def draw(self):
"""
Renders the classification report; must be called after score.
"""
# Perform display related manipulations on the confusion matrix data
cm_display = self.confusion_matrix_
# Convert confusion matrix to percent of each row, i.e. the
# predicted as a percent of true in each class.
if self.percent is True:
# Note: div_safe function returns 0 instead of NAN.
cm_display = div_safe(self.confusion_matrix_,
self.class_counts_.reshape(-1, 1))
cm_display = np.round(cm_display * 100, decimals=0)
# Y axis should be sorted top to bottom in pcolormesh
cm_display = cm_display[::-1, ::]
# Get the human readable labels
labels = self._labels()
if labels is None:
labels = self.classes_
# Set up the dimensions of the pcolormesh
n_classes = len(labels)
X, Y = np.arange(n_classes + 1), np.arange(n_classes + 1)
self.ax.set_ylim(bottom=0, top=cm_display.shape[0])
self.ax.set_xlim(left=0, right=cm_display.shape[1])
# Fetch the grid labels from the classes in correct order; set ticks.
xticklabels = labels
yticklabels = labels[::-1]
ticks = np.arange(n_classes) + 0.5
self.ax.set(xticks=ticks, yticks=ticks)
self.ax.set_xticklabels(xticklabels,
rotation="vertical",
fontsize=self.fontsize)
self.ax.set_yticklabels(yticklabels, fontsize=self.fontsize)
# Set data labels in the grid enumerating over all x,y class pairs.
# NOTE: X and Y are one element longer than the confusion matrix, so
# skip the last element in the enumeration to label grids.
for x in X[:-1]:
for y in Y[:-1]:
# Extract the value and the text label
value = cm_display[x, y]
svalue = "{:0.0f}".format(value)
if self.percent:
svalue += "%"
# Determine the grid and text colors
base_color = self.cmap(value / cm_display.max())
text_color = find_text_color(base_color)
# Make zero values more subtle
if cm_display[x, y] == 0:
text_color = CMAP_MUTEDCOLOR
# Add the label to the middle of the grid
cx, cy = x + 0.5, y + 0.5
self.ax.text(
cy,
cx,
svalue,
va="center",
ha="center",
color=text_color,
fontsize=self.fontsize,
)
# Add a dark line on the grid with the diagonal. Note that the
# tick labels have already been reversed.
lc = "k" if xticklabels[x] == yticklabels[y] else "w"
self._edgecolors.append(lc)
# Draw the heatmap with colors bounded by vmin,vmax
vmin = 0.00001
vmax = 99.999 if self.percent is True else cm_display.max()
self.ax.pcolormesh(
X,
Y,
cm_display,
vmin=vmin,
vmax=vmax,
edgecolor=self._edgecolors,
cmap=self.cmap,
linewidth="0.01",
)
# Return the axes being drawn on
return self.ax
def ConfusionMatrixViz(self, percent=True):
"""
Renders the classification report; must be called after score.
"""
labels = [0, 1]
confusion_matrix_ = confusion_matrix_metric(self.y_true,
self.y_pred,
labels=labels)
class_counts_ = dict(zip(*np.unique(self.y_true, return_counts=True)))
# Make array of only the classes actually being used.
# Needed because sklearn confusion_matrix only returns counts for
# selected classes but percent should be calculated on all classes
selected_class_counts = []
for c in labels:
try:
selected_class_counts.append(class_counts_[c])
except KeyError:
selected_class_counts.append(0)
class_counts_ = np.array(selected_class_counts)
# Perform display related manipulations on the confusion matrix data
cm_display = confusion_matrix_
# Convert confusion matrix to percent of each row, i.e. the
# predicted as a percent of true in each class.
if percent is True:
# Note: div_safe function returns 0 instead of NAN.
cm_display = div_safe(confusion_matrix_,
class_counts_.reshape(-1, 1))
cm_display = np.round(cm_display * 100, decimals=0)
# Y axis should be sorted top to bottom in pcolormesh
cm_display = cm_display[::-1, ::]
# Set up the dimensions of the pcolormesh
n_classes = len(self.classes)
X, Y = np.arange(n_classes + 1), np.arange(n_classes + 1)
fig, ax = plt.subplots(ncols=1, nrows=1)
ax.set_ylim(bottom=0, top=cm_display.shape[0])
ax.set_xlim(left=0, right=cm_display.shape[1])
# Fetch the grid labels from the classes in correct order; set ticks.
xticklabels = self.classes
yticklabels = self.classes[::-1]
ticks = np.arange(n_classes) + 0.5
ax.set(xticks=ticks, yticks=ticks)
ax.set_xticklabels(xticklabels,
rotation="vertical",
fontsize=self.fontsize)
ax.set_yticklabels(yticklabels, fontsize=self.fontsize)
# Set data labels in the grid enumerating over all x,y class pairs.
# NOTE: X and Y are one element longer than the confusion matrix, so
# skip the last element in the enumeration to label grids.
for x in X[:-1]:
for y in Y[:-1]:
# Extract the value and the text label
value = cm_display[x, y]
svalue = "{:0.0f}".format(value)
if percent:
svalue += "%"
# Determine the grid and text colors
base_color = self.cmap(value / cm_display.max())
text_color = find_text_color(base_color)
# Make zero values more subtle
if cm_display[x, y] == 0:
text_color = "0.75"
# Add the label to the middle of the grid
cx, cy = x + 0.5, y + 0.5
ax.text(
cy,
cx,
svalue,
va="center",
ha="center",
color=text_color,
fontsize=self.fontsize,
)
# Add a dark line on the grid with the diagonal. Note that the
# tick labels have already been reversed.
lc = "k" if xticklabels[x] == yticklabels[y] else "w"
self._edgecolors.append(lc)
# Draw the heatmap with colors bounded by vmin,vmax
vmin = 0.00001
vmax = 99.999 if percent is True else cm_display.max()
ax.pcolormesh(
X,
Y,
cm_display,
vmin=vmin,
vmax=vmax,
edgecolor=self._edgecolors,
cmap=self.cmap,
linewidth="0.01",
)
ax.set_title("Confusion Matrix for {}".format(self.name))
ax.set_ylabel("True Class")
ax.set_xlabel("Predicted Class")
# Call tight layout to maximize readability
fig.tight_layout()
fig.savefig(self.path_to_save + "/ConfusionMatrix_" + self.name +
".pdf")
# Return the axes being drawn on
return ax