def densenettrain():
print("densenet begin")
model = fc.get_model(50176)
root = '/Users/lizhenhao/Desktop/helloworld/im/densenet'
weights_savedir = "/Users/lizhenhao/PycharmProjects/laserFilm/weights/freeze/densenet"
acc_savepath = "/Users/lizhenhao/PycharmProjects/laserFilm/weights/freeze/densenet/acc.png"
loss_savepath = "/Users/lizhenhao/PycharmProjects/laserFilm/weights/freeze/densenet/loss.png"
train_f, train_y = getdata_fromtxt(root + '/train.txt')
vali_f, vali_y = getdata_fromtxt(root + '/vali.txt')
test_f, test_y = getdata_fromtxt(root + '/test.txt')
mycall = mycallback(test_f, test_y, weights_savedir)
model.compile(loss=tf.keras.losses.binary_crossentropy,
optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
metrics=[tf.keras.metrics.binary_accuracy])
model.fit(train_f, train_y, epochs=20, shuffle=True, validation_data=(vali_f, vali_y), callbacks=[mycall])
draw.draw_acc(mycall.train_acc, mycall.vali_acc, mycall.test_acc, mycall.epoch, acc_savepath)
draw.draw_loss(mycall.train_loss, mycall.vali_loss, mycall.test_loss, mycall.epoch, loss_savepath)
print('densenet done')
return model
loss = loss + (y_new[i] - y_p) * (y_new[i] - y_p) / 2
all_da = all_da + da(y_new[i], y_p, x_new[i])
all_db = all_db + db(y_new[i], y_p)
#loss_ = calc_loss(a = a,b=b,x=np.array(x),y=np.array(y))
all_d = np.array([all_da / len(x_new), all_db / len(x_new)])
#all_d = np.array([all_da, all_db])
loss = loss / len(x_new)
# 在图1的曲面上绘制 loss 点
draw.draw_curved_surface_loss(a, b, loss, curved_surface)
# 在图2的等高线上绘制 loss 点
draw.draw_equal_altitude_loss(a, b, last_a, last_b)
# 绘制图3中的回归直线
draw.draw_regression_line(a, b, x, y)
# 绘制图4的loss更新曲线
draw.draw_loss(loss, step, all_loss, all_step)
# 更新参数
last_a = a
last_b = b
E_g = gamma * E_g + (1 - gamma) * (all_db**2)
rms_g = np.sqrt(E_g + epsilon)
E_theta = gamma * E_theta + (1 - gamma) * (theta**2)
rms_theta = np.sqrt(E_theta + epsilon)
theta = -(rms_theta / rms_g) * all_d
[a, b] = [a, b] + theta
if step % 1 == 0: