numpy实现简单的二层网络------前向和反向传播

会推导神经网络的前向和反向传播是深度学习的基础，也许大家在实验使用Tensorflow框架，只需要调用某一个损失函数，传入参数就可以得到损失，或者Mxnet框架，你都不需要调用任何函数，给个标志就可以了。但是，你真的会推导么，如果你不会，千万别说你正在学“深度学习”，由于时间问题，我没有太多的时间阐述基本原理，以及推导公式，加我QQ，我们可以详细的讨论，废话不多说，直接拷贝代码：

def background():
    """
        by: nxg
    """
    import numpy as np
    input = np.array([0.99, 0.01], dtype='float')
    weight_1 = np.array([[0.35, 0.55],
                         [0.90, 0.36]], dtype='float')
    weight_2 = np.array([[0.60, 0.77],
                         [0.96, 0.89]], dtype='float')
    bias = np.array([0.23, 0.79], dtype='float')
    target_output = np.array([0.1, 0.89], dtype='float')
    learning_rate = 0.5
    steps = 10001
    while steps > 0:
        steps -= 1
        # 正向传播
        # layer-1
        net_h1, net_h2 = np.dot(input, weight_1) + bias[0]  # 第一层的计算结果
        out_h1, out_h2 = [1 / (1 + np.exp(-net_h1)), 1 / (1 + np.exp(-net_h2))]  # sigmoid激活后的输出
        # layer-2
        layer_2_input = np.array([out_h1, out_h2], dtype='float')
        net_o1, net_o2 = np.dot(layer_2_input, weight_2) + bias[1]  # 第二层的计算结果
        out_o1, out_o2 = [1 / (1 + np.exp(-net_o1)), 1 / (1 + np.exp(-net_o2))]  # sigmoid激活后的输出
        # 计算梯度,反向传播
        total_loss = sum([pow(target_output-real_output, 2)/2 for target_output, real_output
                          in zip(target_output, [out_o1, out_o2])])
        print('网络的输出：step:%d loss=%f out_o1=%f out_o2=%f ' % (steps, total_loss, out_o1, out_o2))
        # gradient for layer-2
        gradient_w5 = -1 * (target_output[0] - out_o1) * out_o1 * (1 - out_o1) * out_h1
        gradient_w6 = -1 * (target_output[0] - out_o1) * out_o1 * (1 - out_o1) * out_h2
        gradient_w7 = -1 * (target_output[1] - out_o2) * out_o2 * (1 - out_o2) * out_h1
        gradient_w8 = -1 * (target_output[1] - out_o2) * out_o2 * (1 - out_o2) * out_h2

        # gradient for layer-1
        gradient_w1 = -1 * (target_output[0] - out_o1) * out_o1 * (1 - out_o1) * weight_2[0][0] * out_h1 * \
                      (1 - out_h1)*input[0] - (target_output[1] - out_o2) * out_o2 * (1 - out_o2) * weight_2[1][0] * \
                      out_h1 * (1 - out_h1)*input[0]

        gradient_w2 = -1 * (target_output[0] - out_o1) * out_o1 * (1 - out_o1) * weight_2[0][0] * out_h1 * \
                      (1 - out_h1) * input[1] - (target_output[1] - out_o2) * out_o2 * (1 - out_o2) * weight_2[1][0] * \
                      out_h1 * (1 - out_h1)*input[1]

        gradient_w3 = -1 * (target_output[0] - out_o1) * out_o1 * (1 - out_o1) * weight_2[0][1] * out_h2 * \
                      (1 - out_h2) * input[0] - (target_output[1] - out_o2) * out_o2 * (1 - out_o2) * weight_2[1][1] * \
                      out_h2 * (1 - out_h2)*input[0]

        gradient_w4 = -1 * (target_output[0] - out_o1) * out_o1 * (1 - out_o1) * weight_2[0][1] * out_h2 * \
                      (1 - out_h2) * input[1] - (target_output[1] - out_o2) * out_o2 * (1 - out_o2) * weight_2[1][1] * \
                      out_h2 * (1 - out_h2)*input[1]

        # update weight
        # layer -1
        weight_1[0][0] = weight_1[0][0] - learning_rate * gradient_w1
        weight_1[1][0] = weight_1[0][1] - learning_rate * gradient_w2
        weight_1[0][1] = weight_1[1][0] - learning_rate * gradient_w3
        weight_1[1][1] = weight_1[1][1] - learning_rate * gradient_w4

        # layer-2
        weight_2[0][0] = weight_2[0][0] - learning_rate * gradient_w5
        weight_2[1][0] = weight_2[0][1] - learning_rate * gradient_w6
        weight_2[0][1] = weight_2[1][0] - learning_rate * gradient_w7
        weight_2[1][1] = weight_2[1][1] - learning_rate * gradient_w8
        # print('weight_1:', weight_1)
        # print('weight_2:', weight_2)
        if steps == 1:
            print('网络的输出：out_o1=%f out_o2=%f loss=%f' % (out_o1, out_o2, total_loss))
            break

background()

实验结果：

网络的输出：step:10000 loss=0.290651 out_o1=0.862149 out_o2=0.869266
网络的输出：step:9999 loss=0.274318 out_o1=0.840335 out_o2=0.866760
网络的输出：step:9998 loss=0.269865 out_o1=0.834205 out_o2=0.864076
网络的输出：step:9997 loss=0.265266 out_o1=0.827812 out_o2=0.861333
网络的输出：step:9996 loss=0.260522 out_o1=0.821148 out_o2=0.858532
网络的输出：step:9995 loss=0.255635 out_o1=0.814208 out_o2=0.855674
网络的输出：step:9994 loss=0.250608 out_o1=0.806986 out_o2=0.852762
网络的输出：step:9993 loss=0.245444 out_o1=0.799480 out_o2=0.849797
网络的输出：step:9992 loss=0.240147 out_o1=0.791684 out_o2=0.846783
网络的输出：step:9991 loss=0.234724 out_o1=0.783598 out_o2=0.843721
网络的输出：step:9990 loss=0.229181 out_o1=0.775221 out_o2=0.840615
网络的输出：step:9989 loss=0.223527 out_o1=0.766554 out_o2=0.837469
网络的输出：step:9988 loss=0.217770 out_o1=0.757598 out_o2=0.834287
网络的输出：step:9987 loss=0.211921 out_o1=0.748359 out_o2=0.831073
网络的输出：step:9986 loss=0.205993 out_o1=0.738843 out_o2=0.827832
网络的输出：step:9985 loss=0.199997 out_o1=0.729058 out_o2=0.824571
网络的输出：step:9984 loss=0.193949 out_o1=0.719014 out_o2=0.821293
网络的输出：step:9983 loss=0.187864 out_o1=0.708723 out_o2=0.818006
网络的输出：step:9982 loss=0.181756 out_o1=0.698202 out_o2=0.814717
网络的输出：step:9981 loss=0.175644 out_o1=0.687465 out_o2=0.811431
网络的输出：step:9980 loss=0.169544 out_o1=0.676533 out_o2=0.808156
网络的输出：step:9979 loss=0.163475 out_o1=0.665428 out_o2=0.804901
网络的输出：step:9978 loss=0.157454 out_o1=0.654172 out_o2=0.801671
网络的输出：step:9977 loss=0.151499 out_o1=0.642791 out_o2=0.798476
网络的输出：step:9976 loss=0.145628 out_o1=0.631311 out_o2=0.795324
网络的输出：step:9975 loss=0.139857 out_o1=0.619762 out_o2=0.792222
网络的输出：step:9974 loss=0.134202 out_o1=0.608173 out_o2=0.789179
网络的输出：step:9973 loss=0.128679 out_o1=0.596573 out_o2=0.786202
网络的输出：step:9972 loss=0.123301 out_o1=0.584993 out_o2=0.783300
网络的输出：step:9971 loss=0.118081 out_o1=0.573462 out_o2=0.780479
网络的输出：step:9970 loss=0.113027 out_o1=0.562011 out_o2=0.777748
网络的输出：step:9969 loss=0.108150 out_o1=0.550668 out_o2=0.775113
网络的输出：step:9968 loss=0.103456 out_o1=0.539460 out_o2=0.772580
网络的输出：step:9967 loss=0.098950 out_o1=0.528413 out_o2=0.770154
网络的输出：step:9966 loss=0.094635 out_o1=0.517549 out_o2=0.767840
网络的输出：step:9965 loss=0.090512 out_o1=0.506890 out_o2=0.765643
网络的输出：step:9964 loss=0.086581 out_o1=0.496454 out_o2=0.763567
网络的输出：step:9963 loss=0.082839 out_o1=0.486258 out_o2=0.761613
网络的输出：step:9962 loss=0.079284 out_o1=0.476315 out_o2=0.759784
网络的输出：step:9961 loss=0.075912 out_o1=0.466635 out_o2=0.758081
网络的输出：step:9960 loss=0.072717 out_o1=0.457228 out_o2=0.756504
网络的输出：step:9959 loss=0.069692 out_o1=0.448100 out_o2=0.755053
网络的输出：step:9958 loss=0.066831 out_o1=0.439253 out_o2=0.753728
网络的输出：step:9957 loss=0.064128 out_o1=0.430691 out_o2=0.752527
网络的输出：step:9956 loss=0.061573 out_o1=0.422411 out_o2=0.751447
网络的输出：step:9955 loss=0.059160 out_o1=0.414414 out_o2=0.750487
网络的输出：step:9954 loss=0.056881 out_o1=0.406696 out_o2=0.749642
网络的输出：step:9953 loss=0.054729 out_o1=0.399251 out_o2=0.748911
网络的输出：step:9952 loss=0.052695 out_o1=0.392075 out_o2=0.748289
网络的输出：step:9951 loss=0.050773 out_o1=0.385162 out_o2=0.747773
网络的输出：step:9950 loss=0.048955 out_o1=0.378503 out_o2=0.747358
网络的输出：step:9949 loss=0.047236 out_o1=0.372092 out_o2=0.747040
网络的输出：step:9948 loss=0.045608 out_o1=0.365921 out_o2=0.746815
网络的输出：step:9947 loss=0.044065 out_o1=0.359981 out_o2=0.746679
网络的输出：step:9946 loss=0.042603 out_o1=0.354264 out_o2=0.746627
网络的输出：step:9945 loss=0.041215 out_o1=0.348761 out_o2=0.746654
网络的输出：step:9944 loss=0.039897 out_o1=0.343464 out_o2=0.746756
网络的输出：step:9943 loss=0.038643 out_o1=0.338364 out_o2=0.746930
网络的输出：step:9942 loss=0.037451 out_o1=0.333454 out_o2=0.747171
网络的输出：step:9941 loss=0.036314 out_o1=0.328725 out_o2=0.747474
网络的输出：step:9940 loss=0.035231 out_o1=0.324169 out_o2=0.747836
网络的输出：step:9939 loss=0.034198 out_o1=0.319779 out_o2=0.748252
网络的输出：step:9938 loss=0.033210 out_o1=0.315547 out_o2=0.748720
网络的输出：step:9937 loss=0.032266 out_o1=0.311466 out_o2=0.749236
网络的输出：step:9936 loss=0.031363 out_o1=0.307529 out_o2=0.749796
网络的输出：step:9935 loss=0.030498 out_o1=0.303730 out_o2=0.750396
网络的输出：step:9934 loss=0.029668 out_o1=0.300062 out_o2=0.751034
网络的输出：step:9933 loss=0.028872 out_o1=0.296520 out_o2=0.751707
网络的输出：step:9932 loss=0.028108 out_o1=0.293097 out_o2=0.752412
网络的输出：step:9931 loss=0.027374 out_o1=0.289788 out_o2=0.753146
网络的输出：step:9930 loss=0.026668 out_o1=0.286588 out_o2=0.753907
网络的输出：step:9929 loss=0.025989 out_o1=0.283492 out_o2=0.754692
网络的输出：step:9928 loss=0.025335 out_o1=0.280495 out_o2=0.755498
网络的输出：step:9927 loss=0.024704 out_o1=0.277592 out_o2=0.756325
网络的输出：step:9926 loss=0.024096 out_o1=0.274780 out_o2=0.757169
网络的输出：step:9925 loss=0.023509 out_o1=0.272054 out_o2=0.758029
网络的输出：step:9924 loss=0.022943 out_o1=0.269411 out_o2=0.758904
网络的输出：step:9923 loss=0.022396 out_o1=0.266847 out_o2=0.759790
网络的输出：step:9922 loss=0.021868 out_o1=0.264358 out_o2=0.760688
网络的输出：step:9921 loss=0.021356 out_o1=0.261941 out_o2=0.761594
网络的输出：step:9920 loss=0.020862 out_o1=0.259593 out_o2=0.762509
网络的输出：step:9919 loss=0.020383 out_o1=0.257311 out_o2=0.763430
网络的输出：step:9918 loss=0.019920 out_o1=0.255093 out_o2=0.764357
网络的输出：step:9917 loss=0.019471 out_o1=0.252935 out_o2=0.765288
网络的输出：step:9916 loss=0.019036 out_o1=0.250835 out_o2=0.766223
网络的输出：step:9915 loss=0.018614 out_o1=0.248791 out_o2=0.767160
网络的输出：step:9914 loss=0.018205 out_o1=0.246801 out_o2=0.768098
网络的输出：step:9913 loss=0.017808 out_o1=0.244861 out_o2=0.769037
网络的输出：step:9912 loss=0.017423 out_o1=0.242971 out_o2=0.769975
网络的输出：step:9911 loss=0.017049 out_o1=0.241129 out_o2=0.770913
网络的输出：step:9910 loss=0.016686 out_o1=0.239332 out_o2=0.771850
网络的输出：step:9909 loss=0.016334 out_o1=0.237578 out_o2=0.772784
网络的输出：step:9908 loss=0.015991 out_o1=0.235867 out_o2=0.773715
网络的输出：step:9907 loss=0.015658 out_o1=0.234196 out_o2=0.774644
网络的输出：step:9906 loss=0.015334 out_o1=0.232564 out_o2=0.775568
网络的输出：step:9905 loss=0.015019 out_o1=0.230970 out_o2=0.776489
网络的输出：step:9904 loss=0.014713 out_o1=0.229412 out_o2=0.777405
网络的输出：step:9903 loss=0.014415 out_o1=0.227889 out_o2=0.778316
网络的输出：step:9902 loss=0.014124 out_o1=0.226400 out_o2=0.779222
网络的输出：step:9901 loss=0.013842 out_o1=0.224943 out_o2=0.780122
网络的输出：step:9900 loss=0.013567 out_o1=0.223518 out_o2=0.781017
网络的输出：step:9899 loss=0.013299 out_o1=0.222122 out_o2=0.781905
网络的输出：step:9898 loss=0.013038 out_o1=0.220756 out_o2=0.782788
网络的输出：step:9897 loss=0.012784 out_o1=0.219419 out_o2=0.783664
网络的输出：step:9896 loss=0.012536 out_o1=0.218109 out_o2=0.784533
网络的输出：step:9895 loss=0.012295 out_o1=0.216825 out_o2=0.785395
网络的输出：step:9894 loss=0.012060 out_o1=0.215567 out_o2=0.786251
网络的输出：step:9893 loss=0.011830 out_o1=0.214333 out_o2=0.787099
网络的输出：step:9892 loss=0.011607 out_o1=0.213124 out_o2=0.787941
网络的输出：step:9891 loss=0.011388 out_o1=0.211938 out_o2=0.788775
网络的输出：step:9890 loss=0.011175 out_o1=0.210774 out_o2=0.789601
网络的输出：step:9889 loss=0.010968 out_o1=0.209633 out_o2=0.790421
网络的输出：step:9888 loss=0.010765 out_o1=0.208513 out_o2=0.791233
网络的输出：step:9887 loss=0.010567 out_o1=0.207413 out_o2=0.792037
网络的输出：step:9886 loss=0.010374 out_o1=0.206333 out_o2=0.792834
网络的输出：step:9885 loss=0.010185 out_o1=0.205273 out_o2=0.793624
网络的输出：step:9884 loss=0.010001 out_o1=0.204231 out_o2=0.794405
网络的输出：step:9883 loss=0.009821 out_o1=0.203208 out_o2=0.795180
网络的输出：step:9882 loss=0.009646 out_o1=0.202203 out_o2=0.795947
网络的输出：step:9881 loss=0.009474 out_o1=0.201215 out_o2=0.796706
网络的输出：step:9880 loss=0.009306 out_o1=0.200244 out_o2=0.797458
网络的输出：step:9879 loss=0.009143 out_o1=0.199289 out_o2=0.798202
网络的输出：step:9878 loss=0.008982 out_o1=0.198350 out_o2=0.798938
网络的输出：step:9877 loss=0.008826 out_o1=0.197427 out_o2=0.799668
网络的输出：step:9876 loss=0.008673 out_o1=0.196518 out_o2=0.800390
网络的输出：step:9875 loss=0.008523 out_o1=0.195625 out_o2=0.801104
网络的输出：step:9874 loss=0.008377 out_o1=0.194746 out_o2=0.801811
网络的输出：step:9873 loss=0.008234 out_o1=0.193881 out_o2=0.802511
网络的输出：step:9872 loss=0.008094 out_o1=0.193029 out_o2=0.803204
网络的输出：step:9871 loss=0.007957 out_o1=0.192191 out_o2=0.803889
网络的输出：step:9870 loss=0.007823 out_o1=0.191366 out_o2=0.804568
网络的输出：step:9869 loss=0.007692 out_o1=0.190553 out_o2=0.805239
网络的输出：step:9868 loss=0.007564 out_o1=0.189753 out_o2=0.805903
网络的输出：step:9867 loss=0.007438 out_o1=0.188965 out_o2=0.806560
网络的输出：step:9866 loss=0.007316 out_o1=0.188188 out_o2=0.807211
网络的输出：step:9865 loss=0.007195 out_o1=0.187423 out_o2=0.807855
网络的输出：step:9864 loss=0.007078 out_o1=0.186670 out_o2=0.808491
网络的输出：step:9863 loss=0.006962 out_o1=0.185927 out_o2=0.809122
网络的输出：step:9862 loss=0.006850 out_o1=0.185195 out_o2=0.809745
网络的输出：step:9861 loss=0.006739 out_o1=0.184473 out_o2=0.810362
网络的输出：step:9860 loss=0.006631 out_o1=0.183762 out_o2=0.810972
网络的输出：step:9859 loss=0.006525 out_o1=0.183060 out_o2=0.811577
网络的输出：step:9858 loss=0.006421 out_o1=0.182369 out_o2=0.812174
网络的输出：step:9857 loss=0.006319 out_o1=0.181687 out_o2=0.812766
网络的输出：step:9856 loss=0.006219 out_o1=0.181014 out_o2=0.813351
网络的输出：step:9855 loss=0.006121 out_o1=0.180351 out_o2=0.813930
网络的输出：step:9854 loss=0.006026 out_o1=0.179697 out_o2=0.814503
网络的输出：step:9853 loss=0.005932 out_o1=0.179051 out_o2=0.815070
网络的输出：step:9852 loss=0.005840 out_o1=0.178414 out_o2=0.815631
网络的输出：step:9851 loss=0.005750 out_o1=0.177785 out_o2=0.816186
网络的输出：step:9850 loss=0.005661 out_o1=0.177165 out_o2=0.816735
网络的输出：step:9849 loss=0.005574 out_o1=0.176553 out_o2=0.817279
网络的输出：step:9848 loss=0.005489 out_o1=0.175948 out_o2=0.817817
网络的输出：step:9847 loss=0.005406 out_o1=0.175352 out_o2=0.818349
网络的输出：step:9846 loss=0.005324 out_o1=0.174763 out_o2=0.818876
网络的输出：step:9845 loss=0.005244 out_o1=0.174181 out_o2=0.819398
网络的输出：step:9844 loss=0.005165 out_o1=0.173607 out_o2=0.819914
网络的输出：step:9843 loss=0.005088 out_o1=0.173040 out_o2=0.820424
网络的输出：step:9842 loss=0.005012 out_o1=0.172480 out_o2=0.820930
网络的输出：step:9841 loss=0.004938 out_o1=0.171927 out_o2=0.821430
网络的输出：step:9840 loss=0.004865 out_o1=0.171381 out_o2=0.821925
网络的输出：step:9839 loss=0.004793 out_o1=0.170842 out_o2=0.822415
网络的输出：step:9838 loss=0.004723 out_o1=0.170309 out_o2=0.822900
网络的输出：step:9837 loss=0.004654 out_o1=0.169782 out_o2=0.823381
网络的输出：step:9836 loss=0.004586 out_o1=0.169262 out_o2=0.823856
网络的输出：step:9835 loss=0.004520 out_o1=0.168747 out_o2=0.824326
网络的输出：step:9834 loss=0.004454 out_o1=0.168239 out_o2=0.824792
网络的输出：step:9833 loss=0.004390 out_o1=0.167737 out_o2=0.825253
网络的输出：step:9832 loss=0.004327 out_o1=0.167241 out_o2=0.825709
网络的输出：step:9831 loss=0.004266 out_o1=0.166750 out_o2=0.826161
网络的输出：step:9830 loss=0.004205 out_o1=0.166265 out_o2=0.826608
网络的输出：step:9829 loss=0.004145 out_o1=0.165786 out_o2=0.827051
网络的输出：step:9828 loss=0.004087 out_o1=0.165312 out_o2=0.827489
网络的输出：step:9827 loss=0.004029 out_o1=0.164843 out_o2=0.827924
网络的输出：step:9826 loss=0.003973 out_o1=0.164380 out_o2=0.828353
网络的输出：step:9825 loss=0.003917 out_o1=0.163922 out_o2=0.828779
网络的输出：step:9824 loss=0.003862 out_o1=0.163469 out_o2=0.829200
网络的输出：step:9823 loss=0.003809 out_o1=0.163021 out_o2=0.829617
网络的输出：step:9822 loss=0.003756 out_o1=0.162578 out_o2=0.830031
网络的输出：step:9821 loss=0.003704 out_o1=0.162140 out_o2=0.830440
网络的输出：step:9820 loss=0.003653 out_o1=0.161706 out_o2=0.830845
网络的输出：step:9819 loss=0.003603 out_o1=0.161277 out_o2=0.831246
网络的输出：step:9818 loss=0.003554 out_o1=0.160853 out_o2=0.831643
网络的输出：step:9817 loss=0.003506 out_o1=0.160433 out_o2=0.832037
网络的输出：step:9816 loss=0.003458 out_o1=0.160018 out_o2=0.832427
网络的输出：step:9815 loss=0.003412 out_o1=0.159607 out_o2=0.832813
网络的输出：step:9814 loss=0.003366 out_o1=0.159201 out_o2=0.833195
网络的输出：step:9813 loss=0.003321 out_o1=0.158798 out_o2=0.833574
网络的输出：step:9812 loss=0.003276 out_o1=0.158400 out_o2=0.833949
网络的输出：step:9811 loss=0.003232 out_o1=0.158006 out_o2=0.834320
网络的输出：step:9810 loss=0.003190 out_o1=0.157617 out_o2=0.834688
网络的输出：step:9809 loss=0.003147 out_o1=0.157231 out_o2=0.835053
网络的输出：step:9808 loss=0.003106 out_o1=0.156849 out_o2=0.835414
网络的输出：step:9807 loss=0.003065 out_o1=0.156471 out_o2=0.835772
网络的输出：step:9806 loss=0.003025 out_o1=0.156096 out_o2=0.836127
网络的输出：step:9805 loss=0.002985 out_o1=0.155726 out_o2=0.836478
网络的输出：step:9804 loss=0.002946 out_o1=0.155359 out_o2=0.836826
网络的输出：step:9803 loss=0.002908 out_o1=0.154996 out_o2=0.837171
网络的输出：step:9802 loss=0.002870 out_o1=0.154636 out_o2=0.837512
网络的输出：step:9801 loss=0.002833 out_o1=0.154280 out_o2=0.837851
网络的输出：step:9800 loss=0.002796 out_o1=0.153928 out_o2=0.838186
网络的输出：step:9799 loss=0.002761 out_o1=0.153579 out_o2=0.838518
网络的输出：step:9798 loss=0.002725 out_o1=0.153233 out_o2=0.838848
网络的输出：step:9797 loss=0.002690 out_o1=0.152891 out_o2=0.839174
网络的输出：step:9796 loss=0.002656 out_o1=0.152552 out_o2=0.839497
网络的输出：step:9795 loss=0.002622 out_o1=0.152216 out_o2=0.839818
网络的输出：step:9794 loss=0.002589 out_o1=0.151883 out_o2=0.840135
网络的输出：step:9793 loss=0.002556 out_o1=0.151554 out_o2=0.840450
网络的输出：step:9792 loss=0.002524 out_o1=0.151227 out_o2=0.840762
网络的输出：step:9791 loss=0.002493 out_o1=0.150904 out_o2=0.841071
网络的输出：step:9790 loss=0.002461 out_o1=0.150584 out_o2=0.841378
网络的输出：step:9789 loss=0.002431 out_o1=0.150267 out_o2=0.841682
网络的输出：step:9788 loss=0.002400 out_o1=0.149952 out_o2=0.841983
网络的输出：step:9787 loss=0.002371 out_o1=0.149641 out_o2=0.842281
网络的输出：step:9786 loss=0.002341 out_o1=0.149332 out_o2=0.842577
网络的输出：step:9785 loss=0.002312 out_o1=0.149027 out_o2=0.842870
网络的输出：step:9784 loss=0.002284 out_o1=0.148724 out_o2=0.843161
网络的输出：step:9783 loss=0.002256 out_o1=0.148423 out_o2=0.843449
网络的输出：step:9782 loss=0.002228 out_o1=0.148126 out_o2=0.843735
网络的输出：step:9781 loss=0.002201 out_o1=0.147831 out_o2=0.844018
网络的输出：step:9780 loss=0.002174 out_o1=0.147539 out_o2=0.844299
网络的输出：step:9779 loss=0.002148 out_o1=0.147250 out_o2=0.844578
网络的输出：step:9778 loss=0.002122 out_o1=0.146963 out_o2=0.844854
网络的输出：step:9777 loss=0.002096 out_o1=0.146678 out_o2=0.845127
网络的输出：step:9776 loss=0.002071 out_o1=0.146396 out_o2=0.845399
网络的输出：step:9775 loss=0.002046 out_o1=0.146117 out_o2=0.845668
网络的输出：step:9774 loss=0.002022 out_o1=0.145840 out_o2=0.845935
网络的输出：step:9773 loss=0.001997 out_o1=0.145565 out_o2=0.846200
网络的输出：step:9772 loss=0.001974 out_o1=0.145293 out_o2=0.846462
网络的输出：step:9771 loss=0.001950 out_o1=0.145023 out_o2=0.846722
网络的输出：step:9770 loss=0.001927 out_o1=0.144756 out_o2=0.846980
网络的输出：step:9769 loss=0.001904 out_o1=0.144491 out_o2=0.847236
网络的输出：step:9768 loss=0.001882 out_o1=0.144228 out_o2=0.847490
网络的输出：step:9767 loss=0.001859 out_o1=0.143967 out_o2=0.847742
网络的输出：step:9766 loss=0.001838 out_o1=0.143709 out_o2=0.847991
网络的输出：step:9765 loss=0.001816 out_o1=0.143452 out_o2=0.848239
网络的输出：step:9764 loss=0.001795 out_o1=0.143198 out_o2=0.848485
网络的输出：step:9763 loss=0.001774 out_o1=0.142946 out_o2=0.848728
网络的输出：step:9762 loss=0.001753 out_o1=0.142696 out_o2=0.848970
网络的输出：step:9761 loss=0.001733 out_o1=0.142448 out_o2=0.849210
网络的输出：step:9760 loss=0.001713 out_o1=0.142203 out_o2=0.849447
网络的输出：step:9759 loss=0.001693 out_o1=0.141959 out_o2=0.849683
网络的输出：step:9758 loss=0.001673 out_o1=0.141717 out_o2=0.849917
网络的输出：step:9757 loss=0.001654 out_o1=0.141477 out_o2=0.850149
网络的输出：step:9756 loss=0.001635 out_o1=0.141240 out_o2=0.850380
网络的输出：step:9755 loss=0.001617 out_o1=0.141004 out_o2=0.850608
网络的输出：step:9754 loss=0.001598 out_o1=0.140770 out_o2=0.850835
网络的输出：step:9753 loss=0.001580 out_o1=0.140538 out_o2=0.851059
网络的输出：step:9752 loss=0.001562 out_o1=0.140308 out_o2=0.851283
网络的输出：step:9751 loss=0.001544 out_o1=0.140079 out_o2=0.851504
网络的输出：step:9750 loss=0.001527 out_o1=0.139853 out_o2=0.851723
网络的输出：step:9749 loss=0.001509 out_o1=0.139628 out_o2=0.851941
网络的输出：step:9748 loss=0.001492 out_o1=0.139405 out_o2=0.852157
网络的输出：step:9747 loss=0.001476 out_o1=0.139184 out_o2=0.852372
网络的输出：step:9746 loss=0.001459 out_o1=0.138965 out_o2=0.852585
网络的输出：step:9745 loss=0.001443 out_o1=0.138747 out_o2=0.852796
网络的输出：step:9744 loss=0.001427 out_o1=0.138531 out_o2=0.853006
网络的输出：step:9743 loss=0.001411 out_o1=0.138317 out_o2=0.853213
网络的输出：step:9742 loss=0.001395 out_o1=0.138104 out_o2=0.853420
网络的输出：step:9741 loss=0.001380 out_o1=0.137893 out_o2=0.853625
网络的输出：step:9740 loss=0.001364 out_o1=0.137684 out_o2=0.853828
网络的输出：step:9739 loss=0.001349 out_o1=0.137476 out_o2=0.854030
网络的输出：step:9738 loss=0.001334 out_o1=0.137270 out_o2=0.854230
网络的输出：step:9737 loss=0.001320 out_o1=0.137065 out_o2=0.854428
网络的输出：step:9736 loss=0.001305 out_o1=0.136862 out_o2=0.854626
网络的输出：step:9735 loss=0.001291 out_o1=0.136661 out_o2=0.854821
网络的输出：step:9734 loss=0.001277 out_o1=0.136461 out_o2=0.855016
网络的输出：step:9733 loss=0.001263 out_o1=0.136262 out_o2=0.855208
网络的输出：step:9732 loss=0.001249 out_o1=0.136066 out_o2=0.855400
网络的输出：step:9731 loss=0.001235 out_o1=0.135870 out_o2=0.855590
网络的输出：step:9730 loss=0.001222 out_o1=0.135676 out_o2=0.855778
网络的输出：step:9729 loss=0.001209 out_o1=0.135484 out_o2=0.855965
网络的输出：step:9728 loss=0.001196 out_o1=0.135293 out_o2=0.856151
网络的输出：step:9727 loss=0.001183 out_o1=0.135103 out_o2=0.856335
网络的输出：step:9726 loss=0.001170 out_o1=0.134915 out_o2=0.856518
网络的输出：step:9725 loss=0.001157 out_o1=0.134728 out_o2=0.856700
网络的输出：step:9724 loss=0.001145 out_o1=0.134542 out_o2=0.856881
网络的输出：step:9723 loss=0.001133 out_o1=0.134358 out_o2=0.857060
网络的输出：step:9722 loss=0.001121 out_o1=0.134175 out_o2=0.857237
网络的输出：step:9721 loss=0.001109 out_o1=0.133994 out_o2=0.857414
网络的输出：step:9720 loss=0.001097 out_o1=0.133814 out_o2=0.857589
网络的输出：step:9719 loss=0.001085 out_o1=0.133635 out_o2=0.857763
网络的输出：step:9718 loss=0.001074 out_o1=0.133457 out_o2=0.857936
网络的输出：step:9717 loss=0.001062 out_o1=0.133281 out_o2=0.858107
网络的输出：step:9716 loss=0.001051 out_o1=0.133106 out_o2=0.858278
网络的输出：step:9715 loss=0.001040 out_o1=0.132932 out_o2=0.858447
网络的输出：step:9714 loss=0.001029 out_o1=0.132760 out_o2=0.858615
网络的输出：step:9713 loss=0.001018 out_o1=0.132588 out_o2=0.858781
网络的输出：step:9712 loss=0.001008 out_o1=0.132418 out_o2=0.858947
网络的输出：step:9711 loss=0.000997 out_o1=0.132249 out_o2=0.859111
网络的输出：step:9710 loss=0.000987 out_o1=0.132082 out_o2=0.859274
网络的输出：step:9709 loss=0.000976 out_o1=0.131915 out_o2=0.859436
网络的输出：step:9708 loss=0.000966 out_o1=0.131750 out_o2=0.859597
网络的输出：step:9707 loss=0.000956 out_o1=0.131586 out_o2=0.859757
网络的输出：step:9706 loss=0.000946 out_o1=0.131422 out_o2=0.859916
网络的输出：step:9705 loss=0.000936 out_o1=0.131261 out_o2=0.860073
网络的输出：step:9704 loss=0.000927 out_o1=0.131100 out_o2=0.860230
网络的输出：step:9703 loss=0.000917 out_o1=0.130940 out_o2=0.860385
网络的输出：step:9702 loss=0.000908 out_o1=0.130781 out_o2=0.860540
网络的输出：step:9701 loss=0.000898 out_o1=0.130624 out_o2=0.860693
网络的输出：step:9700 loss=0.000889 out_o1=0.130468 out_o2=0.860845
网络的输出：step:9699 loss=0.000880 out_o1=0.130312 out_o2=0.860996
网络的输出：step:9698 loss=0.000871 out_o1=0.130158 out_o2=0.861147
网络的输出：step:9697 loss=0.000862 out_o1=0.130005 out_o2=0.861296
网络的输出：step:9696 loss=0.000853 out_o1=0.129853 out_o2=0.861444
网络的输出：step:9695 loss=0.000845 out_o1=0.129701 out_o2=0.861591
网络的输出：step:9694 loss=0.000836 out_o1=0.129551 out_o2=0.861737
网络的输出：step:9693 loss=0.000828 out_o1=0.129402 out_o2=0.861882
网络的输出：step:9692 loss=0.000819 out_o1=0.129254 out_o2=0.862026
网络的输出：step:9691 loss=0.000811 out_o1=0.129107 out_o2=0.862170
网络的输出：step:9690 loss=0.000803 out_o1=0.128961 out_o2=0.862312
网络的输出：step:9689 loss=0.000795 out_o1=0.128815 out_o2=0.862453
网络的输出：step:9688 loss=0.000787 out_o1=0.128671 out_o2=0.862594
网络的输出：step:9687 loss=0.000779 out_o1=0.128528 out_o2=0.862733
网络的输出：step:9686 loss=0.000771 out_o1=0.128386 out_o2=0.862872
网络的输出：step:9685 loss=0.000763 out_o1=0.128244 out_o2=0.863009
网络的输出：step:9684 loss=0.000755 out_o1=0.128104 out_o2=0.863146
网络的输出：step:9683 loss=0.000748 out_o1=0.127964 out_o2=0.863282
网络的输出：step:9682 loss=0.000740 out_o1=0.127826 out_o2=0.863417
网络的输出：step:9681 loss=0.000733 out_o1=0.127688 out_o2=0.863551
网络的输出：step:9680 loss=0.000726 out_o1=0.127551 out_o2=0.863684
网络的输出：step:9679 loss=0.000719 out_o1=0.127415 out_o2=0.863816
网络的输出：step:9678 loss=0.000711 out_o1=0.127280 out_o2=0.863948
网络的输出：step:9677 loss=0.000704 out_o1=0.127146 out_o2=0.864078
网络的输出：step:9676 loss=0.000697 out_o1=0.127013 out_o2=0.864208
网络的输出：step:9675 loss=0.000691 out_o1=0.126880 out_o2=0.864337
网络的输出：step:9674 loss=0.000684 out_o1=0.126749 out_o2=0.864465
网络的输出：step:9673 loss=0.000677 out_o1=0.126618 out_o2=0.864592
网络的输出：step:9672 loss=0.000670 out_o1=0.126488 out_o2=0.864718
网络的输出：step:9671 loss=0.000664 out_o1=0.126359 out_o2=0.864844
网络的输出：step:9670 loss=0.000657 out_o1=0.126230 out_o2=0.864969
网络的输出：step:9669 loss=0.000651 out_o1=0.126103 out_o2=0.865093
网络的输出：step:9668 loss=0.000645 out_o1=0.125976 out_o2=0.865216
网络的输出：step:9667 loss=0.000638 out_o1=0.125850 out_o2=0.865339
网络的输出：step:9666 loss=0.000632 out_o1=0.125725 out_o2=0.865460
网络的输出：step:9665 loss=0.000626 out_o1=0.125601 out_o2=0.865581
网络的输出：step:9664 loss=0.000620 out_o1=0.125478 out_o2=0.865701
网络的输出：step:9663 loss=0.000614 out_o1=0.125355 out_o2=0.865821
网络的输出：step:9662 loss=0.000608 out_o1=0.125233 out_o2=0.865939
网络的输出：step:9661 loss=0.000602 out_o1=0.125112 out_o2=0.866057
网络的输出：step:9660 loss=0.000596 out_o1=0.124991 out_o2=0.866174
网络的输出：step:9659 loss=0.000590 out_o1=0.124872 out_o2=0.866291
网络的输出：step:9658 loss=0.000585 out_o1=0.124753 out_o2=0.866406
网络的输出：step:9657 loss=0.000579 out_o1=0.124634 out_o2=0.866521
网络的输出：step:9656 loss=0.000573 out_o1=0.124517 out_o2=0.866636
网络的输出：step:9655 loss=0.000568 out_o1=0.124400 out_o2=0.866749
网络的输出：step:9654 loss=0.000563 out_o1=0.124284 out_o2=0.866862
网络的输出：step:9653 loss=0.000557 out_o1=0.124169 out_o2=0.866974
网络的输出：step:9652 loss=0.000552 out_o1=0.124054 out_o2=0.867086
网络的输出：step:9651 loss=0.000547 out_o1=0.123940 out_o2=0.867197
网络的输出：step:9650 loss=0.000541 out_o1=0.123827 out_o2=0.867307
网络的输出：step:9649 loss=0.000536 out_o1=0.123714 out_o2=0.867416
网络的输出：step:9648 loss=0.000531 out_o1=0.123602 out_o2=0.867525
网络的输出：step:9647 loss=0.000526 out_o1=0.123491 out_o2=0.867633
网络的输出：step:9646 loss=0.000521 out_o1=0.123380 out_o2=0.867740
网络的输出：step:9645 loss=0.000516 out_o1=0.123270 out_o2=0.867847
网络的输出：step:9644 loss=0.000511 out_o1=0.123161 out_o2=0.867953
网络的输出：step:9643 loss=0.000506 out_o1=0.123053 out_o2=0.868059
网络的输出：step:9642 loss=0.000502 out_o1=0.122945 out_o2=0.868164
网络的输出：step:9641 loss=0.000497 out_o1=0.122837 out_o2=0.868268
网络的输出：step:9640 loss=0.000492 out_o1=0.122731 out_o2=0.868372
网络的输出：step:9639 loss=0.000488 out_o1=0.122625 out_o2=0.868475
网络的输出：step:9638 loss=0.000483 out_o1=0.122519 out_o2=0.868577
网络的输出：step:9637 loss=0.000479 out_o1=0.122414 out_o2=0.868679
网络的输出：step:9636 loss=0.000474 out_o1=0.122310 out_o2=0.868780
网络的输出：step:9635 loss=0.000470 out_o1=0.122207 out_o2=0.868881
网络的输出：step:9634 loss=0.000465 out_o1=0.122104 out_o2=0.868980
网络的输出：step:9633 loss=0.000461 out_o1=0.122002 out_o2=0.869080
网络的输出：step:9632 loss=0.000457 out_o1=0.121900 out_o2=0.869179
网络的输出：step:9631 loss=0.000452 out_o1=0.121799 out_o2=0.869277
网络的输出：step:9630 loss=0.000448 out_o1=0.121698 out_o2=0.869374
网络的输出：step:9629 loss=0.000444 out_o1=0.121598 out_o2=0.869472
网络的输出：step:9628 loss=0.000440 out_o1=0.121499 out_o2=0.869568
网络的输出：step:9627 loss=0.000436 out_o1=0.121400 out_o2=0.869664
网络的输出：step:9626 loss=0.000432 out_o1=0.121302 out_o2=0.869759
网络的输出：step:9625 loss=0.000428 out_o1=0.121204 out_o2=0.869854
网络的输出：step:9624 loss=0.000424 out_o1=0.121107 out_o2=0.869948
网络的输出：step:9623 loss=0.000420 out_o1=0.121010 out_o2=0.870042
网络的输出：step:9622 loss=0.000416 out_o1=0.120914 out_o2=0.870135
网络的输出：step:9621 loss=0.000412 out_o1=0.120819 out_o2=0.870228
网络的输出：step:9620 loss=0.000408 out_o1=0.120724 out_o2=0.870320
网络的输出：step:9619 loss=0.000405 out_o1=0.120630 out_o2=0.870411
网络的输出：step:9618 loss=0.000401 out_o1=0.120536 out_o2=0.870502
网络的输出：step:9617 loss=0.000397 out_o1=0.120442 out_o2=0.870593
网络的输出：step:9616 loss=0.000394 out_o1=0.120350 out_o2=0.870683
网络的输出：step:9615 loss=0.000390 out_o1=0.120257 out_o2=0.870772
网络的输出：step:9614 loss=0.000386 out_o1=0.120166 out_o2=0.870861
网络的输出：step:9613 loss=0.000383 out_o1=0.120074 out_o2=0.870950
网络的输出：step:9612 loss=0.000379 out_o1=0.119984 out_o2=0.871038
网络的输出：step:9611 loss=0.000376 out_o1=0.119894 out_o2=0.871125
网络的输出：step:9610 loss=0.000373 out_o1=0.119804 out_o2=0.871212
网络的输出：step:9609 loss=0.000369 out_o1=0.119715 out_o2=0.871299
网络的输出：step:9608 loss=0.000366 out_o1=0.119626 out_o2=0.871384
网络的输出：step:9607 loss=0.000363 out_o1=0.119538 out_o2=0.871470
网络的输出：step:9606 loss=0.000359 out_o1=0.119450 out_o2=0.871555
网络的输出：step:9605 loss=0.000356 out_o1=0.119363 out_o2=0.871639
网络的输出：step:9604 loss=0.000353 out_o1=0.119276 out_o2=0.871723
网络的输出：step:9603 loss=0.000350 out_o1=0.119190 out_o2=0.871807
网络的输出：step:9602 loss=0.000346 out_o1=0.119104 out_o2=0.871890
网络的输出：step:9601 loss=0.000343 out_o1=0.119019 out_o2=0.871973
网络的输出：step:9600 loss=0.000340 out_o1=0.118934 out_o2=0.872055
网络的输出：step:9599 loss=0.000337 out_o1=0.118850 out_o2=0.872137
网络的输出：step:9598 loss=0.000334 out_o1=0.118766 out_o2=0.872218
网络的输出：step:9597 loss=0.000331 out_o1=0.118682 out_o2=0.872299
网络的输出：step:9596 loss=0.000328 out_o1=0.118599 out_o2=0.872379
网络的输出：step:9595 loss=0.000325 out_o1=0.118517 out_o2=0.872459
网络的输出：step:9594 loss=0.000322 out_o1=0.118435 out_o2=0.872538
网络的输出：step:9593 loss=0.000319 out_o1=0.118353 out_o2=0.872617
网络的输出：step:9592 loss=0.000317 out_o1=0.118272 out_o2=0.872696
网络的输出：step:9591 loss=0.000314 out_o1=0.118191 out_o2=0.872774
网络的输出：step:9590 loss=0.000311 out_o1=0.118111 out_o2=0.872852
网络的输出：step:9589 loss=0.000308 out_o1=0.118031 out_o2=0.872929
网络的输出：step:9588 loss=0.000306 out_o1=0.117952 out_o2=0.873006
网络的输出：step:9587 loss=0.000303 out_o1=0.117873 out_o2=0.873082
网络的输出：step:9586 loss=0.000300 out_o1=0.117794 out_o2=0.873159
网络的输出：step:9585 loss=0.000297 out_o1=0.117716 out_o2=0.873234
网络的输出：step:9584 loss=0.000295 out_o1=0.117638 out_o2=0.873309
网络的输出：step:9583 loss=0.000292 out_o1=0.117561 out_o2=0.873384
网络的输出：step:9582 loss=0.000290 out_o1=0.117484 out_o2=0.873459
网络的输出：step:9581 loss=0.000287 out_o1=0.117407 out_o2=0.873533
网络的输出：step:9580 loss=0.000285 out_o1=0.117331 out_o2=0.873606
网络的输出：step:9579 loss=0.000282 out_o1=0.117255 out_o2=0.873679
网络的输出：step:9578 loss=0.000280 out_o1=0.117180 out_o2=0.873752
网络的输出：step:9577 loss=0.000277 out_o1=0.117105 out_o2=0.873825
网络的输出：step:9576 loss=0.000275 out_o1=0.117031 out_o2=0.873897
网络的输出：step:9575 loss=0.000272 out_o1=0.116957 out_o2=0.873968
网络的输出：step:9574 loss=0.000270 out_o1=0.116883 out_o2=0.874039
网络的输出：step:9573 loss=0.000268 out_o1=0.116809 out_o2=0.874110
网络的输出：step:9572 loss=0.000265 out_o1=0.116737 out_o2=0.874181
网络的输出：step:9571 loss=0.000263 out_o1=0.116664 out_o2=0.874251
网络的输出：step:9570 loss=0.000261 out_o1=0.116592 out_o2=0.874321
网络的输出：step:9569 loss=0.000258 out_o1=0.116520 out_o2=0.874390
网络的输出：step:9568 loss=0.000256 out_o1=0.116448 out_o2=0.874459
网络的输出：step:9567 loss=0.000254 out_o1=0.116377 out_o2=0.874528
网络的输出：step:9566 loss=0.000252 out_o1=0.116307 out_o2=0.874596
网络的输出：step:9565 loss=0.000249 out_o1=0.116236 out_o2=0.874664
网络的输出：step:9564 loss=0.000247 out_o1=0.116166 out_o2=0.874731
网络的输出：step:9563 loss=0.000245 out_o1=0.116097 out_o2=0.874798
网络的输出：step:9562 loss=0.000243 out_o1=0.116028 out_o2=0.874865
网络的输出：step:9561 loss=0.000241 out_o1=0.115959 out_o2=0.874932
网络的输出：step:9560 loss=0.000239 out_o1=0.115890 out_o2=0.874998
网络的输出：step:9559 loss=0.000237 out_o1=0.115822 out_o2=0.875064
网络的输出：step:9558 loss=0.000235 out_o1=0.115754 out_o2=0.875129
网络的输出：step:9557 loss=0.000233 out_o1=0.115687 out_o2=0.875194
网络的输出：step:9556 loss=0.000231 out_o1=0.115620 out_o2=0.875259
网络的输出：step:9555 loss=0.000229 out_o1=0.115553 out_o2=0.875323
网络的输出：step:9554 loss=0.000227 out_o1=0.115486 out_o2=0.875388
网络的输出：step:9553 loss=0.000225 out_o1=0.115420 out_o2=0.875451
网络的输出：step:9552 loss=0.000223 out_o1=0.115354 out_o2=0.875515
网络的输出：step:9551 loss=0.000221 out_o1=0.115289 out_o2=0.875578
网络的输出：step:9550 loss=0.000219 out_o1=0.115224 out_o2=0.875641
网络的输出：step:9549 loss=0.000217 out_o1=0.115159 out_o2=0.875703
网络的输出：step:9548 loss=0.000215 out_o1=0.115095 out_o2=0.875765
网络的输出：step:9547 loss=0.000213 out_o1=0.115031 out_o2=0.875827
网络的输出：step:9546 loss=0.000212 out_o1=0.114967 out_o2=0.875888
网络的输出：step:9545 loss=0.000210 out_o1=0.114903 out_o2=0.875950
网络的输出：step:9544 loss=0.000208 out_o1=0.114840 out_o2=0.876010
网络的输出：step:9543 loss=0.000206 out_o1=0.114777 out_o2=0.876071
网络的输出：step:9542 loss=0.000204 out_o1=0.114715 out_o2=0.876131
网络的输出：step:9541 loss=0.000203 out_o1=0.114653 out_o2=0.876191
网络的输出：step:9540 loss=0.000201 out_o1=0.114591 out_o2=0.876251
网络的输出：step:9539 loss=0.000199 out_o1=0.114529 out_o2=0.876310
网络的输出：step:9538 loss=0.000198 out_o1=0.114468 out_o2=0.876369
网络的输出：step:9537 loss=0.000196 out_o1=0.114407 out_o2=0.876428
网络的输出：step:9536 loss=0.000194 out_o1=0.114346 out_o2=0.876486
网络的输出：step:9535 loss=0.000193 out_o1=0.114286 out_o2=0.876544
网络的输出：step:9534 loss=0.000191 out_o1=0.114226 out_o2=0.876602
网络的输出：step:9533 loss=0.000189 out_o1=0.114166 out_o2=0.876660
网络的输出：step:9532 loss=0.000188 out_o1=0.114107 out_o2=0.876717
网络的输出：step:9531 loss=0.000186 out_o1=0.114047 out_o2=0.876774
网络的输出：step:9530 loss=0.000185 out_o1=0.113989 out_o2=0.876830
网络的输出：step:9529 loss=0.000183 out_o1=0.113930 out_o2=0.876887
网络的输出：step:9528 loss=0.000181 out_o1=0.113872 out_o2=0.876943
网络的输出：step:9527 loss=0.000180 out_o1=0.113814 out_o2=0.876999
网络的输出：step:9526 loss=0.000178 out_o1=0.113756 out_o2=0.877054
网络的输出：step:9525 loss=0.000177 out_o1=0.113699 out_o2=0.877109
网络的输出：step:9524 loss=0.000175 out_o1=0.113641 out_o2=0.877164
网络的输出：step:9523 los