ss检测中的模型训练方式

[Arktische]

你好，我在本地试了一下ss检测的代码，最后得到的效果不是特别好，包数量不够时，loss也不稳定，所以想针对本地环境自己训练一下模型。请问代码中的矩阵是如何计算的

detect_shadowsocks.py

F = np.array([549.85092876, 1.84631161, 7.24625686, 5.0383797, 6.99853457,
              1.50255445, 4.36626127, 4.95866935, 3.45665574, 4.55042411,
              2.32397034, 3.42282341, 1.97602002, 3.23253569, 3.45725634,
              3.71534851, 4.62689791, 3.8438733, 1.74597793, 1.36485671,
              4.27549965, 1.29409605, 3.6081702, 2.18414709, 0.80476456,
              2.22344187, 3.6635435, 0.66864245, 3.01349593, 1.60893295,
              1.93159699, 3.60502456, 1.51572153, 3.20920403, 2.96585503,
              0.59570594, 3.6369488, 1.91328589, 2.45297406, 2.87422478,
              3.62653438, 2.61505699, 0.93285202, 2.61087341, 1.0433738,
              2.81240329, 2.69193529, 2.7447757, 3.50277366, 1.91170636,
              1.89645859, 3.37464325, 3.87443489, 2.88008518, 3.09536179,
              4.17115956, 2.26157271, 4.12317685, 2.33176768, 3.05101856,
              3.93357842, 2.7372064, 8.06253887, 3.46319319, 2.9656001,
              1.70803287, 0.67326659, 6.17000221, 3.37711465, 4.16827182,
              4.22913834, 3.33330792, 1.6061819, 3.9014807, 2.45484721,
              3.18764991, 4.50298305, 4.84597927, 3.87708731, 1.43924832,
              5.53988964, 5.20902506, 3.43908033, 5.42388361, 5.51823775,
              2.60303553, 2.68981832, 1.97306915, 4.75873684, 4.62081544,
              7.77197192, 7.52523643, 4.95305522, 5.05568544, 3.79383539,
              10.80279768, 12.90162243, 15.52545897, 7.25273777, 6.2725986])

H = np.array([4.75563571e-01, 8.88488127e-03, 5.70964833e-02, 1.67959122e-02,
              8.80072137e-03, 9.87075443e-03, 5.63871356e-03, 1.19747520e-02,
              5.84310189e-03, 3.06582507e-03, 7.08145476e-03, 8.71656147e-03,
              4.09978960e-03, 3.22212203e-03, 3.07784791e-03, 2.93357379e-03,
              3.78719567e-03, 2.39254584e-03, 2.18815750e-03, 3.00571085e-03,
              1.64712955e-03, 1.13014728e-03, 5.56657650e-03, 1.55094680e-03,
              4.97745717e-03, 1.05801022e-03, 1.10610159e-03, 1.25037571e-03,
              1.51487827e-03, 1.63510670e-03, 7.09347761e-04, 1.21430718e-03,
              1.53892396e-03, 1.03396453e-03, 1.03396453e-03, 8.65644725e-04,
              8.89690412e-04, 1.50285543e-03, 1.35858130e-03, 6.25187857e-04,
              8.17553351e-04, 6.37210700e-04, 9.73850316e-04, 6.97324917e-04,
              3.96753832e-04, 1.09407875e-03, 5.77096483e-04, 2.88548242e-04,
              1.26239856e-03, 3.17006312e-01])

[ytgui]

oh 大兄弟你吓我一跳，我这个仓库已经废弃了，现在用来日常记笔记哈哈~

请注意代码中的这两行：

square_loss_freq = np.abs(freq_feature - F)
square_loss_hist = np.abs(hist_feature - H)

因此 F 与 H 即是 freq_feature 与 hist_feature 的目标值

所以，

1. 采集一定量的已知 http(s) 数据，计算 F 与 H
2. 对未知数据计算 freq_feature hist_feature，与 1 比较

以上，谢邀 @Arktische