def reliability_hnh(hnh_array, save_path=None, title="可靠性图"):
'''
根据中间结果计算
:param th:
:param save_path:
:return:
'''
total_grade_num = hnh_array[:, 0]
observed_grade_num = hnh_array[:, 1]
ngrade = len(total_grade_num)
grade = 1 / ngrade
total_num = np.sum(total_grade_num)
under = np.zeros_like(total_grade_num)
under[:] = total_grade_num[:]
under[total_grade_num == 0] = 1
ob_rate = observed_grade_num / under
ob_rate[total_grade_num == 0] = IV
ob_rate_noIV = set_plot_IV(ob_rate)
ob_rate[total_grade_num == 0] = np.nan
index_iv = np.where(total_grade_num == 0)
line_x = np.arange(0, 1.00, grade)
prefect_line_y = np.arange(0, 1.00, grade)
climate_line_y = np.ones_like(line_x) * np.sum(
observed_grade_num) / total_num
x = np.arange(grade / 2, 1, grade)
fig = plt.figure(figsize=(5, 5.1))
grid_plt = plt.GridSpec(5, 1, hspace=0)
ax1 = plt.subplot(grid_plt[0:4, 0])
plt.plot(x, ob_rate_noIV, "--", linewidth=0.5, color="k")
x_iv = x[index_iv[0]]
ob_rate_noIV_iv = ob_rate_noIV[index_iv[0]]
plt.plot(x_iv, ob_rate_noIV_iv, "x", color='k')
plt.plot(x, ob_rate, marker=".", markersize="10", label="实际预报", color="r")
plt.plot(line_x, prefect_line_y, '--', label="完美预报", color="k")
plt.plot(line_x, climate_line_y, ':', label="无技巧预报", color="k")
plt.setp(ax1.get_xticklabels(), visible=False)
plt.ylim(0.0, 1)
plt.ylabel("实况的发生比例")
plt.legend(loc=2)
plt.title(title)
ax2 = plt.subplot(grid_plt[4, 0], sharex=ax1)
plt.bar(x, total_grade_num, width=0.03)
#plt.setp(ax2.get_xticklabels())
plt.ylabel("样本数")
plt.xlim(0.0, 1.0)
plt.xticks(np.arange(0.1, 1.01, 0.1))
plt.xlabel("预测的概率")
if save_path is None:
plt.show()
else:
plt.savefig(save_path)
print("检验结果已以图片形式保存至" + save_path)
plt.close()
def comprehensive_hnh(hnh_array, member_list=None,vmax = None,log_y = False, save_path=None,show = False,dpi = 300, title="概率预报综合检验图"):
'''
:param th_array:
:param save_path:
:return:
'''
sup_fontsize = 10
fig = plt.figure(figsize=(8, 5.6),dpi = dpi)
grade_count = hnh_array.shape[-2]
grade = 1 / grade_count
if grade_count < 1:
print('grade_count输入错误，不能小于1')
return
grade_list = np.arange(0, 1, grade).tolist()
grade_list.append(1.1)
shape = list(hnh_array.shape)
new_th_array = hnh_array.reshape((-1, len(grade_list) - 1, 2))
new_th_array_shape = new_th_array.shape
label = []
legend_num = new_th_array_shape[0]
if member_list is None:
if legend_num == 1:
label.append('预报')
else:
for i in range(legend_num):
label.append('预报' + str(i + 1))
else:
label.extend(member_list)
color_list = meteva.base.tool.color_tools.get_color_list(legend_num)
bar_width = 1.0 / (grade_count * (legend_num + 3))
ymax = -9999
mark_line_x = []
mark_line_y = []
for line in range(legend_num):
total_grade_num = new_th_array[line, :, 0]
observed_grade_num = new_th_array[line, :, 1]
total_num = np.sum(total_grade_num)
under = np.zeros_like(total_grade_num)
under[:] = total_grade_num[:]
under[total_grade_num == 0] = 1
ob_rate = observed_grade_num / under
ob_rate[total_grade_num == 0] = IV
ob_rate_noIV = set_plot_IV(ob_rate)
ob_rate[total_grade_num == 0] = np.nan
index_iv = np.where(total_grade_num == 0)
not_observed_grade_num = total_grade_num - observed_grade_num
ngrade = len(total_grade_num)
grade = 1 / ngrade
x = np.arange(grade / 2, 1, grade)
line_x = np.arange(0, 1.01, 0.1)
prefect_line_y = np.arange(0, 1.01, 0.1)
climate_line_y = np.ones_like(line_x) * np.sum(observed_grade_num) / total_num
grid_plt = plt.GridSpec(6, 2)
if line == 0:
plt.suptitle(title,fontsize= sup_fontsize,y=0.95)
plt.subplots_adjust(wspace=0.2, hspace=1)
ax3 = plt.subplot(grid_plt[0:4, 1])
ax4 = plt.subplot(grid_plt[4:, :])
ax1 = plt.subplot(grid_plt[0:4, 0])
ax1.plot(x, ob_rate_noIV, "--", linewidth=0.5, color="k")
x_iv = x[index_iv[0]]
ob_rate_noIV_iv = ob_rate_noIV[index_iv[0]]
ax1.plot(x_iv, ob_rate_noIV_iv, "x", color='k',label = None)
if line == 0:
ax1.plot(line_x, prefect_line_y, '--', label="完美", color="k")
ax1.plot(line_x, climate_line_y, ':', label="无技巧", color="k")
ax1.plot(x, ob_rate, marker=".", markersize="10", color=color_list[line],label = None)
x1 = x + (line -legend_num / 2 + 0.5) * bar_width
total_hap = np.sum(observed_grade_num)
hfmc = np.zeros((len(total_grade_num), 4))
for i in range(ngrade):
hfmc[i, 0] = np.sum(observed_grade_num[i:])
hfmc[i, 1] = np.sum(total_grade_num[i:]) - hfmc[i, 0]
hfmc[i, 2] = total_hap - hfmc[i, 0]
hfmc[i, 3] = total_num - (hfmc[i, 0] + hfmc[i, 1] + hfmc[i, 2])
far = [1]
far.extend(pofd_hfmc(hfmc).tolist())
far.append(0)
pod = [1]
pod.extend(pod_hfmc(hfmc).tolist())
pod.append(0)
far = np.array(far)
pod = np.array(pod)
if line == 0:
ax3.plot([0, 1], [0, 1], ":", color="k", linewidth=1, label="无技巧")
ax3.plot(far, pod, color=color_list[line], linewidth=2, label=label[line])
#ax4.bar(x1, observed_grade_num, width=bar_width*0.8, color=color_list[line],label = label[line])
ax4.bar(x1, (not_observed_grade_num+ observed_grade_num), width=bar_width*0.8,edgecolor=color_list[line],label =label[line])
ax4.set_xlabel("预测的概率",fontsize= sup_fontsize *0.9)
ax4.set_ylabel("样本数",fontsize= sup_fontsize *0.9)
ymax = max(np.max(observed_grade_num+not_observed_grade_num), ymax)
mark_line_x.append(x1)
mark_line_y.append((not_observed_grade_num+ observed_grade_num)* x)
ax1.set_xlim(0.0, 1)
ax1.set_ylim(0.0, 1)
ax1.set_xlabel("预测的概率",fontsize= sup_fontsize *0.9)
ax1.set_ylabel("实况的发生比例",fontsize= sup_fontsize *0.9)
ax1.legend(loc=2,fontsize= sup_fontsize *0.9)
ax3.set_xlabel("空报率",fontsize= sup_fontsize *0.9)
ax3.set_ylabel("命中率",fontsize= sup_fontsize *0.9)
ax3.set_ylim(0.0, 1.0)
ax3.set_xlim(0.0, 1.0)
ax3.legend(loc=4,fontsize= sup_fontsize *0.9)
ax4.set_xlim(0.0, 1)
ax4.set(title='\n')
ax4.set_xticks(np.arange(0,1.01,1/grade_count))
#ax4.legend(loc="upper right", ncol=2)
#mark_line_x = np.array(mark_line_x)
#mark_line_y = np.array(mark_line_y)
#mark_line_x = mark_line_x.T.flatten()
#mark_line_y = mark_line_y.T.flatten()
#ax4.plot(mark_line_x, mark_line_y, '.', color='k',markersize = bar_width * 300)
#lines = ax4.get_children()
#ax4.legend([lines[0], lines[grade_count],lines[grade_count * legend_num * 2 ]], ['观测正例', '观测负例',"合理比例"],
# loc='upper center', bbox_to_anchor=(0.5, 1.0), ncol=3, prop={'size': 6},fontsize= sup_fontsize *0.9)
if log_y: ax4.set_yscale('log')
if vmax is None:
ax4.set_ylim(0.0, ymax * 1.5)
else:
ax4.set_ylim(0.0, vmax)
if save_path is None:
show = True
else:
plt.savefig(save_path,bbox_inches='tight')
print("检验结果已以图片形式保存至" + save_path)
if show is True:
plt.show()
plt.close()
def reliability_hnh(hnh_array, member_list=None,vmax = None,log_y = False, save_path=None,show = False,dpi = 300, title="可靠性图"):
'''
根据中间结果计算
:param th:
:param save_path:
:return:
'''
sup_fontsize = 10
grade_count = hnh_array.shape[-2]
grade = 1 / grade_count
if grade_count < 1:
print('grade_count输入错误，不能小于1')
return
grade_list = np.arange(0, 1, grade).tolist()
grade_list.append(1.1)
hnh_array = hnh_array.reshape((-1, len(grade_list) - 1, 2))
new_hnh_array_shape = hnh_array.shape
label = []
legend_num = new_hnh_array_shape[0]
if member_list is None:
if legend_num== 1:
label.append('预报')
else:
for i in range(legend_num):
label.append('预报' + str(i + 1))
else:
label.extend(member_list)
color_list = meteva.base.tool.color_tools.get_color_list(legend_num)
for line in range(new_hnh_array_shape[0]):
total_grade_num = hnh_array[line, :, 0]
observed_grade_num = hnh_array[line, :, 1]
ngrade = len(total_grade_num)
grade = 1 / ngrade
total_num = np.sum(total_grade_num)
under = np.zeros_like(total_grade_num)
under[:] = total_grade_num[:]
under[total_grade_num == 0] = 1
ob_rate = observed_grade_num / under
ob_rate[total_grade_num == 0] = IV
ob_rate_noIV = set_plot_IV(ob_rate)
ob_rate[total_grade_num == 0] = np.nan
index_iv = np.where(total_grade_num == 0)
line_x = np.arange(0, 1.00, grade)
prefect_line_y = np.arange(0, 1.00, grade)
climate_line_y = np.ones_like(line_x) * np.sum(observed_grade_num) / total_num
x = np.arange(grade / 2, 1, grade)
if line == 0:
fig = plt.figure(figsize=(5.5, 6.1),dpi = dpi)
grid_plt = plt.GridSpec(5, 1, hspace=0)
ax1 = plt.subplot(grid_plt[0:4, 0])
plt.plot(line_x, prefect_line_y, '--', label="完美", color="k")
plt.plot(line_x, climate_line_y, ':', label="无技巧", color="k")
plt.plot(x, ob_rate_noIV, "--", linewidth=0.5, color="k")
x_iv = x[index_iv[0]]
ob_rate_noIV_iv = ob_rate_noIV[index_iv[0]]
plt.plot(x_iv, ob_rate_noIV_iv, "x", color='k')
plt.plot(x, ob_rate, marker=".", markersize="10", label=label[line], color=color_list[line])
plt.setp(ax1.get_xticklabels(), visible=False)
plt.ylim(0.0, 1)
plt.ylabel("正样本比例",fontsize = sup_fontsize * 0.9)
plt.legend(loc=2,fontsize = sup_fontsize * 0.9)
plt.title(title,fontsize = sup_fontsize)
bar_width = 0.8 / (grade_count * (legend_num+2))
for line in range(new_hnh_array_shape[0]):
total_grade_num = hnh_array[line, :, 0]
observed_grade_num = hnh_array[line, :, 1]
ngrade = len(total_grade_num)
grade = 1 / ngrade
under = np.zeros_like(total_grade_num)
under[:] = total_grade_num[:]
under[total_grade_num == 0] = 1
ob_rate = observed_grade_num / under
ob_rate[total_grade_num == 0] = IV
ob_rate[total_grade_num == 0] = np.nan
x = np.arange(grade / 2, 1, grade)
if line == 0:
ax2 = plt.subplot(grid_plt[4, 0], sharex=ax1)
#x1 = x - 0.01 + (line + 1.5) * 0.01
x1 = x + (line - legend_num/2 + 0.5) * bar_width
plt.bar(x1, total_grade_num, width=bar_width*0.8,color=color_list[line])
plt.ylabel("样本数",fontsize = sup_fontsize * 0.9)
if log_y: plt.yscale('log')
plt.xlim(0.0, 1.0)
plt.xticks(np.arange(0.1, 1.01, 0.1))
plt.xlabel("预测的概率",fontsize = sup_fontsize * 0.9)
if vmax is not None:
plt.ylim(0,vmax)
if save_path is None:
show = True
else:
plt.savefig(save_path,bbox_inches='tight')
print("检验结果已以图片形式保存至" + save_path)
if show is True:
plt.show()
plt.close()
def comprehensive_hnh(th_array, save_path=None, title="概率预报综合检验图"):
'''
:param th_array:
:param save_path:
:return:
'''
total_grade_num = th_array[:, 0]
observed_grade_num = th_array[:, 1]
total_num = np.sum(total_grade_num)
under = np.zeros_like(total_grade_num)
under[:] = total_grade_num[:]
under[total_grade_num == 0] = 1
ob_rate = observed_grade_num / under
ob_rate[total_grade_num == 0] = IV
ob_rate_noIV = set_plot_IV(ob_rate)
ob_rate[total_grade_num == 0] = np.nan
index_iv = np.where(total_grade_num == 0)
not_observed_grade_num = total_grade_num - observed_grade_num
ngrade = len(total_grade_num)
grade = 1 / ngrade
x = np.arange(grade / 2, 1, grade)
line_x = np.arange(0, 1.01, 0.1)
prefect_line_y = np.arange(0, 1.01, 0.1)
climate_line_y = np.ones_like(line_x) * np.sum(
observed_grade_num) / total_num
fig = plt.figure(figsize=(10, 7))
title_lines = len(title.split("\n"))
plt.suptitle(title, y=0.90 + 0.03 * title_lines)
plt.subplots_adjust(wspace=0.2, hspace=1)
grid_plt = plt.GridSpec(6, 2)
ax1 = plt.subplot(grid_plt[0:4, 0])
plt.plot(x, ob_rate_noIV, "--", linewidth=0.5, color="k")
x_iv = x[index_iv[0]]
ob_rate_noIV_iv = ob_rate_noIV[index_iv[0]]
plt.plot(x_iv, ob_rate_noIV_iv, "x", color='k')
plt.plot(x, ob_rate, marker=".", markersize="10", label="实际预报", color="r")
plt.plot(line_x, prefect_line_y, '--', label="完美预报", color="k")
plt.plot(line_x, climate_line_y, ':', label="无技巧预报", color="k")
plt.xlim(0.0, 1.0)
plt.ylim(0.0, 1)
plt.xlabel("预测的概率")
plt.ylabel("实况的发生比例")
plt.legend(loc=2)
ax2 = plt.subplot(grid_plt[4:, 0])
plt.bar(x, total_grade_num, width=0.03)
#plt.setp(ax2.get_xticklabels())
plt.xlim(0.0, 1.0)
plt.ylabel("样本数")
plt.xlabel("预测的概率")
total_hap = np.sum(observed_grade_num)
hfmc = np.zeros((len(total_grade_num), 4))
for i in range(ngrade):
hfmc[i, 0] = np.sum(observed_grade_num[i:])
hfmc[i, 1] = np.sum(total_grade_num[i:]) - hfmc[i, 0]
hfmc[i, 2] = total_hap - hfmc[i, 0]
hfmc[i, 3] = total_num - (hfmc[i, 0] + hfmc[i, 1] + hfmc[i, 2])
far = [1]
far.extend(pofd_hfmc(hfmc).tolist())
far.append(0)
pod = [1]
pod.extend(pod_hfmc(hfmc).tolist())
pod.append(0)
far = np.array(far)
pod = np.array(pod)
ax3 = plt.subplot(grid_plt[0:4, 1])
plt.plot(far, pod, color="blue", linewidth=2, label="实际预报")
plt.plot([0, 1], [0, 1], ":", color="k", linewidth=1, label="无技巧预报")
plt.xlabel("空报率")
plt.ylabel("命中率")
plt.ylim(0.0, 1.0)
plt.xlim(0.0, 1.0)
plt.legend(loc=4)
ax4 = plt.subplot(grid_plt[4:, 1])
plt.bar(x - 0.01,
not_observed_grade_num / total_num,
width=0.02,
edgecolor='r',
fill=False,
label="未发生")
plt.bar(x + 0.01,
observed_grade_num / total_num,
width=0.02,
color='b',
label="已发生")
plt.xlabel("预测的概率")
plt.ylabel("占总样本数的比例")
ymax = max(np.max(observed_grade_num / total_num),
np.max(not_observed_grade_num / total_num)) * 1.5
plt.ylim(0.0, ymax)
plt.xlim(0.0, 1.0)
plt.legend(loc="upper right", ncol=2)
if save_path is None:
plt.show()
else:
plt.savefig(save_path)
print("检验结果已以图片形式保存至" + save_path)
plt.close()