以上代码给我返回了六个权重数组。我理解weights[0]是一个2×20的连接集合，但weights[1]只是数组中的20个成员，可能是每个神经元一个。但我以为权重是连接之间的，所以这到底是什么？
结果如下：
Model: "sequential_11"_________________________________________________________________Layer (type) Output Shape Param # =================================================================flatten_11 (Flatten) (None, 2) 0 _________________________________________________________________dense_33 (Dense) (None, 20) 60 _________________________________________________________________dense_34 (Dense) (None, 20) 420 _________________________________________________________________dense_35 (Dense) (None, 1) 21 =================================================================Total params: 501Trainable params: 501Non-trainable params: 0_________________________________________________________________None<tf.Variable 'dense_33/kernel:0' shape=(2, 20) dtype=float32, numpy=array([[-0.08592772, -0.4262397 , 0.32593143, 0.40175033, -0.11370629, 0.29291457, -0.33887625, 0.09051579, -0.11669007, 0.15766495, -0.03898111, 0.47355425, 0.4038219 , -0.16283795, -0.52166206, 0.08563775, 0.10119641, 0.35014063, -0.29258126, 0.11257637], [ 0.32310146, -0.00564504, -0.39950165, 0.3422314 , -0.1736508 , -0.15470237, 0.03384084, -0.50031585, 0.17582124, -0.20669848, 0.38023835, 0.45190394, 0.22054166, -0.3583283 , -0.31276733, -0.42144495, 0.05265975, 0.28793246, -0.12343103, -0.52028173]], dtype=float32)><tf.Variable 'dense_33/bias:0' shape=(20,) dtype=float32, numpy=array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)><tf.Variable 'dense_34/kernel:0' shape=(20, 20) dtype=float32, numpy=array([[ 0.3640272 , -0.1974021 , -0.18908244, 0.36629468, -0.19073088, 0.3649723 , 0.20782089, 0.3469664 , 0.07804716, 0.31226724, -0.29463375, 0.1686036 , -0.20033775, 0.30975252, 0.02273259, -0.20266692, -0.27747104, 0.20879221, -0.01603091, -0.18493965], [ 0.35550284, -0.05491427, 0.17644823, -0.33692163, -0.19445068, -0.04921806, -0.32028157, -0.25789088, -0.2392104 , 0.2923655 , -0.13487688, 0.07437733, -0.08911327, -0.05451488, 0.26065683, -0.15244626, 0.16566378, 0.35106057, 0.2891826 , -0.31660333], [ 0.27196383, 0.06386155, -0.00255299, -0.08238894, 0.10117772, -0.1466422 , -0.3506349 , -0.17026383, 0.37896347, 0.38407606, 0.11256388, 0.04563713, -0.37850204, -0.02221993, 0.10061872, 0.3318779 , 0.02184781, -0.2571288 , -0.3272443 , -0.30387318], [-0.30832642, -0.26533905, -0.17655498, -0.17255014, -0.25053996, -0.23460795, 0.35648036, 0.25864762, 0.2033844 , -0.1375912 , -0.21034713, 0.23340106, 0.07856712, 0.06940717, 0.3642025 , -0.10192937, 0.3174355 , -0.16317467, 0.12438256, 0.34762514], [ 0.29233575, -0.23410663, 0.17361456, -0.35682717, 0.29746616, -0.13978545, 0.19865829, -0.32398444, 0.27369195, 0.29182 , 0.28845608, 0.3714251 , -0.00226831, -0.11382625, -0.03799275, -0.38083866, 0.13849735, -0.17412637, 0.30680603, -0.32791764], [-0.36856598, -0.01098448, 0.2209788 , -0.01641467, 0.36460286, 0.3742503 , -0.07144001, -0.32689905, -0.2800351 , -0.26420033, -0.3203124 , 0.22266299, 0.05407029, 0.20716977, 0.23186374, 0.34451336, -0.3665755 , -0.08111835, 0.02044231, 0.22657269], [-0.18087737, -0.09122089, -0.3162348 , -0.01350608, 0.2994557 , 0.00759923, 0.07653233, -0.11245179, -0.06106046, 0.09489083, -0.34051555, -0.0210776 , -0.3720226 , -0.08034962, -0.3628871 , -0.08755568, 0.13865143, -0.13755408, 0.18153298, 0.23439962], [-0.36453527, -0.3077588 , 0.06971669, 0.23991793, -0.32902858, -0.1256682 , -0.37355578, -0.22176625, -0.06080669, 0.12455881, 0.02237046, 0.21177506, 0.05803809, -0.07626435, -0.36375207, 0.13273174, 0.15075874, -0.18664922, 0.20256019, 0.17832053], [-0.09238327, 0.03065437, 0.04975492, 0.03068706, -0.01132107, 0.04134732, 0.2726786 , 0.09169459, 0.16609359, -0.26199952, 0.34235936, -0.3293307 , -0.2625829 , 0.05643666, -0.19363837, -0.09321746, 0.15029383, 0.1271655 , -0.13643244, -0.1260187 ], [-0.10736805, 0.08597881, -0.28592098, 0.32719833, 0.25863254, -0.35738683, 0.28420174, 0.07898697, 0.12083912, 0.24187142, 0.20364356, 0.16368687, 0.3372751 , -0.11902198, 0.29610634, -0.26228833, 0.26691556, 0.02676412, 0.20875496, 0.2742722 ], [-0.01111042, 0.01864234, -0.3684872 , 0.25593793, 0.05572906, -0.27395982, -0.18536313, -0.28665859, 0.33866453, -0.04277194, 0.31874043, 0.17231691, 0.26513118, 0.2841534 , 0.38413507, 0.32093495, -0.1821885 , 0.3448484 , 0.06886706, 0.05471361], [ 0.27029324, 0.17785454, -0.3417698 , -0.18585834, 0.13658857, 0.25487036, -0.34464136, -0.31934893, -0.07558686, 0.12984264, -0.12386304, 0.33101034, 0.2395941 , -0.35495222, -0.38362566, 0.02023152, -0.38369113, 0.10231277, -0.00923318, -0.2564116 ], [ 0.3738134 , -0.13709581, -0.02727005, -0.38571945, -0.17950383, -0.08438393, -0.35748094, -0.02563897, -0.26492482, -0.3148442 , 0.27745587, 0.01215285, 0.0338603 , 0.22927964, -0.26310933, 0.17490405, 0.15125847, 0.33357888, -0.10504535, 0.09216848], [-0.18665901, 0.14127249, -0.31025392, 0.3109604 , 0.11353964, 0.1544854 , -0.0628956 , 0.2526992 , -0.38535342, -0.35054773, 0.3639174 , -0.22744954, 0.2787813 , 0.25469422, -0.24284746, 0.2586198 , 0.38151866, 0.14534372, -0.07336038, 0.35205972], [-0.0222263 , -0.09051144, 0.19810867, 0.2596696 , 0.16493648, -0.15432249, -0.12816939, 0.26651537, 0.33925128, 0.24480599, -0.20886998, -0.23604779, 0.35640693, -0.1257923 , -0.3385602 , 0.37019014, -0.34767368, -0.20191407, 0.05838048, -0.3322008 ], [-0.22849075, 0.31127506, -0.1032331 , 0.03278631, -0.3802262 , 0.06519806, -0.10763076, -0.23816115, 0.29874003, -0.17749721, -0.10582674, -0.03064901, 0.18550068, 0.08624834, 0.09579298, -0.305739 , 0.00272122, 0.14033073, 0.22830683, -0.17147864], [-0.23749031, -0.36036015, -0.15639098, 0.16943222, 0.33908015, -0.18797807, -0.31251115, 0.13584453, -0.10717931, -0.11736256, 0.17281443, 0.1897279 , -0.35898107, 0.21381551, 0.3051238 , 0.12489098, 0.29044586, -0.20301346, -0.25790715, 0.04153055], [-0.25941753, -0.3507824 , 0.34750968, -0.04910356, -0.1914334 , -0.22343925, 0.3420688 , 0.38251758, -0.09309632, 0.3546936 , -0.22427556, -0.24499758, 0.00074324, -0.06633586, 0.1922136 , 0.11927372, -0.19837731, -0.23528719, -0.26004478, -0.24688683], [-0.03148285, -0.32766464, -0.2530514 , 0.1765365 , 0.26583946, -0.18146862, -0.20307828, -0.07899943, -0.10167924, -0.05031338, 0.03324467, -0.27283487, -0.3672278 , -0.24607424, -0.15097658, 0.18689764, -0.32162574, 0.10529301, -0.2671068 , -0.29016626], [ 0.23749584, -0.10357189, 0.03281826, 0.30171496, 0.3568563 , -0.27596533, 0.3714081 , 0.04653817, 0.14261234, -0.18491131, 0.3152057 , 0.23654068, 0.07070702, 0.0457052 , -0.17505457, 0.15374076, -0.03659964, 0.2212556 , -0.05467528, 0.160887 ]], dtype=float32)><tf.Variable 'dense_34/bias:0' shape=(20,) dtype=float32, numpy=array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)><tf.Variable 'dense_35/kernel:0' shape=(20, 1) dtype=float32, numpy=array([[ 0.44639164], [ 0.46424013], [-0.36026257], [-0.19160783], [-0.236644 ], [ 0.36841106], [ 0.5083434 ], [-0.00797582], [ 0.25151885], [-0.51940155], [-0.03726539], [-0.15949944], [-0.2284751 ], [ 0.4611426 ], [ 0.2685169 ], [ 0.1900658 ], [-0.4574982 ], [ 0.22935611], [-0.46032292], [-0.28261 ]], dtype=float32)><tf.Variable 'dense_35/bias:0' shape=(1,) dtype=float32, numpy=array([0.], dtype=float32)>回答：
model.weights 返回所有权重。您的理解是正确的。 weights[0] 是一个2×20的连接集合，weights[1] 是对应偏置的权重。因为您使用的是 tf.keras.layers.Dense，默认情况下 use_bias=True。所以，weight[1] 指的是第一层Dense层的偏置权重。其他Dense层的情况也一样。尝试将其中任何一层设置为 use_bias=False，您会注意到差异。例如 –
model = keras.Sequential([ keras.layers.Flatten(input_shape=(2,)), keras.layers.Dense(20, activation=tf.nn.relu, use_bias=False), keras.layers.Dense(20, activation=tf.nn.relu), keras.layers.Dense(1) ]) len(model.weights) 这会返回一个包含5个元素的列表。它忽略了第一层的偏置权重。
