Open starkdm opened 4 years ago
I solve the problem of regression. 5 values are predicted from the 20 previous values of 2 exogenous variables. All values in data and labels are in the range from 0 to 1. What could be the reason for so large error?
model.prototxt
name: "MLP" layer { name: "input" type: "Input" top: "data" top: "labels" input_param { shape: { dim: 5000 dim: 2 dim: 1 dim: 20 } shape: { dim: 5000 dim: 1 dim: 1 dim: 5 } } include { phase: TRAIN } } layer { name: "input" type: "Input" top: "data" input_param { shape: { dim: 100 dim: 2 dim: 1 dim: 20 } } include { phase: TEST } } layer { name: "fc1" type: "InnerProduct" bottom: "data" top: "fc1" inner_product_param { num_output: 50 weight_filler { type: "msra" } bias_filler { type: "constant" value: 1 } } } layer { name: "s1" type: "Sigmoid" bottom: "fc1" top: "s1" } layer { name: "fc2" type: "InnerProduct" bottom: "s1" top: "fc2" inner_product_param { num_output: 5 weight_filler { type: "msra" } bias_filler { type: "constant" value: 1 } } } layer { name: "output" type: "Sigmoid" bottom: "fc2" top: "s2" } layer { name: "loss" type: "EuclideanLoss" bottom: "s2" bottom: "labels" top: "loss" include { phase: TRAIN } }
solver.prototxt
net: "model.prototxt" base_lr: 0.01 lr_policy: "fixed" max_iter: 500 momentum: 0.45 solver_mode: CPU display: 10 snapshot_prefix: "mlp" type: "Adam"
Output
I1005 21:38:24.438515 18200 solver.cpp:218] Iteration 0 (0 iter/s, 0.033s/10 iters), loss = 2.1907e+131 I1005 21:38:24.438515 18200 solver.cpp:237] Train net output #0: loss = 2.1907e+131 (* 1 = 2.1907e+131 loss) I1005 21:38:24.440623 18200 sgd_solver.cpp:105] Iteration 0, lr = 0.01 I1005 21:38:24.689244 18200 solver.cpp:218] Iteration 10 (40 iter/s, 0.25s/10 iters), loss = 2.1907e+131 I1005 21:38:24.690243 18200 solver.cpp:237] Train net output #0: loss = 2.1907e+131 (* 1 = 2.1907e+131 loss) I1005 21:38:24.691956 18200 sgd_solver.cpp:105] Iteration 10, lr = 0.01 I1005 21:38:25.043009 18200 solver.cpp:218] Iteration 20 (28.4091 iter/s, 0.352s/10 iters), loss = 2.1907e+131 I1005 21:38:25.044751 18200 solver.cpp:237] Train net output #0: loss = 2.1907e+131 (* 1 = 2.1907e+131 loss) I1005 21:38:25.045539 18200 sgd_solver.cpp:105] Iteration 20, lr = 0.01 I1005 21:38:25.680462 18200 solver.cpp:218] Iteration 30 (15.7233 iter/s, 0.636s/10 iters), loss = 2.1907e+131 I1005 21:38:25.699206 18200 solver.cpp:237] Train net output #0: loss = 2.1907e+131 (* 1 = 2.1907e+131 loss)
Operating system: Windows 10
I solve the problem of regression. 5 values are predicted from the 20 previous values of 2 exogenous variables. All values in data and labels are in the range from 0 to 1. What could be the reason for so large error?
model.prototxt
solver.prototxt
Output
Operating system: Windows 10