def score1(sta_ob_and_fos0,method,s = None,g = None,gll = None,para1 = None,para2 = None,plot = "line",show = False,excel_path = None):
if s is not None:
if g is not None:
if g == "last_range" or g == "last_step":
s["drop_last"] = False
else:
s["drop_last"] = True
sta_ob_and_fos = sele_by_dict(sta_ob_and_fos0, s)
if type(method) == str:
method = globals().get(method)
if method == meteva.method.FSS_time:
if g == "dtime":
print("FSS_time 检验时，参数group_by不能选择dtime")
return
sta_ob_and_fos_list,group_list_list1 = group(sta_ob_and_fos,g,gll)
data_name = meteva.base.get_stadata_names(sta_ob_and_fos)
fo_num = len(data_name) -1
ensemble_score_method = [meteva.method.cr]
group_num = len(sta_ob_and_fos_list)
if para1 is None:
para_num = 1
else:
para_name_list = []
mutil_list = [meteva.method.ts_multi,meteva.method.bias_multi,meteva.method.ets_multi,
meteva.method.mr_multi,meteva.method.far_multi]
if method in mutil_list:
para_num = len(para1)+1
para_name_list = ["<" + str(para1[0])]
for i in range(len(para1)-1):
para_name_list.append("["+str(para1[i]) + ","+str(para1[i+1])+")")
para_name_list.append(">=" + str(para1[-1]))
else:
para_num = len(para1)
for i in range(len(para1)):
para_name_list.append(para1[i])
sta_result = None
if method == meteva.method.FSS_time:
#统计dtime的集合
dtime_list = list(set(sta_ob_and_fos["dtime"].values.tolist()))
dtime_list.sort()
ndtime = len(dtime_list)
result= []
for sta_ob_and_fo in sta_ob_and_fos_list:
# 将观测和预报数据重新整理成FSS_time所需格式
ob = in_member_list(sta_ob_and_fo,[data_name[0]])
ob_dtimes = None
for k in range(ndtime):
dtimek = dtime_list[k]
sta_obk = in_dtime_list(ob,[dtimek])
set_stadata_names(sta_obk,[data_name[0]+ str(dtimek)])
ob_dtimes = combine_on_leve_time_id(ob_dtimes,sta_obk)
result1 = []
#print(ob_dtimes)
ob_array = ob_dtimes.values[:,6:]
for j in range(fo_num):
fo = in_member_list(sta_ob_and_fo, [data_name[j+1]])
fo_dtimes = None
for k in range(ndtime):
dtimek = dtime_list[k]
sta_fok = in_dtime_list(fo, [dtimek])
set_stadata_names(sta_fok, [data_name[j+1] + str(dtimek)])
fo_dtimes = combine_on_leve_time_id(fo_dtimes, sta_fok)
fo_array = fo_dtimes.values[:,6:]
#调用检验程序
if para1 is None:
para1 = [1e-30]
result2 = FSS_time(ob_array, fo_array, para1, para2)
result1.append(result2)
result.append(result1)
result = np.array(result) #将数据转换成数组
result = result.squeeze()
else:
nead_lon_lat = False
if g == "id" and gll is None:
nead_lon_lat = True
lon_lat_list = []
if method in ensemble_score_method:
result = np.zeros((group_num,para_num))
for i in range(group_num):
sta = sta_ob_and_fos_list[i]
#if(len(sta.index) == 0):
# result[i,:] = meteva.base.IV
#else:
ob = sta[data_name[0]].values
fo = sta[data_name[1:]].values
if para1 is None:
result[i, :] = method(ob, fo)
else:
result[i,:] = method(ob, fo,para1)
if nead_lon_lat:
lon_lat_list.append(sta.iloc[0,[4,5]])
else:
if fo_num ==0:
result = np.zeros((group_num,para_num))
else:
result = np.zeros((group_num,fo_num,para_num))
for i in range(group_num):
#print(group_num)
sta = sta_ob_and_fos_list[i]
#if(len(sta.index) == 0):
# result[i,:] = meteva.base.IV
#else:
ob = sta[data_name[0]].values
if fo_num>0:
for j in range(fo_num):
fo = sta[data_name[j+1]].values
if para1 is None:
result[i, j] = method(ob, fo)
else:
result[i,j] = method(ob, fo,para1)
else:
if para1 is None:
result[i] = method(ob, None)
else:
result[i] = method(ob, None,para1)
if nead_lon_lat:
lon_lat_list.append(sta.iloc[0,[4,5]])
# 将结果输出到excel
if excel_path is not None:
meteva.base.creat_path(excel_path)
if fo_num ==0:
fo_num = 1
result.reshape([group_num,fo_num,para_num])
group_dict ={"group":group_list_list1}
model_dict = {"member":data_name[1:]}
para_dict = {"threshold":para_name_list}
name_dict_list = [group_dict,model_dict,para_dict]
meteva.base.write_array_to_excel(result,excel_path,name_dict_list)
result = result.squeeze()
if nead_lon_lat:
df = pd.DataFrame(lon_lat_list)
df["id"] = group_list_list1
df["level"] = np.NAN
df["time"] = np.NAN
df["dtime"] = np.NAN
if fo_num ==0:
if para1 is None:
df["ob"] = result
else:
if isinstance(para1,list):
for i in range(len(para1)):
para = para1[i]
df[para] = result[:,i]
else:
df[para1] = result
else:
result = result.reshape(group_num,fo_num,para_num)
for j in range(fo_num):
if para1 is None:
df[data_name[1+j]] = result[:,j,0]
else:
if isinstance(para1, list):
for i in range(len(para1)):
para = para1[i]
df[data_name[1+j]+"_"+str(para)] = result[:,j, i]
else:
df[data_name[1 + j]] = result[:,j,0]
sta_result = sta_data(df)
if show:
if plot == "line":
if g == "dtime":
x = np.arange(len(group_list_list1))
plt.plot(result,label = data_name[1])
plt.xticks(x,group_list_list1)
plt.xlabel("预报时效")
plt.ylabel("ME")
plt.legend()
plt.show()
return result,group_list_list1,sta_result
def score(sta_ob_and_fos0,
method,
s=None,
g=None,
gll=None,
group_name_list=None,
plot=None,
save_path=None,
show=False,
dpi=300,
title="",
excel_path=None,
**kwargs):
if s is not None:
if g is not None:
if g == "last_range" or g == "last_step":
s["drop_last"] = False
else:
s["drop_last"] = True
sta_ob_and_fos = sele_by_dict(sta_ob_and_fos0, s)
if type(method) == str:
method = globals().get(method)
if method == meteva.method.FSS_time:
if g == "dtime":
print("FSS_time 检验时，参数group_by不能选择dtime")
return
sta_ob_and_fos_list, group_list_list1 = group(sta_ob_and_fos, g, gll)
group_num = len(sta_ob_and_fos_list)
data_name = meteva.base.get_stadata_names(sta_ob_and_fos_list[0])
if method.__name__.find("ob_fo") >= 0:
fo_name = data_name
else:
ensemble_score_method = [meteva.method.cr]
if method in ensemble_score_method:
fo_name = [""]
else:
fo_name = data_name[1:]
fo_num = len(fo_name)
#等级参数的确定
if "grade_list" not in kwargs.keys():
mutil_list = [
meteva.method.ts_multi, meteva.method.bias_multi,
meteva.method.ets_multi, meteva.method.mr_multi,
meteva.method.far_multi
]
if method in mutil_list:
# 如果是多分类检验，但又没有设置分级方法，就需要从数据中获得全局的种类
values = sta_ob_and_fos.iloc[:, 6:].flatten()
index_list = list(set(values))
if len(index_list) > 30:
print("自动识别的样本类别超过30种，判断样本为连续型变量，grade_list不能缺省")
return
index_list.sort()
grade_list = []
for i in range(len(index_list) - 1):
grade_list.append((index_list[i] + index_list[i + 1]) / 2)
kwargs["grade_list"] = grade_list
if "grade_list" in kwargs.keys():
grades = kwargs["grade_list"]
grade_names = []
mutil_list1 = [
meteva.method.ts_multi, meteva.method.bias_multi,
meteva.method.ets_multi, meteva.method.mr_multi,
meteva.method.far_multi
]
mutil_list2 = [
meteva.method.accuracy, meteva.method.hk, meteva.method.hss
]
if method in mutil_list1:
grade_names = ["<" + str(grades[0])]
for i in range(len(grades) - 1):
grade_names.append("[" + str(grades[i]) + "," +
str(grades[i + 1]) + ")")
grade_names.append(">=" + str(grades[-1]))
elif method in mutil_list2:
grade_names = ["0"]
else:
for i in range(len(grades)):
grade_names.append(grades[i])
else:
grade_names = ["0"]
grade_num = len(grade_names)
if method == meteva.method.FSS_time:
#统计dtime的集合
dtime_list = list(set(sta_ob_and_fos["dtime"].values.tolist()))
dtime_list.sort()
ndtime = len(dtime_list)
result = []
for sta_ob_and_fo in sta_ob_and_fos_list:
# 将观测和预报数据重新整理成FSS_time所需格式
ob = in_member_list(sta_ob_and_fo, [data_name[0]])
ob_dtimes = None
for k in range(ndtime):
dtimek = dtime_list[k]
sta_obk = in_dtime_list(ob, [dtimek])
set_stadata_names(sta_obk, [data_name[0] + str(dtimek)])
ob_dtimes = combine_on_leve_time_id(ob_dtimes, sta_obk)
result1 = []
#print(ob_dtimes)
ob_array = ob_dtimes.values[:, 6:]
for j in range(fo_num):
fo = in_member_list(sta_ob_and_fo, [data_name[j + 1]])
fo_dtimes = None
for k in range(ndtime):
dtimek = dtime_list[k]
sta_fok = in_dtime_list(fo, [dtimek])
set_stadata_names(sta_fok,
[data_name[j + 1] + str(dtimek)])
fo_dtimes = combine_on_leve_time_id(fo_dtimes, sta_fok)
fo_array = fo_dtimes.values[:, 6:]
#调用检验程序
result2 = FSS_time(ob_array, fo_array, **kwargs)
result1.append(result2)
result.append(result1)
result = np.array(result) #将数据转换成数组
result = result.squeeze()
else:
result_list = []
for i in range(group_num):
sta = sta_ob_and_fos_list[i]
#if(len(sta.index) == 0):
# result[i,:] = meteva.base.IV
#else:
ob = sta[data_name[0]].values
fo = sta[data_name[1:]].values.T
result1 = method(ob, fo, **kwargs)
result_list.append(result1)
result = np.array(result_list)
if plot is not None or excel_path is not None:
result_plot = result.reshape((group_num, fo_num, grade_num))
name_list_dict = {}
if g is None:
group_dict_name = "group_name"
else:
group_dict_name = g
#设置分组名称
if group_name_list is not None:
if group_num == len(group_name_list):
name_list_dict[group_dict_name] = group_name_list
else:
print("group_name_list参数中包含的分组名称个数和实际分组个数不匹配")
else:
if not isinstance(group_list_list1, list):
group_list_list1 = [group_list_list1]
name_list_dict[group_dict_name] = get_group_name(group_list_list1)
#设置成员名称
name_list_dict["member"] = fo_name
#设置等级名称
name_list_dict["grade"] = grade_names
keys = list(name_list_dict.keys())
if fo_num == 1:
if grade_num > 1:
legend = keys[0]
axis = keys[2]
else:
axis = keys[0]
legend = keys[1]
else:
if group_num == 1:
if grade_num > 1:
legend = keys[1]
axis = keys[2]
else:
legend = keys[2]
axis = keys[1]
else:
legend = keys[1]
axis = keys[0]
ylabel = method.__name__.upper()
bigthan0_method = [
meteva.method.ts, meteva.method.ob_fo_hr, meteva.method.ob_fo_std,
meteva.method.ts_multi, meteva.method.s,
meteva.method.pc_of_sun_rain, meteva.method.bias_multi,
meteva.method.bias, meteva.method.pc, meteva.method.mr,
meteva.method.far, meteva.method.tc, meteva.method.roc_auc,
meteva.method.r, meteva.method.sr, meteva.method.cr,
meteva.method.pod, meteva.method.pofd, meteva.method.mse,
meteva.method.rmse, meteva.method.mae
]
if method in bigthan0_method:
vmin = 0
else:
vmin = None
if plot is not None:
if plot == "bar":
meteva.base.plot_tools.bar(result_plot,
name_list_dict,
legend=legend,
axis=axis,
vmin=vmin,
ylabel=ylabel,
save_path=save_path,
show=show,
dpi=dpi,
title=title)
else:
meteva.base.plot_tools.plot(result_plot,
name_list_dict,
legend=legend,
axis=axis,
vmin=vmin,
ylabel=ylabel,
save_path=save_path,
show=show,
dpi=dpi,
title=title)
if excel_path is not None:
meteva.base.write_array_to_excel(result_plot,
excel_path,
name_list_dict,
index=axis,
columns=legend)
result = result.squeeze()
return result, group_list_list1
def score(sta_ob_and_fos,
method,
group_by=None,
group_list_list=None,
para1=None,
para2=None):
if type(method) == str:
method = globals().get(method)
if method == meteva.method.FSS_time:
if group_by == "dtime":
print("FSS_time 检验时，参数group_by不能选择dtime")
return
sta_ob_and_fos_list, group_list_list1 = group(sta_ob_and_fos, group_by,
group_list_list)
data_name = meteva.base.get_stadata_names(sta_ob_and_fos)
fo_num = len(data_name) - 1
ensemble_score_method = [meteva.method.cr]
group_num = len(sta_ob_and_fos_list)
if para1 is None:
para_num = 1
else:
para_num = len(para1)
if method in ensemble_score_method:
result = np.zeros((group_num, para_num))
for i in range(group_num):
sta = sta_ob_and_fos_list[i]
if (len(sta.index) == 0):
result[i, :] = meteva.base.IV
else:
ob = sta[data_name[0]].values
fo = sta[data_name[1:]].values
if para1 is None:
result[i, :] = method(ob, fo)
else:
result[i, :] = method(ob, fo, para1)
elif method == meteva.method.FSS_time:
#统计dtime的集合
dtime_list = list(set(sta_ob_and_fos["dtime"].values.tolist()))
dtime_list.sort()
ndtime = len(dtime_list)
result = []
for sta_ob_and_fo in sta_ob_and_fos_list:
# 将观测和预报数据重新整理成FSS_time所需格式
ob = in_member_list(sta_ob_and_fo, [data_name[0]])
ob_dtimes = None
for k in range(ndtime):
dtimek = dtime_list[k]
sta_obk = in_dtime_list(ob, [dtimek])
set_stadata_names(sta_obk, [data_name[0] + str(dtimek)])
ob_dtimes = combine_on_leve_time_id(ob_dtimes, sta_obk)
result1 = []
#print(ob_dtimes)
ob_array = ob_dtimes.values[:, 6:]
for j in range(fo_num):
fo = in_member_list(sta_ob_and_fo, [data_name[j + 1]])
fo_dtimes = None
for k in range(ndtime):
dtimek = dtime_list[k]
sta_fok = in_dtime_list(fo, [dtimek])
set_stadata_names(sta_fok,
[data_name[j + 1] + str(dtimek)])
fo_dtimes = combine_on_leve_time_id(fo_dtimes, sta_fok)
fo_array = fo_dtimes.values[:, 6:]
#调用检验程序
if para1 is None:
para1 = [1e-30]
result2 = FSS_time(ob_array, fo_array, para1, para2)
result1.append(result2)
result.append(result1)
result = np.array(result) #将数据转换成数组
else:
if fo_num == 0:
result = np.zeros((group_num, para_num))
else:
result = np.zeros((group_num, fo_num, para_num))
for i in range(group_num):
#print(group_num)
sta = sta_ob_and_fos_list[i]
if (len(sta.index) == 0):
result[i, :] = meteva.base.IV
else:
ob = sta[data_name[0]].values
if fo_num > 0:
for j in range(fo_num):
fo = sta[data_name[j + 1]].values
if para1 is None:
result[i, j] = method(ob, fo)
else:
result[i, j] = method(ob, fo, para1)
else:
if para1 is None:
result[i] = method(ob, None)
else:
result[i] = method(ob, None, para1)
result = result.squeeze()
return result, group_list_list1
