def plot_EM_results():
# 绘制测试结果图像
from param_options import args_parser
from plot_indexes_resluts import plot_indicator_results
from util_tools import mkdir
args = args_parser()
args.dataset_path = './constructed_datasets_6_30(1)/'
exp_name = 'EM'
result_save_file = './results/' + exp_name + '/'
plots_file = './plot_results/' + exp_name + '/'
print(
'====================== Save every component indicator result ========================'
)
for indicator in ['rmse', 'd2', 'r2']:
print('{} results'.format(indicator))
fig_save_path = plots_file + indicator + '/'
mkdir(fig_save_path)
for component in args.select_dim:
results_logdir = [result_save_file + indicator + '/']
plot_indicator_results(results_logdir,
fig_save_path,
component,
xaxis='station',
leg=['EM'])
print(">>> Finished save resluts figures!")
def save_model(self, G, D, save_file='', station_name=''):
mkdir('./saved_model/' + save_file + '/')
torch.save(
D.state_dict(), './saved_model/' + save_file + '/independent_' +
station_name + '_D.pkl')
torch.save(
G.state_dict(), './saved_model/' + save_file + '/independent_' +
station_name + '_G.pkl')
def show_fed_eval_results(args, exp_name, results_saved_file,
results_plot_file, cross_validation_sets):
result_save_root = './{}/'.format(results_saved_file) + exp_name + '/'
plots_save_root = './{}/'.format(results_plot_file) + exp_name + '/'
indicator_list = ['rmse', 'd2', 'r2', 'all_rmse']
leg = ['Federated', 'Independent']
for indicator_name in indicator_list:
# 建立保存结果的文件夹
indicator_avg_results_csv_save_fpth = result_save_root + 'avg_' + indicator_name + '/'
for mode in leg:
mkdir(indicator_avg_results_csv_save_fpth + mode + '/')
# 计算每个数据集的几次实验的均值
for c in range(cross_validation_sets):
results_logdir = [
result_save_root + 'datasets_' + str(c) + '/' +
indicator_name + '/' + model_name + '/'
for model_name in ['fed', 'idpt']
]
compute_avg_of_data_in_file(args, c, results_logdir,
indicator_avg_results_csv_save_fpth,
indicator_name, leg)
# 绘制测试结果图像
print(
'\033[0;34;40m [Visulize] Save every component indicator result \033[0m'
)
print('[Visulize] {} results'.format(indicator_name))
results_logdir = [
result_save_root + 'avg_' + indicator_name + '/' + model_name + '/'
for model_name in leg
]
fig_save_path = plots_save_root + indicator_name + '/'
csv_save_fpth = result_save_root + 'avg_' + indicator_name + '/'
mkdir(fig_save_path)
# 将每个数据集计算的均值再计算总的均值和绘制方差均值线图
plot_indicator_avg_results_m(results_logdir,
fig_save_path,
'station',
indicator_name,
csv_save_fpth,
select_dim=arg.select_dim)
# plot_fed_avg_acc(results_logdir, indicator_name, fig_save_path)
# plot_indicator_results(results_logdir, fig_save_path, indicator_name)
print(
"\033[0;34;40m >>>[Visulize] Finished save {} resluts figures! \033[0m"
.format(indicator_name))
def plot_federated_indicator_avg_results(dataset_number):
args = args_parser()
# dataset_number = 'one_mi_v1((A5)_1)'
results_saved_file = 'results_one_dn'
results_plot_file = 'plot_results_one_dn'
exp_name = '{}_latest'.format(dataset_number, )
result_save_root = './{}/'.format(results_saved_file) + exp_name + '/'
plots_save_root = './{}/'.format(results_plot_file) + exp_name + '/'
indicator_name = 'all_rmse'
leg = ['Fed', 'Independent']
results_logdir = [
result_save_root + 'avg_' + indicator_name + '/' + model_name + '/'
for model_name in leg
]
fig_save_path = plots_save_root + indicator_name + '/'
csv_save_fpth = result_save_root + 'avg_' + indicator_name + '/'
mkdir(fig_save_path)
# 将每个数据集计算的均值再计算总的均值和绘制方差均值线图
plot_indicator_avg_results_m(results_logdir, fig_save_path, 'station',
indicator_name, csv_save_fpth)
test_param_name = 'missing_rate'
Dname_prefix = '(A{})'
Dname_suffix = 'norm_1r_1P'
elif generate_condition == 'one_time':
params_test_list = [1]
test_param_name = 'One_time'
for param in params_test_list:
if generate_condition == 'missing_rate':
args.missing_ratios = [param / 100] * len(args.selected_stations)
dataset_name = Dname_prefix.format(param, param,
param) + '_' + Dname_suffix
print(
'===================== Missing ratio {}========================'
.format(param))
elif generate_condition == 'one_time':
dataset_name = dataset_name_temp + '_' + test_param_name
dataset_path = './constructed_datasets/{}/'.format(dataset_name)
# 构造所有站点不同的缺失率数据
for exp_n in range(exp_num):
args.dataset_path = dataset_path + str(exp_n) + '/'
mkdir(args.dataset_path)
all_station_datasets = generate_selected_station_save_datasets(
args, if_random=True)
# 载入数据集（用于测试函数）
# all_station_datasets = get_saved_datasets(args)
print(all_station_datasets)
def plot_missing_data_statistics(dataset_name_l, item_missing_stat_l,
data_missing_number_l, r_stat_l, c_stat_l,
data_len_l, select_dim):
# 用于汇总行和列每个站点数据集有多少出连续缺失
r_stat_sum_l = []
c_stat_sum_l = []
# 用于汇总行和列每个站点数据集有多少数据点是连续缺失
c_mp_num_l = []
# 表示一个站点数据集总的缺失率
fig, ax = plt.subplots()
y = [
data_missing_number_l[i] / (data_len_l[i] * len(select_dim)) * 100
for i in range(len(dataset_name_l))
]
ax.plot(dataset_name_l, y)
ax.tick_params(labelsize=13)
ax.set_xlabel('Statioin', size=13)
ax.set_ylabel('Missing rate(%)', size=13)
# ax.set_title('Missing rate of each station')
save_path = 'E:/zx/Fed-AQ/experiments_results/Figures/Dataset_overview_v2/'
# save_path = save_path + dataset_name + '/'
mkdir(save_path)
plt.savefig(save_path + 'missing_rate' + '.eps')
plt.savefig(save_path + 'missing_rate' '.svg')
# 保存到本地json
save2json(save_path + 'missing_rate.json', y, dataset_name_l)
# 每个污染物成分的缺失分析
plt.style.use('ggplot')
fig1, ax1 = plt.subplots()
for dataset_name, item_missing_stat, data_len in zip(
dataset_name_l, item_missing_stat_l, data_len_l):
y = [item_missing_stat[d] / data_len * 100 for d in select_dim]
ax1.plot(select_dim, y, label=dataset_name)
# 保存到本地json
save2json(save_path + dataset_name + '_pollutants.json', y, select_dim)
ax1.tick_params(labelsize=13)
ax1.set_xlabel('Pollutants', size=13)
ax1.set_ylabel('Missing rate(%)', size=13)
# ax1.set_title('Missing rate of each pollutant')
ax1.legend()
# save_path = 'E:/zx/Fed-AQ/experiments_results/Figures/Dataset_overview/'
# save_path = save_path + dataset_name + '/'
# mkdir(save_path)
plt.savefig(save_path + 'pollutants_stat' + '.eps')
plt.savefig(save_path + 'pollutants_stat' '.svg')
# 行缺失
fig2, ax2 = plt.subplots()
for i, dataset_name, r_stat in zip(range(len(dataset_name_l)),
dataset_name_l, r_stat_l):
# bar graphs
y = [v for v in r_stat.values()]
x = [k for k in r_stat.keys()]
y = np.array(y)
x = np.array(x)
width = 0.25
ax2.bar(x + width * i, y, width, label=dataset_name)
ax2.tick_params(labelsize=13)
ax2.set_xlabel('Gap lengths', size=13)
ax2.set_ylabel('Number', size=13)
# ax2.set_title(' Row-wise area')
ax2.set_xticks(x + width * i / 2)
ax2.set_xticklabels(x)
ax2.legend()
r_stat_sum = np.sum(y) # 总共多少个行缺失
r_stat_sum_l.append(r_stat_sum)
save_name = save_path + '/Row-wise-missing-data-stat'
plt.savefig(save_name + '.eps')
plt.savefig(save_name + '.svg')
# 列缺失
fig3, ax3 = plt.subplots()
for i, dataset_name, c_stat in zip(range(len(dataset_name_l)),
dataset_name_l, c_stat_l):
# bar graphs
# y = [v for v in c_stat.values()]
# x = [k for k in c_stat.keys()]
x_split = ['2', '3', '4', '5', '6-15', '16-30', '>30']
y_split = [0 for _ in range(len(x_split))]
# 对列缺失进行段落划分，否则统计显示很不方便
c_missing_num = 0
for x_, y_ in zip(c_stat.keys(), c_stat.values()):
c_missing_num += x_ * y_
if x_ == 2:
y_split[0] += y_
if x_ == 3:
y_split[1] += y_
if x_ == 4:
y_split[2] += y_
if x_ == 5:
y_split[3] += y_
if 5 < x_ <= 15:
y_split[4] += y_
if 15 < x_ <= 30:
y_split[5] += y_
if 30 < x_:
y_split[6] += y_
y = np.array(y_split)
x = np.array(np.arange(len(x_split)))
width = 0.25
ax3.bar(x + width * i, y, width, label=dataset_name)
ax3.tick_params(labelsize=13)
ax3.set_xlabel('Gap lengths', size=13)
ax3.set_ylabel('Number', size=13)
# ax3.set_title(' Column-wise area')
ax3.set_xticks(x + width * i / 2)
ax3.set_xticklabels(x_split)
plt.xticks(rotation=-15) # 设置x轴标签旋转角度
plt.tick_params(axis='x', labelsize=10) # 设置x轴标签大小
ax3.legend()
c_stat_sum = np.sum(y) # 总共多少个列缺失
c_stat_sum_l.append(c_stat_sum)
c_mp_num_l.append(c_missing_num)
save_name = save_path + '/Column-wise-missing-data-stat'
plt.savefig(save_name + '.eps')
plt.savefig(save_name + '.svg')
# 行和列缺失的总的个数
fig4, ax4 = plt.subplots()
for i, dataset_name, r_stat_sum, c_stat_sum in zip(
range(len(dataset_name_l)), dataset_name_l, r_stat_sum_l,
c_stat_sum_l):
y = [r_stat_sum, c_stat_sum]
x_ticklable = ['row', 'column']
x = np.arange(len(x_ticklable))
width = 0.25
ax4.bar(x + width * i, y, width, label=dataset_name)
ax4.tick_params(labelsize=13)
ax4.set_ylabel('Number', size=13)
# ax4.set_title('Numbers of missing row and column')
ax4.set_xticks(x + width * i / 2)
ax4.set_xticklabels(x_ticklable)
ax4.legend()
save_name = save_path + '/Sum-missing-data-stat'
plt.savefig(save_name + '.eps')
plt.savefig(save_name + '.svg')
# 行和列缺失的总的占数据集的比例
fig5, ax5 = plt.subplots()
for i, dataset_name, r_stat_sum, c_missing_num, data_len in zip(
range(len(dataset_name_l)), dataset_name_l, r_stat_sum_l,
c_mp_num_l, data_len_l):
y = [
r_stat_sum / data_len * 100,
c_missing_num / (data_len * len(select_dim)) * 100
]
x_ticklable = ['row', 'column']
x = np.arange(len(x_ticklable))
width = 0.25
ax5.plot(x_ticklable, y, label=dataset_name)
# 保存到本地json
save2json(save_path + dataset_name + '_rc.json', y, x_ticklable)
ax5.tick_params(labelsize=13)
ax5.set_ylabel('Ratio(%)', size=13)
# ax5.set_title('Ratio of missing data in row and column gaps')
# ax5.set_xticks(x + width * i / 2)
# ax5.set_xticklabels(x_ticklable)
ax5.legend()
save_name = save_path + '/Ratio-of-missing-data-stat_for_r&c'
plt.savefig(save_name + '.eps')
plt.savefig(save_name + '.svg')
# dataset_number = '30'
# exp_name = 'EM_6_dn({})'.format(dataset_number)
for param in params_test_list:
print('** {} params test: {} **'.format(test_param_name, param))
# dataset_number = 'one_mi_v1((A{})_1r)'.format(param)
dataset_number = 'one_mi((A5_A20_A30)_111)'
exp_name = 'EM_{}_lastest_1'.format(dataset_number)
result_save_file = './{}/'.format(results_saved_file) + exp_name + '/'
# 用于统计各种指标，建立相对应的文件夹
for ind in indicator_list:
test_result_save_path = result_save_file + ind + '/'
mkdir(test_result_save_path)
for i in range(cross_validation_sets_num):
print('============= Start training at datasets {} =============='.
format(i))
# 用于统计各种指标，建立相对应的文件夹
# result_save_file = './{}/'.format(results_saved_file) + exp_name + '/datasets_{}/'.format(i)
# plots_file = './plot_results/' + exp_name + '/datasets_{}/'.format(i)
# 当前数据集
args.dataset_path = './constructed_datasets/{}/{}/'.format(
dataset_number, i)
# 载入数据
station_datasets = get_saved_datasets(args)
rmse_on_test_results = []
'G_hidden_n': args.G_hidden_dim,
'D_hidden_n': args.D_hidden_dim,
'activate_function': 'ReLU',
'optimizer': 'Adam',
'idpt_d_lr': args.d_lr,
'idpt_g_lr': args.g_lr,
'lr_decay': args.g_lr_decay,
'decay_step': args.g_lr_decay_step,
'clip_value': args.clip_value,
'p_hint': args.p_hint,
'alpha': args.alpha
}
# 存储主文件路径
result_save_main_file = './{}/'.format(results_saved_file) + exp_name + '/'
mkdir(result_save_main_file)
# 保存参数配置
params_save_name = result_save_main_file + 'params_settings.json'
with open(params_save_name, 'w+') as jsonf:
json.dump(ex_params_settings, jsonf)
for i in range(cross_validation_sets):
print('============= Start training at datasets {} =============='.format(i))
# 用于统计各种指标，建立相对应的文件夹
result_save_file = './{}/'.format(results_saved_file) + exp_name + '/datasets_{}/'.format(i)
for index in indicator_list:
for model_name in model_name_list:
test_result_save_path = result_save_file + index + '/' + model_name
mkdir(test_result_save_path)
'optimizer': 'Adam',
'fed_d_lr': args.fed_d_lr,
'fed_g_lr': args.fed_g_lr,
'idpt_d_lr': args.d_lr,
'idpt_g_lr': args.g_lr,
'lr_decay': args.g_lr_decay,
'decay_step': args.g_lr_decay_step,
'clip_value': args.clip_value,
'p_hint': args.p_hint,
'alpha': args.alpha
}
# 存储主文件路径
result_save_main_file = './{}/'.format(
results_saved_file) + exp_name + '/'
mkdir(result_save_main_file)
# 保存参数配置
params_save_name = result_save_main_file + 'params_settings.json'
with open(params_save_name, 'w+') as jsonf:
json.dump(ex_params_settings, jsonf)
for i in range(cross_validation_sets):
print(
'\033[1;32m ============= Start training at datasets {} ==============\033[0m'
.format(i))
# 用于统计各种指标，建立相对应的文件夹
result_save_file = './{}/'.format(
results_saved_file) + exp_name + '/datasets_{}/'.format(i)
for index in indicator_list:
def run_multi_cross_validation_datasets_test():
"""
用于计算每个数据集的学习训练测试结果的总学习测试结果的平均
:return:
"""
args = args_parser()
# 做实验
exp_total_time = 10
cross_validation_sets = 5
exp_name = 'FedWeightAvg(soft)_6_dn(15)_100_32(0.001_0.001_bs_128)_32(phint_0.95)'
indicator_list = ['rmse', 'd2', 'r2', 'all_rmse']
# for i in range(cross_validation_sets):
# print('============= Start training at datasets {} =============='.format(i))
# # 用于统计各种指标，建立相对应的文件夹
# result_save_file = './results_v2/' + exp_name + '/datasets_{}/'.format(i)
# plots_file = './plot_results/' + exp_name + '/datasets_{}/'.format(i)
# for exp_t in range(exp_total_time):
# # 当前数据集
# args.dataset_path = './constructed_datasets_6_dn(5)/{}/'.format(i)
# fed_gain_test_exp(args, result_save_file, i)
# print('>>> Finished training & testing!')
result_save_root = './results_v3/' + exp_name + '/'
plots_save_root = './plot_results/' + exp_name + '/'
indicator_name = 'all_rmse'
leg = ['Fed', 'Independent']
'''
# 建立保存结果的文件夹
indicator_results_csv_save_fpth = result_save_root + indicator_name + '/'
for mode in leg:
mkdir(indicator_results_csv_save_fpth+mode+'/')
# 计算每个数据集的几次实验的均值
for c in range(cross_validation_sets):
results_logdir = [result_save_root + 'datasets_' + str(c) + '/' + indicator_name + '/' + model_name + '/'
for model_name in ['fed', 'idpt']]
compute_avg_of_data_in_file(args, c, results_logdir, indicator_results_csv_save_fpth,
indicator_name)
'''
# 绘制测试结果图像
print(
'====================== Save every component indicator result ========================'
)
print('{} results'.format(indicator_name))
results_logdir = [
result_save_root + indicator_name + '/' + model_name + '/'
for model_name in leg
]
fig_save_path = plots_save_root + indicator_name + '/'
csv_save_fpth = result_save_root + indicator_name + '/'
mkdir(fig_save_path)
# 计算几次实验的最后的平均值并画出均值和方差曲线图
plot_indicator_avg_results_m(results_logdir, fig_save_path, 'station',
indicator_name, csv_save_fpth)
# plot_indicator_results(results_logdir, fig_save_path, indicator_name)
print(">>> Finished save resluts figures!")
def plot_all_algorithm_indicator_results(args,
indicator_list,
results_file_saved_path,
fig_save_file='',
leg=None):
"""
用于绘制所有算法的指标结果图
:param args:
:param indicator_list:
:param results_file_saved_path:
:param fig_save_file:
:param leg:
:return:
"""
data = None
if leg is None:
leg = ['Fed', 'Independent']
# 用于图像保存
if fig_save_file == '':
fig_save_fpath = results_file_saved_path + 'All_indicator_avg_results/'
else:
fig_save_fpath = fig_save_file + 'All_indicator_avg_results/'
csv_save_fpath = results_file_saved_path + 'All_indicator_avg_results/'
mkdir(fig_save_fpath) # 新建文件夹用于保存图片
mkdir(csv_save_fpath) # 新建文件夹用于保存数据
# 绘制测试结果图像
print(
'====================== Save every component indicator result ========================'
)
for indicator in indicator_list:
print('{} results'.format(indicator))
if indicator != 'all_rmse':
_results_logdir = [
results_file_saved_path + indicator + '/' + model_name + '/'
for model_name in ['fed', 'idpt', 'em']
]
datas = get_all_datasets(_results_logdir, leg)
if isinstance(datas, list):
data = pd.concat(datas)
unit_sets = data['Unit'].values.tolist()
unit_sets = set(unit_sets)
indicator_avg_list = []
for mode in leg:
avg_t = 0
for u in unit_sets:
fed_avg_data = data[data.Condition == mode]
fed_avg_data = fed_avg_data[fed_avg_data.Unit == u][
args.select_dim].values
avg_t += fed_avg_data
indicator_avg = avg_t / len(unit_sets)
indicator_avg_list.append(indicator_avg)
# 保存到本地
fed_save_csv_pt = csv_save_fpath + mode + '_' + indicator + '_avg_resluts.csv'
save_all_avg_results(fed_save_csv_pt, indicator_avg,
args.select_dim, args.selected_stations)
print('***** ' + mode + ' avg: ', indicator_avg)
c = len(args.select_dim)
r = len(args.selected_stations)
x = [h for h in range(r)]
# plot style
fig, axs = plt.subplots(2, 3, constrained_layout=True)
for i in range(c):
for j in range(len(indicator_avg_list)):
axs[i // 3, i % 3].plot(x,
indicator_avg_list[j][:, i],
label=leg[j])
axs[i // 3, i % 3].set_xlabel('Station')
axs[i // 3, i % 3].set_ylabel(indicator)
axs[i // 3, i % 3].set_title(args.select_dim[i])
# if indicator == 'rmse':
# axs[i // 3, i % 3].set_ylim(0, 0.1)
# else:
# axs[i // 3, i % 3].set_ylim(0, 1.0)
# axs[i // 3, i % 3].legend(loc='upper right', fontsize=8)
plt.legend(loc='upper right', fontsize=8)
save_path = fig_save_fpath + indicator + '_avg_resluts.svg'
plt.savefig(save_path)
# plot_indicator_results(results_logdir, fig_save_path, component)
print('')
elif indicator == 'all_rmse':
# 计算all_rmse值，并保存到表格
_results_logdir = [results_file_saved_path + 'all_rmse/']
plot_indicator_results(_results_logdir,
fig_save_fpath,
'all_rmse',
xaxis='station',
leg=leg)
plt.clf()
plt.close()
return True
# dataset_number = 'one_mi((A5_B10_E15)_111)'
dataset_name = '(A5_A10_A15)_nCO_532r_One_time'
# dataset_name = '(1P10_2P20_3P30)_532r_One_time'
for param in params_test_list:
print('** {} params test: {} **'.format(test_param_name, param))
if training_model == 'Many_time':
dataset_name = Dname_prefix.format(param, param,
param) + '_' + Dname_suffix
# dataset_name = 'one_mi_v1((A{})_1r_v3)'.format(param)
exp_name = 'C_Test_{}_FedWGAI_T1'.format(dataset_name)
# 存储主文件路径
result_save_main_file = './{}/'.format(
results_saved_file) + exp_name + '/'
mkdir(result_save_main_file)
for i in range(cross_validation_sets):
print('============= Start training at datasets {} =============='.
format(i))
# 用于统计各种指标，建立相对应的文件夹
result_save_file = './{}/'.format(
results_saved_file) + exp_name + '/datasets_{}/'.format(i)
for index in indicator_list:
for model_name in ['fed', 'idpt']:
test_result_save_path = result_save_file + index + '/' + model_name
mkdir(test_result_save_path)
for exp_t in range(exp_total_time):
# 当前数据集
format(i))
args.dataset_path = './constructed_datasets/{}/{}/'.format(
dataset_name, i)
station_datasets = get_saved_datasets(args)
# 用于统计各种指标，建立相对应的文件夹
result_save_file = './{}/'.format(
results_saved_file) + exp_name + '/datasets_{}/'.format(i)
# result_save_file = './{}/'.format(results_saved_file) + exp_name + '/'
# 用于统计各种指标，建立相对应的文件夹
for index in indicator_list:
test_result_save_path = result_save_file + index + '/'
mkdir(test_result_save_path)
for exp_t in range(exp_total_time):
all_rmse_on_test_results = [] # 用于存储每个数据集测试结果
# 对每个数据集的每个站点数据进行计算
for dataset, station in zip(station_datasets,
args.selected_stations):
# r = Low_rank_completion(dataset, l=1, mu=0.02, initial=0) # LRC method
# Linear, Spline, Cubic, MultiBayesian, RandomTrees
if method_name == 'Linear':
a_rmse = pandas_linear_methods(dataset) # Linear
elif method_name == 'Spline':
a_rmse = pandas_spline_method(dataset) # Spline
elif method_name == 'KNN':
a_rmse = pandas_nearest_method(dataset) # KNN
elif method_name == 'Cubic':
