示例#1
0
 def show_bias(self):
     if len(self.weights) is 1:
         mp = MyPlot()
         mp.set_labels('Step', 'Bias')
         mp.show_list(self.biases)
     else:
         print('Cannot show the bias! Call print_bias mehtod.')
示例#2
0
    def show_weight(self):
        print('shape=', self.weights)

        if len(self.weights[0]) is 1:
            mp = MyPlot()
            mp.set_labels('Step', 'Weight')
            mp.show_list(self.weights)
        else:
            print('Cannot show the weight! Call print_weight method.')
示例#3
0
    def __init__(self, *args, **kwargs):
        super(Wnd, self).__init__(*args, **kwargs)

        self.t = None
        self.osc = None

        self.input_cbox = None
        self.output_cbox = None

        left = VContainer()
        right = VContainer()

        self.slider = QtWidgets.QSlider(Qt.Horizontal)
        self.slider.setMinimum(20)
        self.slider.setMaximum(500)
        left.addWidget(self.slider)
        self.slider.valueChanged.connect(self.change_freq)

        self.addWidget(left)
        self.addWidget(right)

        devices = HContainer()
        devices.addWidget(self.inputCBox())
        devices.addWidget(self.outputCBox())
        left.addWidget(devices)

        spf = wave.open("wav.wav", "r")
        signal = spf.readframes(-1)
        signal = np.frombuffer(signal, "int16")
        self.signal = signal

        self.graphWidget = MyPlot()
        left.addWidget(self.graphWidget)

        self.fft = Spec()
        # self.fft.setYRange(0, 100000, padding=0)
        left.addWidget(self.fft)

        self.buttons = HContainer()

        btn = QtWidgets.QPushButton()
        btn.setText('Record')
        self.buttons.addWidget(btn)
        btn.clicked.connect(self.get_device_num)

        btn = QtWidgets.QPushButton()
        btn.setText('Play')
        self.buttons.addWidget(btn)
        btn.clicked.connect(self.play_file)

        left.addWidget(self.buttons)

        self.logger = LogWrite()
        right.addWidget(self.logger)
示例#4
0
#----- a neuron
w = tf.Variable(tf.random_normal([2, 1]))
b = tf.Variable(tf.random_normal([1]))
hypo = tf.matmul(x, w) + b
#-----

cost = tf.reduce_mean((hypo - y) * (hypo - y))

train = tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(cost)

sess = tf.Session()
sess.run(tf.global_variables_initializer())

costs = []

for i in range(2001):
    sess.run(train)

    if i % 50 == 0:
        print('hypo:', sess.run(hypo), '|', sess.run(w), sess.run(b),
              sess.run(cost))

        costs.append(sess.run(cost))

hypo2 = tf.matmul([[4., 4]], w) + b
print(sess.run(hypo2))

p = MyPlot()
p.show_list(costs)
示例#5
0
 def show_error(self):
     mp = MyPlot()
     mp.set_labels('Step', 'Error')
     mp.show_list(self.costs)
class LineParser(object):
    __rules = [
        'CREATEVAR',
        'SETVAR',
        'CALCULATE',
        'CREATEPLOT',
        'ADDTOPLOT',
        'SHOWPLOT',
        'CREATEFCT',
        'ENDFCT',
        'CALL']
    __function_create_start = None
    __variables = Variable()
    __plots = MyPlot()
    __functions = Function()

    def __init__(self, line):
        self.line = line
        self.error = 0
        self.rule_in_line = ''

    def update_globals(self, globals):
        pass
    def get_rule(self):
        for __rule in LineParser.__rules:
            __location = -1
            __matching_rules = 0
            while __matching_rules <=2: #search multiple identical rules in the same line
                __location = self.line.find(__rule, __location + 1)
                if __location == -1:
                    break
                else:
                    __matching_rules += 1
                    if __matching_rules == 1:
                        self.rule_in_line = __rule
                        self.error += 1
                    else:
                        self.error += 1
    def implement_rule(self):
        if LineParser.__function_create_start == None:
            if self.rule_in_line == 'CREATEVAR':
                LineParser.__variables.CreateVar((self.line.split(':')[1]).split('\n')[0])
                MyPlot.UpdateGlobals(Variable.variables)
            elif self.rule_in_line == 'SETVAR':
                LineParser.__variables.SetVar(self.line.split(':')[1], (self.line.split(':')[2]).split('\n')[0])
                MyPlot.UpdateGlobals(Variable.variables)
            elif self.rule_in_line == 'CALCULATE':
                LineParser.__variables.Calculate(self.line.split(':')[1], (self.line.split(':')[2]).split('\n')[0])
                MyPlot.UpdateGlobals(Variable.variables)
            elif self.rule_in_line == 'CREATEPLOT':
                LineParser.__plots.CreatePlot((self.line.split(':')[1]).split('\n')[0])
            elif self.rule_in_line == 'ADDTOPLOT':
                LineParser.__plots.AddToPlot(self.line.split(':')[1], self.line.split(':')[2], (self.line.split(':')[3]).split('\n')[0])
            elif self.rule_in_line == 'SHOWPLOT':
                LineParser.__plots.ShowPlot((self.line.split(':')[1]).split('\n')[0])
            elif self.rule_in_line == 'CREATEFCT':
                LineParser.__function_create_start = (self.line.split(':')[1]).split('\n')[0]
                LineParser.__functions.CreateFunction((self.line.split(':')[1]).split('\n')[0])
            elif self.rule_in_line == 'CALL':
                LineParser.__functions.CallFunction((self.line.split(':')[1]).split('\n')[0])
            else:
                pass
        else:
            if self.rule_in_line == 'CREATEVAR':
                LineParser.__functions.AddToFunction(LineParser.__function_create_start, self.line)
            elif self.rule_in_line == 'SETVAR':
                LineParser.__functions.AddToFunction(LineParser.__function_create_start, self.line)
            elif self.rule_in_line == 'CALCULATE':
                LineParser.__functions.AddToFunction(LineParser.__function_create_start, self.line)
            elif self.rule_in_line == 'CREATEPLOT':
                LineParser.__functions.AddToFunction(LineParser.__function_create_start, self.line)
            elif self.rule_in_line == 'ADDTOPLOT':
                LineParser.__functions = Function()
                LineParser.__functions.AddToFunction(LineParser.__function_create_start, self.line)
            elif self.rule_in_line == 'SHOWPLOT':
                LineParser.__functions.AddToFunction(LineParser.__function_create_start, self.line)
            elif self.rule_in_line == 'ENDFCT':
                LineParser.__function_create_start = None
示例#7
0
from myplot import MyPlot
import sys

data = [[1, 2, 3, 4, 5], [6, 7, 8, 9, 10]]
print(data)

print('Number of arguments:', len(sys.argv), 'arguments.')
print('Argument List:', str(sys.argv))

#type_of_graph = int(input("What is the type of graph you want. \n Press 1 for histogram \n Press 2 for bar chart \n Press 3 for scatter plot. \n Press 4 for line plot \n Press 5 for histogram\n Press 6 for pie chart"))
#{'b', 'g', 'r', 'c', 'm', 'y', 'k', 'w'}

type_of_graph = int(sys.argv[1])
myplt = MyPlot(type_of_graph, str(sys.argv[2]), data)
myplt.plot_graph()
#fig, axs = plt.subplots(2, 2, figsize=(5, 5))
#axs[0, 0].hist(data[0])
#axs[1, 0].scatter(data[0], data[1])
#axs[0, 1].plot(data[0], data[1])
#axs[1, 1].hist2d(data[0], data[1])
示例#8
0
        self.addPoint(state.PosAct)


# A worker class for the ros node
# To be deprecated soon
class WorkerListener(Thread):
    def __init__(self):
        Thread.__init__(self)

    def run(self):
        rospy.ready(NAME, anonymous=True)
        rospy.spin()


channel = MyChannel()
channel2 = MyChannelAct()
rospy.TopicSub('ARM_L_PAN_state', RotaryJointState, channel.callback)
rospy.TopicSub('ARM_L_PAN_state', RotaryJointState, channel2.callback)
app = wx.PySimpleApp(0)
worker = WorkerListener()
worker.start()
frame = wx.Frame(None, -1, "")
panel = MyPlot(frame)
panel.addChannel(channel)
panel.addChannel(channel2)
frame.Show()
print '>>>>>>>>>>>>>>'
app.MainLoop()
print '>>>>>>>>>>>>>>'
#rospy.spin()
示例#9
0
from myplot import MyPlot

x_data = [1]
y_data = [1]

#----- a neuron
w = tf.Variable(tf.random_normal([1]))
b = tf.Variable(tf.random_normal([1]))
hypo = w * x_data + b
#-----

cost = (hypo - y_data) ** 2

train = tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(cost)

sess = tf.Session()
sess.run(tf.global_variables_initializer())

costs = []

for i in range(1001):
    sess.run(train)

    if i % 50 == 0:
        print(sess.run(w), sess.run(b), sess.run(cost))
        costs.append(sess.run(cost))

gildong = MyPlot()
gildong.show_list(costs)