Python:从matplotlib热图及其图例中保留Numpy NaN值

Question

我有一个numpy数组，需要绘制成一个热图。numpy数组还将包含我需要从绘图中排除的NaN值。我在其他帖子中被告知，numpy自动屏蔽了情节中的NaN值，但不知何故对我不起作用。这是一个示例代码

column_labels = list('ABCDEFGH')
row_labels = list('WXYZ')
fig, ax = plt.subplots()
data = np.array([[ 0.96753494,  0.52349944,  0.0254628 ,  0.5104103 ],
         [ 0.07320069,  0.91278731,  0.97094436,  0.70533351],
         [ 0.30162006,  0.49068337,  0.41837729,  0.71139215],
         [ 0.19786101,  0.15882713,  0.59028841,  0.06242765],
         [ 0.51505872,  0.07798389,  0.58790067,  0.44782683],
         [ 0.68975694,  0.53535385,  0.15696023,  0.35641951],
         [ 0.66481995,  0.03576846,  0.9623601 ,  0.96006395],
         [ 0.45865404,  0.50433582,  0.18182575,  0.35126449],])

data[3,:] = np.nan
heatmap = ax.pcolor(data, cmap=plt.cm.seismic)

fig.colorbar(heatmap)
# put the major ticks at the middle of each cell
ax.set_xticks(np.arange(data.shape[1])+0.5, minor=False)
ax.set_yticks(np.arange(data.shape[0])+0.5, minor=False)

# want a more natural, table-like display
ax.invert_yaxis()
ax.xaxis.tick_top()

ax.set_xticklabels(row_labels, minor=False)
ax.set_yticklabels(column_labels, minor=False)
plt.show()

情节看起来像

显然，这与没有Nan的情节非常不同，Nan看起来像

我想完全避免图例中的NaN值，最好用一些符号标记它，如X。我怎样才能做到这一点呢？

Answer 1

nans干扰了pcolor确定data中包含的值的范围，因为

In [72]: data.min(), data.max()
Out[72]: (nan, nan)

您可以通过使用np.nanmin和np.nanmax声明值的范围来解决此问题，以在data中找到最小和最大的非NaN值：

heatmap = ax.pcolor(data, cmap=plt.cm.seismic, 
                    vmin=np.nanmin(data), vmax=np.nanmax(data))

自从

In [73]: np.nanmin(data), np.nanmax(data)
Out[73]: (0.025462800000000001, 0.97094435999999995)

---

import numpy as np
import matplotlib.pyplot as plt

column_labels = list('ABCDEFGH')
row_labels = list('WXYZ')
fig, ax = plt.subplots()
data = np.array([[ 0.96753494,  0.52349944,  0.0254628 ,  0.5104103 ],
         [ 0.07320069,  0.91278731,  0.97094436,  0.70533351],
         [ 0.30162006,  0.49068337,  0.41837729,  0.71139215],
         [ 0.19786101,  0.15882713,  0.59028841,  0.06242765],
         [ 0.51505872,  0.07798389,  0.58790067,  0.44782683],
         [ 0.68975694,  0.53535385,  0.15696023,  0.35641951],
         [ 0.66481995,  0.03576846,  0.9623601 ,  0.96006395],
         [ 0.45865404,  0.50433582,  0.18182575,  0.35126449],])

data[3,:] = np.nan
heatmap = ax.pcolor(data, cmap=plt.cm.seismic, 
                    vmin=np.nanmin(data), vmax=np.nanmax(data))
heatmap.cmap.set_under('black')

bar = fig.colorbar(heatmap, extend='both')

# put the major ticks at the middle of each cell
ax.set_xticks(np.arange(data.shape[1])+0.5, minor=False)
ax.set_yticks(np.arange(data.shape[0])+0.5, minor=False)

# want a more natural, table-like display
ax.invert_yaxis()
ax.xaxis.tick_top()

ax.set_xticklabels(row_labels, minor=False)
ax.set_yticklabels(column_labels, minor=False)
plt.show() 

---

另一个选项（基于Joe Kington's solution）是绘制 在data为NaN的地方有填充标记的矩形块。

上面的示例显示，在NaN值为的单元格中，pcolor的颜色为 尽管南斯是非常负的数字。相反，如果通过pcolora 蒙版数组使蒙版区域保持透明。因此，你可以画 在轴背景补丁ax.patch上的阴影，以显示阴影标记 在蒙面区域。

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches

column_labels = list('ABCDEFGH')
row_labels = list('WXYZ')
fig, ax = plt.subplots()
data = np.array([[ 0.96753494,  0.52349944,  0.0254628 ,  0.5104103 ],
         [ 0.07320069,  0.91278731,  0.97094436,  0.70533351],
         [ 0.30162006,  0.49068337,  0.41837729,  0.71139215],
         [ 0.19786101,  0.15882713,  0.59028841,  0.06242765],
         [ 0.51505872,  0.07798389,  0.58790067,  0.44782683],
         [ 0.68975694,  0.53535385,  0.15696023,  0.35641951],
         [ 0.66481995,  0.03576846,  0.9623601 ,  0.96006395],
         [ 0.45865404,  0.50433582,  0.18182575,  0.35126449],])

data[3,:] = np.nan
data = np.ma.masked_invalid(data)

heatmap = ax.pcolor(data, cmap=plt.cm.seismic, 
                    vmin=np.nanmin(data), vmax=np.nanmax(data))
# https://stackoverflow.com/a/16125413/190597 (Joe Kington)
ax.patch.set(hatch='x', edgecolor='black')
fig.colorbar(heatmap)

# put the major ticks at the middle of each cell
ax.set_xticks(np.arange(data.shape[1])+0.5, minor=False)
ax.set_yticks(np.arange(data.shape[0])+0.5, minor=False)

# want a more natural, table-like display
ax.invert_yaxis()
ax.xaxis.tick_top()

ax.set_xticklabels(row_labels, minor=False)
ax.set_yticklabels(column_labels, minor=False)
plt.show() 

---

如果希望使用多种类型的填充标记，例如一种用于NaNs，另一种用于负值，则可以使用循环添加填充矩形：

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches

column_labels = list('ABCDEFGH')
row_labels = list('WXYZ')
fig, ax = plt.subplots()
data = np.array([[ 0.96753494,  0.52349944,  0.0254628 ,  0.5104103 ],
         [ 0.07320069,  0.91278731,  0.97094436,  0.70533351],
         [ 0.30162006,  0.49068337,  0.41837729,  0.71139215],
         [ 0.19786101,  0.15882713,  0.59028841,  0.06242765],
         [ 0.51505872,  0.07798389,  0.58790067,  0.44782683],
         [ 0.68975694,  0.53535385,  0.15696023,  0.35641951],
         [ 0.66481995,  0.03576846,  0.9623601 ,  0.96006395],
         [ 0.45865404,  0.50433582,  0.18182575,  0.35126449],])
data -= 0.5
data[3,:] = np.nan
data = np.ma.masked_invalid(data)
heatmap = ax.pcolor(data, cmap=plt.cm.seismic, 
                    vmin=np.nanmin(data), vmax=np.nanmax(data))

# https://stackoverflow.com/a/16125413/190597 (Joe Kington)
ax.patch.set(hatch='x', edgecolor='black')

# draw a hatched rectangle wherever the data is negative
# http://matthiaseisen.com/pp/patterns/p0203/
mask = data < 0
for j, i in np.column_stack(np.where(mask)):
      ax.add_patch(
          mpatches.Rectangle(
              (i, j),     # (x,y)
              1,          # width
              1,          # height
              fill=False, 
              edgecolor='blue',
              snap=False,
              hatch='x' # the more slashes, the denser the hash lines 
          ))

fig.colorbar(heatmap)

# put the major ticks at the middle of each cell
ax.set_xticks(np.arange(data.shape[1])+0.5, minor=False)
ax.set_yticks(np.arange(data.shape[0])+0.5, minor=False)

# want a more natural, table-like display
ax.invert_yaxis()
ax.xaxis.tick_top()

ax.set_xticklabels(row_labels, minor=False)
ax.set_yticklabels(column_labels, minor=False)
plt.show()