[人工智能]test1

import sklearn.preprocessing
import tensorflow as tf
import matplotlib.pyplot as plt
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from sklearn.metrics import classification_report
from sklearn.metrics import f1_score,accuracy_score

from network import *
from load import *
from load_2 import *
# import load_2

BS =300
EPOCHS = 500
NUM_C = 21
d_LEARNING_RATE = 0.005
BETA_1 = 0.5




def train():

    #读取网络输入数据
    X_train, y_train, X_test, y_test = load_2()
    # print(X_train.shape)

    # 转换为one_hot类型
    # Y_train_one = tf.keras.utils.to_categorical(y_train,num_classes=NUM_C)
    # Y_test_one = tf.keras.utils.to_categorical(y_test,num_classes=NUM_C)
    # Y_train_one = tf.one_hot(y_train, NUM_C)
    # Y_test_one = tf.one_hot(y_test, NUM_C)


    #构建CNN分类网络
    d = CNN_focus_predict_model()
    #编译CNN分类网络
    d_optimizer = tf.keras.optimizers.Adam(lr=d_LEARNING_RATE, beta_1=BETA_1)
    d.compile(loss='mean_squared_error', optimizer=d_optimizer, metrics=['accuracy']) #loss='categorical_crossentropy'
    #训练方法1
    H = d.fit(X_train, y_train, batch_size=100, epochs=EPOCHS, verbose=2)
    # #训练方法2
    # aug = ImageDataGenerator(rotation_range=30, width_shift_range=0.1,
    #                          height_shift_range=0.1, shear_range=0.2, zoom_range=0.2,
    #                          horizontal_flip=True, fill_mode="nearest")
    # H = d.fit_generator(aug.flow(X_train, Y_train_one, batch_size=BS),steps_per_epoch=len(X_train) // BS,
    #                         epochs=EPOCHS)

    # 评估模型
    score = d.evaluate(X_test, y_test, verbose=0)
    print('Test score:', score[0])
    # print('Test accuracy:', score[1])

    # 测试
    # lb = sklearn.preprocessing.MultiLabelBinarizer()
    # print("------测试网络------")
    # predictions = d.predict(X_test, batch_size=32)
    # print(classification_report(y_test.argmax(axis=1),
    #                             predictions.argmax(axis=1), target_names=lb.classes_))

    #绘制loss曲线
    N1 = np.arange(0, EPOCHS)
    plt.style.use("ggplot")
    plt.figure()
    plt.plot(N1, H.history["loss"], label="train_loss")
    # plt.plot(N1, H.history["accuracy"], label="train_acc")
    plt.title("Training Loss")
    plt.xlabel("Epoch #")
    plt.ylabel("Loss/Accuracy")
    plt.legend()
    plt.savefig('huigui_loss.png')

    #绘制预测结果图（回归问题）
    y_test_len = len(y_test)
    N = np.arange(0, y_test_len)
    Y_pred = d.predict(X_test)

    plt.style.use("ggplot")
    plt.figure()
    plt.scatter(N, y_test, label="sample",color='red',marker =".",s= 10)
    plt.scatter(N, Y_pred, label="predict",color='blue',marker ="." ,s=10)
    # plt.plot(N, Y_pred, label="predict",color='blue',linestyle='--')
    plt.title("Sample points and predicted results")
    plt.legend() #画图lable
    plt.savefig('huigui_sample.png')



    #保存训练网络参数
    d.save_weights("./generator_weight.h5", True)





if __name__ == "__main__":
    train()