【python】matplotlib动态显示

1.matplotlib动态绘图

python在绘图的时候，需要开启 interactive mode。核心代码如下：

    plt.ion(); #开启interactive mode 成功的关键函数
    fig = plt.figure(1);
    
    for i in range(100):
        filepath="E:/Model/weights-improvement-" + str(i + 1) + ".hdf5";
        model.load_weights(filepath);
        #测试数据
        x_new = np.linspace(low, up, 1000);
        y_new = getfit(model,x_new);
        # 显示数据
        plt.clf();
        plt.plot(x,y); 
        plt.scatter(x_sample, y_sample);
        plt.plot(x_new,y_new);
        
        ffpath = "E:/imgs/" + str(i) + ".jpg";
        plt.savefig(ffpath);

        plt.pause(0.01)             # 暂停0.01秒
        
    ani = animation.FuncAnimation(plt.figure(2), update,range(100),init_func=init, interval=500);
    ani.save("E:/test.gif",writer='pillow');
    
    plt.ioff()                 # 关闭交互模式

2.实例

已知下面采样自Sin函数的数据：

	x	y
1	0.093	-0.81
2	0.58	-0.45
3	1.04	-0.007
4	1.55	0.48
5	2.15	0.89
6	2.62	0.997
7	2.71	0.995
8	2.73	0.993
9	3.03	0.916
10	3.14	0.86
11	3.58	0.57
12	3.66	0.504
13	3.81	0.369
14	3.83	0.35
15	4.39	-0.199
16	4.44	-0.248
17	4.6	-0.399
18	5.39	-0.932
19	5.54	-0.975
20	5.76	-0.999

通过一个简单的三层神经网络训练一个Sin函数的拟合器，并可视化模型训练过程的拟合曲线。

2.1 网络训练实现 

主要做的事情是定义一个三层的神经网络，输入层节点数为1，隐藏层节点数为10，输出层节点数为1。

import math;
import random;
from matplotlib import pyplot as plt
from keras.models import Sequential
from keras.layers.core import Dense
from keras.optimizers import Adam
import numpy as np
from keras.callbacks import ModelCheckpoint
import os


#采样函数
def sample(low, up, num):
    data = [];
    for i in range(num):
        #采样
        tmp = random.uniform(low, up);
        data.append(tmp);
    data.sort();
    return data;

#sin函数
def func(x):
    y = [];
    for i in range(len(x)):
        tmp = math.sin(x[i] - math.pi/3);
        y.append(tmp);
    return y;

#获取模型拟合结果
def getfit(model,x):    
    y = [];
    for i in range(len(x)):
        tmp = model.predict([x[i]], 10);
        y.append(tmp[0][0]);
    return y;

#删除同一目录下的所有文件
def del_file(path):
    ls = os.listdir(path)
    for i in ls:
        c_path = os.path.join(path, i)
        if os.path.isdir(c_path):
            del_file(c_path)
        else:
            os.remove(c_path)

if __name__ == '__main__':    
    path = "E:/Model/";
    del_file(path);
    
    low = 0;
    up = 2 * math.pi;
    x = np.linspace(low, up, 1000);
    y = func(x);
    
    # 数据采样
#     x_sample = sample(low,up,20);
    x_sample = [0.09326442022999694, 0.5812590520508311, 1.040490143783586, 1.5504427746047338, 2.1589557183817036, 2.6235357787018407, 2.712578091093361, 2.7379109336528167, 3.0339662651841186, 3.147676812083248, 3.58596337171837, 3.6621496731124314, 3.81130899864203, 3.833092859928872, 4.396611340802901, 4.4481080339256875, 4.609657879057151, 5.399731063412583, 5.54299720786794, 5.764084730699906];
    y_sample = func(x_sample);
    
    # callback
    filepath="E:/Model/weights-improvement-{epoch:00d}.hdf5";
    checkpoint= ModelCheckpoint(filepath, verbose=1, save_best_only=False, mode='max');
    callbacks_list= [checkpoint];
    
    # 建立顺序神经网络层次模型
    model = Sequential();  
    model.add(Dense(10, input_dim=1, init='uniform', activation='relu'));
    model.add(Dense(1, init='uniform', activation='tanh'));
    adam = Adam(lr = 0.05);
    model.compile(loss='mean_squared_error', optimizer=adam, metrics=['accuracy']);
    model.fit(x_sample, y_sample, nb_epoch=1000, batch_size=20,callbacks=callbacks_list);
    
    #测试数据
    x_new = np.linspace(low, up, 1000);
    y_new = getfit(model,x_new);
    
    # 数据可视化
    plt.plot(x,y); 
    plt.scatter(x_sample, y_sample);
    plt.plot(x_new,y_new);
    
    plt.show();

2.2 模型保存

 在神经网络训练的过程中，有一个非常重要的操作，就是将训练过程中模型的参数保存到本地，这是后面拟合过程可视化的基础。训练过程中保存的模型文件，如下图所示。

模型保存的关键在于fit函数中callback函数的设置，注意到，下面的代码，每次迭代，算法都会执行callbacks函数指定的函数列表中的方法。这里，我们的回调函数设置为ModelCheckpoint，其参数如下表所示：

参数	含义
filename	字符串，保存模型的路径
verbose	信息展示模式，0或1 (Epoch 00001: saving model to ...)
mode	‘auto’，‘min’，‘max’
monitor	需要监视的值
save_best_only	当设置为True时，监测值有改进时才会保存当前的模型。在save_best_only=True时决定性能最佳模型的评判准则，例如，当监测值为val_acc时，模式应为max，当监测值为val_loss时，模式应为min。在auto模式下，评价准则由被监测值的名字自动推断
save_weights_only	若设置为True，则只保存模型权重，否则将保存整个模型（包括模型结构，配置信息等）
period	CheckPoint之间的间隔的epoch数

    # callback
    filepath="E:/Model/weights-improvement-{epoch:00d}.hdf5";
    checkpoint= ModelCheckpoint(filepath, verbose=1, save_best_only=False, mode='max');
    callbacks_list= [checkpoint];
    
    # 建立顺序神经网络层次模型
    model = Sequential();  
    model.add(Dense(10, input_dim=1, init='uniform', activation='relu'));
    model.add(Dense(1, init='uniform', activation='tanh'));
    adam = Adam(lr = 0.05);
    model.compile(loss='mean_squared_error', optimizer=adam, metrics=['accuracy']);
    model.fit(x_sample, y_sample, nb_epoch=1000, batch_size=20,callbacks=callbacks_list);

2.3 拟合过程可视化实现 

利用上述保存的模型，我们就可以通过matplotlib实时地显示拟合过程。

import math;
import random;
from matplotlib import pyplot as plt
from keras.models import Sequential
from keras.layers.core import Dense
import numpy as np
import matplotlib.animation as animation
from PIL import Image

#定义kdd99数据预处理函数
def sample(low, up, num):
    data = [];
    for i in range(num):
        #采样
        tmp = random.uniform(low, up);
        data.append(tmp);
    data.sort();
    return data;

def func(x):
    y = [];
    for i in range(len(x)):
        tmp = math.sin(x[i] - math.pi/3);
        y.append(tmp);
    return y;

def getfit(model,x):    
    y = [];
    for i in range(len(x)):
        tmp = model.predict([x[i]], 10);
        y.append(tmp[0][0]);
    return y;

def init():
    fpath = "E:/imgs/0.jpg";
    img = Image.open(fpath);
    plt.axis('off') # 关掉坐标轴为 off
    return plt.imshow(img);

def update(i): 
    fpath = "E:/imgs/" + str(i) + ".jpg";
    img = Image.open(fpath);
    plt.axis('off') # 关掉坐标轴为 off
    return plt.imshow(img);

if __name__ == '__main__':    
    low = 0;
    up = 2 * math.pi;
    x = np.linspace(low, up, 1000);
    y = func(x);
    
    # 数据采样
#     x_sample = sample(low,up,20);
    x_sample = [0.09326442022999694, 0.5812590520508311, 1.040490143783586, 1.5504427746047338, 2.1589557183817036, 2.6235357787018407, 2.712578091093361, 2.7379109336528167, 3.0339662651841186, 3.147676812083248, 3.58596337171837, 3.6621496731124314, 3.81130899864203, 3.833092859928872, 4.396611340802901, 4.4481080339256875, 4.609657879057151, 5.399731063412583, 5.54299720786794, 5.764084730699906];
    y_sample = func(x_sample);
    
    # 建立顺序神经网络层次模型
    model = Sequential();  
    model.add(Dense(10, input_dim=1, init='uniform', activation='relu'));
    model.add(Dense(1, init='uniform', activation='tanh'));
        
    plt.ion(); #开启interactive mode 成功的关键函数
    fig = plt.figure(1);
    
    for i in range(100):
        filepath="E:/Model/weights-improvement-" + str(i + 1) + ".hdf5";
        model.load_weights(filepath);
        #测试数据
        x_new = np.linspace(low, up, 1000);
        y_new = getfit(model,x_new);
        # 显示数据
        plt.clf();
        plt.plot(x,y); 
        plt.scatter(x_sample, y_sample);
        plt.plot(x_new,y_new);
        
        ffpath = "E:/imgs/" + str(i) + ".jpg";
        plt.savefig(ffpath);

        plt.pause(0.01)             # 暂停0.01秒
        
    ani = animation.FuncAnimation(plt.figure(2), update,range(100),init_func=init, interval=500);
    ani.save("E:/test.gif",writer='pillow');
    
    plt.ioff()                 # 关闭交互模式