Execution Time0.74s

Test: TMVA-DNN-RNN-Backpropagation-Cpu (Passed)
Build: master-x86_64-fedora28-gcc8 (sft-fedora-28-1.cern.ch) on 2020-04-21 00:15:26

Test Timing: Passed
Processors1

Show Command Line
Display graphs:

Test output
Testing RNN backward pass on CPU
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 1 input = 4 state = 5 time = 3 with a fixed input and a dense layer
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.05581e-09[NON-XML-CHAR-0x1B][39m
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m3.51475e-10[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m8.68692e-10[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m1.16811e-10[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 2 input = 2 state = 1 time = 1 using a random input
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m1.60859e-10[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (absolute): [NON-XML-CHAR-0x1B][32m0[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m1.05685e-11[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 2 input = 2 state = 3 time = 1 using a random input
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m7.44079e-10[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (absolute): [NON-XML-CHAR-0x1B][32m0[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m9.25083e-11[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 3 input = 5 state = 4 time = 2 using a random input
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.66864e-09[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m2.94421e-09[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m3.19502e-10[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 3 input = 5 state = 4 time = 2 using a random input and full output
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m7.33013e-10[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m8.24819e-09[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m2.14619e-10[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 2 input = 5 state = 10 time = 4 using a random input
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m3.48226e-08[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m3.36546e-08[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.29569e-09[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 64 input = 5 state = 10 time = 5 using a random input
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.20432e-06[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.98357e-07[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m7.46946e-09[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 1 input = 5 state = 10 time = 3 with a fixed input
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m3.23943e-08[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m3.85252e-08[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.29883e-09[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 32 input = 20 state = 10 time = 4 using a random input and a dense layer
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m2.54898e-10[NON-XML-CHAR-0x1B][39m
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m5.84229e-08[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m4.46518e-08[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.00739e-09[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 8 input = 4 state = 5 time = 3 using a random input and a dense layer and full output
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.30596e-09[NON-XML-CHAR-0x1B][39m
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.03022e-09[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m6.70522e-09[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m8.24245e-11[NON-XML-CHAR-0x1B][39m
******************************************************************************************************
Testing Weight Backprop using RNN with batchsize = 2 input = 2 state = 3 time = 2 using a random input and a dense layer and an extra RNN
******************************************************************************************************
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m2.07325e-10[NON-XML-CHAR-0x1B][39m
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.24135e-09[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m4.12272e-08[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m1.20702e-10[NON-XML-CHAR-0x1B][39m
Testing weight input gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m7.60225e-10[NON-XML-CHAR-0x1B][39m
Testing weight state gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][33m1.56456e-09[NON-XML-CHAR-0x1B][39m
Testing bias gradients:      maximum error (relative): [NON-XML-CHAR-0x1B][32m7.31936e-11[NON-XML-CHAR-0x1B][39m
Info in <testRecurrentPropagationCpu>: All tests passed !!!