def __init__(self, parent=None):
super(Main, self).__init__(parent)
QMainWindow.__init__(self)
loadUi('core/window.ui', self)
#self.setupUi(self)
self.setWindowIcon(QIcon('main.ico'))
'''pushButton等控件信号'''
self.time_point_training.clicked.connect(self.clickedTime_point_training)
self.pushButton_pause.clicked.connect(self.clickedPause)
#self.dynamic_display.clicked.connect(self.dynamic_display_click)
self.change_dynamic.clicked.connect(self.change_dynamic_click)
self.change_hot.clicked.connect(self.change_hot_click)
self.commandLink_about.clicked.connect(self.about)
self.time_point.currentIndexChanged.connect(self.currentIndexChanged)
'''newForm'''
self.new_dig.clicked.connect(self.test)
#self.close_signal.connect(self.newForm.close)
'''widget_plot，数据可视化'''
self.main_widget = QWidget(self)
self.widget_plot = matplot.MyccMplCanvas(self.main_widget, width=5, height=4, dpi=100)
self.verticalLayout_cell.addWidget(self.widget_plot)
GlobalMap().set(running_global = 0)
self.initial_value() #初始化全局变量
self.timer = QTimer(self) #动态更新全局变量
self.timer.timeout.connect(self.update_value)
self.global_temp= GlobalMap().get('all')
def loss(self):
#绘制损失度动态图
self.axes.hold(False)
self.a += [GlobalMap().get('times')]
self.b += [GlobalMap().get('loss_now') / 100]
self.axes.plot(self.a, self.b, linewidth=2)
self.axes.set(xlabel="times",
ylabel="loss",
title='Real time training state')
self.draw()
def save_question(self):
'''询问是否使用历史训练数据，提高速度'''
reply = QMessageBox.question(self, '已有历史训练数据',
"你还要重新训练吗?", QMessageBox.Yes |
QMessageBox.No, QMessageBox.No)
if reply == QMessageBox.Yes:
GlobalMap().set(
save_nopass = 0
)
GlobalMap().set(
save_question = 0
)
def update_value(self):
'''更新UI状态显示'''
self.global_temp= GlobalMap().get('all')
try:
if not GlobalMap().get('running_global'):
time.sleep(0.3)
self.timer.stop()
self.initial_value()
#print(self.global_temp)
self.progress_num.setValue(self.global_temp['progress_num'])
self.iter_times.setText(str(self.global_temp['times']))
self.loss_now.setText(str(self.global_temp['loss_now']))
self.used_time.display(self.global_temp['used_time'])
except:
pass
def dig_change_accept(self):
'''23小时各小时变化动态图'''
if GlobalMap().get('running_global'):
QMessageBox.warning(self, "Warning","Is_running!!!\nMaybe you want to click '停止当前任务'")
return
self.initial_value()
GlobalMap().set(running_global = 1, running_status = 2, time_point = self.time_point.currentIndex(), fix_seed = self.fix_seed.isChecked())
self.update_value()
self.timer.start(1000)
self.verticalLayout_cell.removeWidget(self.widget_plot)
self.widget_plot = matplot.MyDig_ChangeMplCanvas(self.main_widget, width=5, height=4, dpi=100)
self.verticalLayout_cell.addWidget(self.widget_plot)
self.widget_plot.start()
def start(self):
self.times = 1
GlobalMap().set(progress_num=0)
self.lists = [] #记录时间
self.list_height = [[], []] #用于记录高密度的区域
self.stat = 0
self.fig.subplots_adjust(left=0.02,
bottom=0.02,
right=0.98,
top=0.92,
hspace=0.04,
wspace=0.02)
self.axes1 = self.fig.add_subplot(211)
self.axes2 = self.fig.add_subplot(212)
for i in range(24):
self.lists += [i]
if exists('db/temp/20180515.csv'):
self.temp1 = DataFrame(read_csv('db/temp/20180515.csv', header=0))
else:
self.temp1 = dataclear(20180515).run()
self.compute_initial_figure()
self.timer = QTimer(self)
self.timer.timeout.connect(self.compute_initial_figure)
self.timer.start(1000)
def pause(self):
try:
self.timer.stop()
self.thread.quit()
except:
pass
GlobalMap().set(running_global=0, )
def clickedTime_point_training(self):
'''点击单时间的预测'''
self.initial_value()
if GlobalMap().get('running_global'):
QMessageBox.warning(self, "Warning","Is_running!!!\nMaybe you want to click '停止当前任务'")
return
GlobalMap().set(running_global = 1, running_status = 0, time_point = self.time_point.currentIndex(), fix_seed = self.fix_seed.isChecked())
self.update_value()
if exists('ckpt/model-'+ str(self.global_temp['time_point']) +'.ckpt.index') and GlobalMap().get('save_question'): #判断模型是否存在
self.save_question()
self.timer.start(600)
self.verticalLayout_cell.removeWidget(self.widget_plot)
self.widget_plot = matplot.MySticMplCanvas(self.main_widget, width=5, height=4, dpi=100)
self.verticalLayout_cell.addWidget(self.widget_plot)
self.widget_plot.start()
def update_figure(self):
if GlobalMap().get('userdefined') == 'Null_': #检测从自定义对话框传递过来的数据
return
if not self.begin: #防止重绘
return
self.begin = 0
GlobalMap().set(time_point=self.times, )
self.times += 1
if self.times == 2:
self.begin = 1
self.temp0 = NN.main([], [],
GlobalMap().get('userdefined')) #传入BP网络预测
return
else:
temp01 = self.temp0
self.temp0 = NN.main([], [], temp01)
imgP = imread('core/map.png', )
self.axes.imshow(imgP,
extent=[112.1495, 112.288, 30.308, 30.391],
aspect='auto')
x = np.array(self.temp0[0])
y = np.array(self.temp0[1])
z = gaussian_kde(self.temp0)(self.temp0) #生成密度颜色
self.axes.hold(True)
self.axes.scatter(x, y, marker='.', s=100, c=z, edgecolors='')
self.axes.grid(True)
self.axes.set_xlim(112.150, 112.284)
self.axes.set_ylim(30.309, 30.385)
self.axes.set(xlabel="longitude", ylabel="latitude")
self.axes.set_title('The ' + str(self.times) +
':00 ’s distribution prediction')
self.draw()
self.begin = 1
self.axes.hold(False)
if self.times >= 2:
self.pause()
def test(self, a, b, c):
'''入口'''
if a == []: #仅用于自定义模式的预测
#print(self.ckpt_path, c)
self.setup([2, self.init_temp['neurons_num'], 2])
self.saver.Saver().restore(self.session, 'ckpt/modelc-' +
str(self.init_temp['time_point']) +
'.ckpt') #存在就从模型中恢复变量
GlobalMap().set(used_time=0, times=0)
return noturn(self.predict(turn(c)))
x_data = turn(a)
y_data = turn(b)
test_data = turn(c)
self.setup([2, self.init_temp['neurons_num'], 2])
if exists(self.ckpt_path +
'.index') and GlobalMap().get('save_nopass'): #判断模型是否存在
try:
self.saver.Saver().restore(self.session,
self.ckpt_path) #存在就从模型中恢复变量
self.loss = reduce_mean(
reduce_sum(square((self.label_layer - self.output_layer)),
reduction_indices=[1]))
loss_now = self.session.run(self.loss,
feed_dict={
self.input_layer: x_data,
self.label_layer: y_data
})
GlobalMap().set(used_time=0, times=0, loss_now=loss_now)
except:
print(2)
self.train(x_data, y_data)
else:
self.train(x_data, y_data)
GlobalMap().set(running_global=0, )
return noturn(self.predict(test_data))
def start(self):
self.times = 1
GlobalMap().set(progress_num=0)
self.lists = []
self.changenum00 = []
self.changenum01 = []
for i in range(24):
self.lists += [i]
if exists('db/temp/20180515.csv'):
self.temp1 = DataFrame(read_csv('db/temp/20180515.csv', header=0))
else:
self.temp1 = dataclear(20180515).run()
self.compute_initial_figure()
self.timer = QTimer(self)
self.timer.timeout.connect(self.compute_initial_figure)
self.timer.start(1000)
def __init__(self, session):
self.session = session #记录session
self.saver = train #提供Saver服务
self.loss = None #记录误差
self.optimizer = None #优化器
self.input_n = 0 #输入维度
self.hidden_n = 0 #中间层个数
self.hidden_size = [] #中间层大小
self.output_n = 0 #输出维度
self.input_layer = None
self.hidden_layers = []
self.output_layer = None
self.label_layer = None
self.init_temp = GlobalMap().get('all') #获取所有前台传递过来的数据
self.auto_times = 0
#print(self.init_temp) #测试用
self.ckpt_path = 'ckpt/model-' + str(
self.init_temp['time_point']) + '.ckpt' #定义训练模型保存位置
if self.init_temp['fix_seed']:
set_random_seed(523456789)
def initial_value(self):
'''初始化全局变量'''
#print(temp)
GlobalMap().set(max_itera = self.max_itera.value(),
neurons_num = self.neurons_num.value(),
dg_initial = self.dg_initial.value(),
dg_adaptive = self.dg_adaptive.value(),
precision = self.precision.value(),
save_question = 1,
save_nopass = 1,
temp_path = 'temp',
fix_auto = self.fix_auto.isChecked()
)
for i in range(24):
if exists('ckpt/model-'+ str(i) +'.ckpt.index'):
'''每个小时点的训练状况'''
{0:self.checkBox_0.setChecked,
1:self.checkBox_1.setChecked,
2:self.checkBox_2.setChecked,
3:self.checkBox_3.setChecked,
4:self.checkBox_4.setChecked,
5:self.checkBox_5.setChecked,
6:self.checkBox_6.setChecked,
7:self.checkBox_7.setChecked,
8:self.checkBox_8.setChecked,
9:self.checkBox_9.setChecked,
10:self.checkBox_10.setChecked,
11:self.checkBox_11.setChecked,
12:self.checkBox_12.setChecked,
13:self.checkBox_13.setChecked,
14:self.checkBox_14.setChecked,
15:self.checkBox_15.setChecked,
16:self.checkBox_16.setChecked,
17:self.checkBox_17.setChecked,
18:self.checkBox_18.setChecked,
19:self.checkBox_19.setChecked,
20:self.checkBox_20.setChecked,
21:self.checkBox_21.setChecked,
22:self.checkBox_22.setChecked,
23:self.checkBox_23.setChecked,
}.get(i)(True)
def compute_initial_figure(self, a, b):
self.timer.stop()
self.axes.hold(True)
self.axes.set_title('The ' + str(GlobalMap().get('time_point')) +
':00 ’s distribution prediction')
x = a[0]
y = a[1]
x0 = b[0]
y0 = b[1]
imgP = imread('core/map.png', )
self.axes.imshow(imgP,
extent=[112.1495, 112.288, 30.308, 30.391],
aspect='auto')
self.axes.plot(x, y, '.b', markersize=3, label='prediction') #预测的结果
self.axes.plot(x0, y0, '.r', markersize=2, label='original') #期望的结果
self.axes.grid(True)
self.axes.legend(shadow=True)
self.axes.set_xlim(112.150, 112.284)
self.axes.set_ylim(30.309, 30.385)
self.axes.set(xlabel="longitude", ylabel="latitude")
self.draw()
def run(self):
if self.working == True:
times = GlobalMap().get('time_point')
GlobalMap().set(progress_num=0)
lists = []
for i in range(24):
lists += [i]
GlobalMap().set(progress_num=2)
if exists('db/temp/20180515.csv') and exists(
'db/temp/20180516.csv'):
temp1 = DataFrame(read_csv('db/temp/20180515.csv', header=0))
temp2 = DataFrame(read_csv('db/temp/20180516.csv', header=0))
else:
temp1 = dataclear(20180515).run()
temp2 = dataclear(20180516).run()
GlobalMap().set(progress_num=5)
temp11 = temp1[temp1['time'] == lists[times - 1]]
temp12 = temp1[temp1['time'] == lists[times]]
temp1112 = merge(temp11, temp12, how='outer', on='id').dropna()
temp21 = temp2[temp2['time'] == lists[times - 1]]
temp22 = temp2[temp2['time'] == lists[times]]
temp2122 = merge(temp21, temp22, how='outer', on='id').dropna()
temp000 = np.vstack((temp1112['x_x'], temp1112['y_x']))
temp001 = np.vstack((temp1112['x_y'], temp1112['y_y']))
temp020 = np.vstack((temp2122['x_x'], temp2122['y_x']))
#temp021 = np.vstack((temp2122['x_y'],temp2122['y_y']))
cells_array = np.array(NN.main(temp000, temp001, temp020))
np.save(GlobalMap().get('temp_path') + '/test' + str(times),
cells_array)
GlobalMap().set(progress_num=100, running_global=0)
self.changed2_signal.emit(1)
def run(self):
if self.working == True:
times = GlobalMap().get('time_point')
GlobalMap().set(progress_num=0)
lists = []
for i in range(24):
lists += [i]
GlobalMap().set(progress_num=2)
if exists('db/temp/20180515.csv') and exists(
'db/temp/20180516.csv'):
temp1 = DataFrame(read_csv('db/temp/20180515.csv', header=0))
temp2 = DataFrame(read_csv('db/temp/20180516.csv', header=0))
else:
temp1 = dataclear(20180515).run()
temp2 = dataclear(20180516).run()
GlobalMap().set(progress_num=5)
temp11 = temp1[temp1['time'] == lists[times - 1]]
temp12 = temp1[temp1['time'] == lists[times]]
temp1112 = merge(temp11, temp12, how='outer', on='id').dropna()
'''begin 另一种数据处理方式'''
# temp1_x = DataFrame(read_csv('db/temp/20180515-x.csv' ,header=0))
# temp1_y = DataFrame(read_csv('db/temp/20180515-y.csv' ,header=0))
# temp2_x = DataFrame(read_csv('db/temp/20180516-x.csv' ,header=0))
# temp2_y = DataFrame(read_csv('db/temp/20180516-y.csv' ,header=0))
#
# temp11_x = temp1_x[str(lists[times])]
# temp11_y = temp1_y[str(lists[times])]
# temp12_x = temp2_x[str(lists[times])]
# temp12_y = temp2_y[str(lists[times])]
#
# temp21_x = temp1_x[str(lists[times+1])]
# temp21_y = temp1_y[str(lists[times+1])]
# temp22_x = temp2_x[str(lists[times+1])]
# temp22_y = temp2_y[str(lists[times+1])]
#
# cc01 = np.vstack((temp11_x, temp11_y))
# cc02 = np.vstack((temp21_x, temp21_y))
# cc03 = np.vstack((temp12_x, temp12_y))
# cc04 = np.vstack((temp22_x, temp22_y))
'''end'''
temp21 = temp2[temp2['time'] == lists[times - 1]]
temp22 = temp2[temp2['time'] == lists[times]]
temp2122 = merge(temp21, temp22, how='outer', on='id').dropna()
temp000 = np.vstack((temp1112['x_x'], temp1112['y_x']))
temp001 = np.vstack((temp1112['x_y'], temp1112['y_y']))
temp020 = np.vstack((temp2122['x_x'], temp2122['y_x']))
temp021 = np.vstack((temp2122['x_y'], temp2122['y_y']))
#用20180515 T 和 T+1 时刻的训练 来预测 20180516 T 时刻 在 T+1 时刻的分布情况
cells_array = np.array(NN.main(temp000, temp001, temp020))
#cells_array = np.array(NN.main(cc01, cc02, cc03 ))
np.save(GlobalMap().get('temp_path') + '/test' + str(times),
cells_array)
GlobalMap().set(progress_num=100)
self.changed_signal.emit(cells_array, temp021)
def train(self, cases, labels):
'''训练'''
limit = self.init_temp['max_itera']
learn_rate = self.init_temp['dg_initial'] * 0.00000001
self.loss = reduce_mean(
reduce_sum(square((self.label_layer - self.output_layer)),
reduction_indices=[1]))
learning_rate = placeholder(float32, shape=[])
#self.train_step = #tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(self.loss)
self.train_step = train.MomentumOptimizer(learning_rate=learning_rate,
momentum=0.9,
use_locking=False,
use_nesterov=True).minimize(
self.loss) #动量优化器
self.session.run(global_variables_initializer())
next_done = 0
starttime = datetime.now()
for i in range(limit):
self.session.run(self.train_step,
feed_dict={
self.input_layer: cases,
self.label_layer: labels,
learning_rate: learn_rate
})
if i % 2 == 0:
if not GlobalMap().get('running_global'):
break
next_done = 1
next_times = i + 1
train_0 = self.train_step
loss0 = self.session.run(self.loss,
feed_dict={
self.input_layer: cases,
self.label_layer: labels
})
#更新前台和处理数据
if next_done == 1 and i % next_times == 0:
loss1 = self.session.run(self.loss,
feed_dict={
self.input_layer: cases,
self.label_layer: labels
})
GlobalMap().set(loss_now=loss0,
times=i,
used_time=(datetime.now() - starttime).seconds)
if loss1 < self.init_temp['precision']:
break
if (i + 1) % 50 == 0 and self.init_temp['fix_auto'] and (
loss0 - loss1 < 0.1 or isnan(loss1)):
self.auto_times += 1
#print(self.auto_times)
GlobalMap().set(progress_num=5 + i * 100 // limit +
10) #进度条
if self.auto_times > 8:
break
if loss1 < loss0: #在动量的基础上调整学习率
learn_rate *= 1.0 + self.init_temp['dg_adaptive']
else:
learn_rate *= 1.0 - self.init_temp['dg_adaptive']
self.train_step = train_0
#print("1:", loss1)
#print("shu:",learn_rate)
next_done = 0
self.saver.Saver().save(self.session, self.ckpt_path)
def pause(self):
GlobalMap().set(running_global=0, )
def accept(self):
self.mun = self.dialog_setting.spinBox.value()
self.cum[0]=append(self.cum[0], uniform(self.x_n0,self.x_n1,size=(1,self.mun)))
self.cum[1]=append(self.cum[1], uniform(self.y_n0,self.y_n1,size=(1,self.mun)))
self.accept_move += [[self.x0, self.y0, self.x1-self.x0, self.y1-self.y0, self.mun]]
GlobalMap().set(userdefined = self.cum, )
def pause(self):
self.timer.stop()
GlobalMap().set(running_global=0)
def compute_initial_figure(self):
if self.times <= 23:
temp11 = self.temp1[self.temp1['time'] == self.lists[self.times -
1]]
temp12 = self.temp1[self.temp1['time'] == self.lists[self.times]]
temp1112 = merge(temp11, temp12, how='outer', on='id')
temp1112 = temp1112[[
'x_x', 'y_x', 'x_y', 'y_y'
]][temp1112['x_x'] != temp1112['x_y']] #np.isnan(temp1112['y_y'])]
temp1112 = temp1112[[
'x_x', 'y_x', 'x_y', 'y_y'
]][np.square(temp1112['x_x'] - temp1112['x_y']) +
np.square(temp1112['y_x'] - temp1112['y_y']) >
np.square(0.0001)] #车至少移动了直线距离100
temp1112 = temp1112[['x_x', 'y_x', 'x_y',
'y_y']][np.isfinite(temp1112['x_x'])]
temp11120 = merge(temp11, temp12, how='outer', on='id')
temp11120 = temp11120[[
'x_x', 'y_x', 'x_y', 'y_y'
]][np.square(temp11120['x_x'] - temp11120['x_y']) +
np.square(temp1112['y_x'] - temp11120['y_y']) >
np.square(0.001)] #车至少移动了直线距离1km
temp11120 = temp11120[['x_x',
'y_x']][np.isfinite(temp11120['x_x'])]
self.changenum00 += [temp1112.shape[0]]
self.changenum01 += [temp11120.shape[0]]
imgP = imread('core/map.png', )
self.axes.imshow(imgP,
extent=[112.1495, 112.288, 30.308, 30.391],
aspect='auto')
x = temp1112['x_x']
y = temp1112['y_x']
np.save(
GlobalMap().get('temp_path') + '/change' + str(self.times) +
'.npy', np.vstack((x, y)))
self.axes.hold(True)
self.axes.scatter(x, y, marker='o', c='', edgecolors='g')
self.axes.grid(True)
self.axes.set_xlim(112.150, 112.284)
self.axes.set_ylim(30.309, 30.385)
self.axes.set(xlabel="longitude", ylabel="latitude")
self.axes.set_title('The ' + str(self.times) +
':00 ’s distribution statistics')
self.draw()
self.axes.hold(False)
if self.times >= 23:
if self.times == 23:
self.axes.bar(range(1, 24),
self.changenum00,
0.8,
color=['g', 'c'],
linewidth=0.1,
yerr=[
0.8,
] * 23,
error_kw=dict(elinewidth=1.5, ecolor='green'))
self.axes.hold(True)
self.axes.plot(range(1, 24),
self.changenum00,
marker="o",
linestyle="-",
color="r")
self.axes.set_xticks(range(1, 24))
self.axes.set(xlabel="hour", ylabel="number")
for x, y in zip(range(1, 24), self.changenum00):
self.axes.text(x, y + 0.5, '%d' % y, ha='center')
self.axes.set_title(
'Number of changes at each time point(At least 100m)')
self.draw()
self.axes.hold(False)
else:
self.pause()
sleep(1)
self.axes.bar(range(1, 24),
self.changenum01,
0.8,
color=['g', 'c'],
linewidth=0.1,
yerr=[
0.8,
] * 23,
error_kw=dict(elinewidth=1.5, ecolor='green'))
self.axes.hold(True)
self.axes.plot(range(1, 24),
self.changenum01,
marker="o",
linestyle="-",
color="r")
self.axes.set_xticks(range(1, 24))
self.axes.set(xlabel="hour", ylabel="number")
for x, y in zip(range(1, 24), self.changenum01):
self.axes.text(x, y + 0.5, '%d' % y, ha='center')
self.axes.set_title(
'Number of changes at each time point(At least 1km)')
self.draw()
self.axes.hold(False)
self.times += 1