When I ran dpm_sovler_skipUQ.py, I found that the last layer would be retrained, but the loss would not decrease with each iteration.
It may indicate that the model cannot fit the data distribution effectively.
iter:0 loss:12148.708984375
iter:1 loss:36523.2890625
iter:2 loss:30084.55078125
iter:3 loss:15620.3291015625
iter:4 loss:34903.0390625
iter:5 loss:7580.31005859375
iter:6 loss:6508.32421875
iter:7 loss:9720.5302734375
iter:8 loss:10204.521484375
iter:9 loss:5467.26318359375
iter:10 loss:30769.896484375
iter:11 loss:8178.75927734375
iter:12 loss:39332.96484375
iter:13 loss:9098.0791015625
iter:14 loss:26085.1015625
iter:15 loss:10706.9111328125
iter:16 loss:72460.9765625
iter:17 loss:3483.43408203125
iter:18 loss:14585.998046875
iter:19 loss:6981.5703125
iter:20 loss:15868.8076171875
iter:21 loss:5725.53466796875
iter:22 loss:35333.390625
iter:23 loss:38813.34375
iter:24 loss:11802.0546875
iter:25 loss:12351.8583984375
iter:26 loss:3665.28662109375
iter:27 loss:1745.161376953125
iter:28 loss:10452.044921875
iter:29 loss:6976.3759765625
iter:30 loss:2253.290283203125
iter:31 loss:4307.98193359375
iter:32 loss:19940.427734375
iter:33 loss:5888.1376953125
iter:34 loss:8116.71875
iter:35 loss:6951.1455078125
iter:36 loss:13301.1171875
iter:37 loss:18389.734375
iter:38 loss:21819.4375
iter:39 loss:4901.2001953125
iter:40 loss:22742.7109375
iter:41 loss:4589.68359375
iter:42 loss:3254.26416015625
iter:43 loss:8937.9521484375
iter:44 loss:9177.1455078125
iter:45 loss:4575.599609375
iter:46 loss:31417.09765625
iter:47 loss:19158.794921875
iter:48 loss:9227.7109375
iter:49 loss:7116.0869140625
iter:50 loss:34749.0703125
iter:51 loss:5798.89453125
iter:52 loss:33096.7890625
iter:53 loss:21321.861328125
iter:54 loss:13240.6328125
iter:55 loss:3758.980224609375
iter:56 loss:10366.9658203125
iter:57 loss:26977.091796875
iter:58 loss:4997.310546875
iter:59 loss:22883.97265625
iter:60 loss:20039.755859375
iter:61 loss:14093.001953125
iter:62 loss:11133.85546875
iter:63 loss:15523.6201171875
iter:64 loss:36102.7734375
iter:65 loss:16064.197265625
iter:66 loss:12511.1923828125
iter:67 loss:3731.33984375
iter:68 loss:27153.359375
iter:69 loss:3833.329833984375
iter:70 loss:6349.7490234375
iter:71 loss:23716.20703125
iter:72 loss:8286.845703125
iter:73 loss:12143.0986328125
iter:74 loss:13969.279296875
iter:75 loss:9628.1298828125
iter:76 loss:6814.1689453125
iter:77 loss:20542.23828125
iter:78 loss:15336.376953125
iter:79 loss:5830.5595703125
iter:80 loss:19006.67578125
iter:81 loss:6234.8046875
iter:82 loss:6880.68505859375
iter:83 loss:25276.509765625
iter:84 loss:14096.61328125
iter:85 loss:5536.50537109375
iter:86 loss:19971.599609375
iter:87 loss:10206.591796875
iter:88 loss:23719.80078125
iter:89 loss:11952.15234375
iter:90 loss:36366.5
iter:91 loss:25210.69921875
iter:92 loss:6288.00048828125
iter:93 loss:14854.703125
iter:94 loss:9977.44140625
iter:95 loss:53876.33984375
iter:96 loss:62251.328125
iter:97 loss:30210.359375
iter:98 loss:7555.1787109375
iter:99 loss:4163.421875
iter:100 loss:38099.0078125
When I ran dpm_sovler_skipUQ.py, I found that the last layer would be retrained, but the loss would not decrease with each iteration.
It may indicate that the model cannot fit the data distribution effectively.