseline if needed
baseline = np.nanmin(
color_data) if color_baseline is None else color_baseline
with np.errstate(invalid='ignore'):
if baseline > 0:
color_data -= baseline
elif baseline < 0:
color_data += -baseline
if color_data.dtype.kind != 'u' and color_baseline is not None:
color_data[color_data < 0] = 0
color_total = nansum_missing(color_data, axis=2)
# dot does not handle nans, so replace with zeros
color_data[np.isnan(data)] = 0
# zero-count pixels will be 0/0, but it's safe to ignore that when dividing
with np.errstate(divide='ignore', invalid='ignore'):
r = (color_data.dot(rs) / color_total).astype(np.uint8)
g = (color_data.dot(gs) / color_total).astype(np.uint8)
b = (color_data.dot(bs) / color_total).astype(np.uint8)
# special case -- to give an appropriate color when min_alpha != 0 and data=0,
# take avg color of all non-nan categories
color_mask = ~np.isnan(data)
cmask_sum = np.sum(color_mask, axis=2)
with np.errstate(divide='ignore', invalid='ignore'):
r2 = (color_mask.dot(rs) / cmask_sum).astype(np.uint8)
g2 = (color_mask.dot(gs) / cmask_sum).astype(np.uint8)
b2 = (color_mask.dot(bs) / cmask_sum).astype(np.uint8)
missing_colors = np.sum(color_data, axis=2) == 0
r = np.where(missing_colors, r2, r)
g = np.where(missing_colors, g2, g)
b = np.where(missing_colors, b2, b)
total = nansum_missing(data, axis=2)
mask = np.isnan(total)
# if span is provided, use it, otherwise produce a span based off the
# min/max of the data
if span is None:
offset = np.nanmin(total)
if total.dtype.kind == 'u' and offset == 0:
mask = mask | (total == 0)
# If at least one element is not masked, use the minimum as the offset
# otherwise the offset remains at zero
if not np.all(mask):
offset = total[total > 0].min()
total = np.where(~mask, total, np.nan)
a_scaled = _normalize_interpolate_how(how)(total - offset, mask)
norm_span = [np.nanmin(a_scaled).item(), np.nanmax(a_scaled).item()]
else:
if how == 'eq_hist':
# For eq_hist to work with span, we'll need to compute the histogram
# only on the specified span's range.
raise ValueError("span is not (yet) valid to use with eq_hist")
# even in fixed-span mode cells with 0 should remain fully transparent
# i.e. a 0 will be fully transparent, but any non-zero number will
# be clipped to the span range and have min-alpha applied
offset = np.array(span, dtype=data.dtype)[0]
if total.dtype.kind == 'u' and np.nanmin(total) == 0:
mask = mask | (total <= 0)
total = np.where(~mask, total, np.nan)
masked_clip_2d(total, mask, *span)
a_scaled = _normalize_interpolate_how(how)(total - offset, mask)
norm_span = _normalize_interpolate_how(how)([0, span[1] - span[0]], 0)
# Interpolate the alpha values
a = interp(a_scaled,
array(norm_span),
array([min_alpha, alpha]),
left=0,
right=255).astype(np.uint8)
values = np.dstack([r, g, b, a]).view(np.uint32).reshape(a.shape)
if cupy and isinstance(values, cupy.ndarray):
# Convert cupy array to numpy for final image
values = cupy.asnumpy(values)
return Image(values,
dims=agg.dims[:-1],
coords=OrderedDict([
(agg.dims[1], agg.coords[agg.dims[1]]),
(agg.dims[0], agg.coords[agg.dims[0]]),
]),
name=name)
def _interpolate(agg, cmap, how, alpha, span, min_alpha, name):
if cupy and isinstance(agg.data, cupy.ndarray):
from ._cuda_utils import masked_clip_2d, interp
else:
from ._cpu_utils import masked_clip_2d
interp = np.interp
if agg.ndim != 2:
raise ValueError("agg must be 2D")
interpolater = _normalize_interpolate_how(how)
data = orient_array(agg).copy()
# Compute mask
if np.issubdtype(data.dtype, np.bool_):
mask = ~data
data = data.astype(np.int8)
else:
if np.issubdtype(data.dtype, np.integer):
mask = data == 0
else:
mask = np.isnan(data)
# Handle case where everything is masked out
if mask.all():
return Image(np.zeros(shape=agg.data.shape, dtype=np.uint32),
coords=agg.coords,
dims=agg.dims,
attrs=agg.attrs,
name=name)
# Handle offset / clip
if span is None:
offset = np.nanmin(data[~mask])
else:
offset = np.array(span, dtype=data.dtype)[0]
masked_clip_2d(data, mask, *span)
# If log/cbrt, could case to float64 right away
# If linear, can keep current type
data -= offset
with np.errstate(invalid="ignore", divide="ignore"):
# Transform data (log, eq_hist, etc.)
data = interpolater(data, mask)
# Transform span
if span is None:
masked_data = np.where(~mask, data, np.nan)
span = np.nanmin(masked_data), np.nanmax(masked_data)
else:
if how == 'eq_hist':
# For eq_hist to work with span, we'll need to store the histogram
# from the data and then apply it to the span argument.
raise ValueError("span is not (yet) valid to use with eq_hist")
span = interpolater([0, span[1] - span[0]], 0)
if isinstance(cmap, Iterator):
cmap = list(cmap)
if isinstance(cmap, list):
rspan, gspan, bspan = np.array(list(zip(*map(rgb, cmap))))
span = np.linspace(span[0], span[1], len(cmap))
r = interp(data, span, rspan, left=255).astype(np.uint8)
g = interp(data, span, gspan, left=255).astype(np.uint8)
b = interp(data, span, bspan, left=255).astype(np.uint8)
a = np.where(np.isnan(data), 0, alpha).astype(np.uint8)
rgba = np.dstack([r, g, b, a])
elif isinstance(cmap, str) or isinstance(cmap, tuple):
color = rgb(cmap)
aspan = np.arange(min_alpha, alpha + 1)
span = np.linspace(span[0], span[1], len(aspan))
r = np.full(data.shape, color[0], dtype=np.uint8)
g = np.full(data.shape, color[1], dtype=np.uint8)
b = np.full(data.shape, color[2], dtype=np.uint8)
a = interp(data, span, aspan, left=0, right=255).astype(np.uint8)
rgba = np.dstack([r, g, b, a])
elif callable(cmap):
# Assume callable is matplotlib colormap
scaled_data = (data - span[0]) / (span[1] - span[0])
if cupy and isinstance(scaled_data, cupy.ndarray):
# Convert cupy array to numpy before passing to matplotlib colormap
scaled_data = cupy.asnumpy(scaled_data)
rgba = cmap(scaled_data, bytes=True)
rgba[:, :, 3] = np.where(np.isnan(scaled_data), 0,
alpha).astype(np.uint8)
else:
raise TypeError("Expected `cmap` of `matplotlib.colors.Colormap`, "
"`list`, `str`, or `tuple`; got: '{0}'".format(
type(cmap)))
img = rgba.view(np.uint32).reshape(data.shape)
if cupy and isinstance(img, cupy.ndarray):
# Convert cupy array to numpy for final image
img = cupy.asnumpy(img)
return Image(img, coords=agg.coords, dims=agg.dims, name=name)
def _colorize(agg, color_key, how, span, min_alpha, name):
if cupy and isinstance(agg.data, cupy.ndarray):
from ._cuda_utils import interp, masked_clip_2d
array = cupy.array
else:
from ._cpu_utils import masked_clip_2d
interp = np.interp
array = np.array
if not agg.ndim == 3:
raise ValueError("agg must be 3D")
cats = agg.indexes[agg.dims[-1]]
if color_key is None:
raise ValueError("Color key must be provided, with at least as many " +
"colors as there are categorical fields")
if not isinstance(color_key, dict):
color_key = dict(zip(cats, color_key))
if len(color_key) < len(cats):
raise ValueError("Insufficient colors provided ({}) for the categorical fields available ({})"
.format(len(color_key), len(cats)))
if not (0 <= min_alpha <= 255):
raise ValueError("min_alpha ({}) must be between 0 and 255".format(min_alpha))
colors = [rgb(color_key[c]) for c in cats]
rs, gs, bs = map(array, zip(*colors))
# Reorient array (transposing the category dimension first)
agg_t = agg.transpose(*((agg.dims[-1],)+agg.dims[:2]))
data = orient_array(agg_t).transpose([1, 2, 0])
total = data.sum(axis=2)
# zero-count pixels will be 0/0, but it's safe to ignore that when dividing
with np.errstate(divide='ignore', invalid='ignore'):
r = (data.dot(rs)/total).astype(np.uint8)
g = (data.dot(gs)/total).astype(np.uint8)
b = (data.dot(bs)/total).astype(np.uint8)
mask = np.isnan(total)
# if span is provided, use it, otherwise produce it a span based off the
# min/max of the data
if span is None:
# Currently masks out zero or negative values, but will need fixing
offset = np.nanmin(total)
if offset == 0 and total.dtype.kind == 'u':
mask = mask | (total <= 0)
# If at least one element is not masked, use the minimum as the offset
# otherwise the offset remains at zero
if not np.all(mask):
offset = total[total > 0].min()
total = np.where(~mask, total, np.nan)
a_scaled = _normalize_interpolate_how(how)(total - offset, mask)
norm_span = [np.nanmin(a_scaled).item(), np.nanmax(a_scaled).item()]
else:
if how == 'eq_hist':
# For eq_hist to work with span, we'll need to store the histogram
# from the data and then apply it to the span argument.
raise ValueError("span is not (yet) valid to use with eq_hist")
# even in fixed-span mode cells with 0 should remain fully transparent
# i.e. a 0 will be fully transparent, but any non-zero number will
# be clipped to the span range and have min-alpha applied
offset = np.array(span, dtype=data.dtype)[0]
if np.nanmin(total) == 0 and total.dtype.kind == 'u':
mask = mask | (total <= 0)
total = np.where(~mask, total, np.nan)
masked_clip_2d(total, mask, *span)
a_scaled = _normalize_interpolate_how(how)(total - offset, mask)
norm_span = _normalize_interpolate_how(how)([0, span[1] - span[0]], 0)
# Interpolate the alpha values
a = interp(a_scaled, array(norm_span),
array([min_alpha, 255]), left=0, right=255).astype(np.uint8)
r[mask] = g[mask] = b[mask] = 255
values = np.dstack([r, g, b, a]).view(np.uint32).reshape(a.shape)
if cupy and isinstance(values, cupy.ndarray):
# Convert cupy array to numpy for final image
values = cupy.asnumpy(values)
return Image(values,
dims=agg.dims[:-1],
coords=OrderedDict([
(agg.dims[1], agg.coords[agg.dims[1]]),
(agg.dims[0], agg.coords[agg.dims[0]]),
]),
name=name)