2017-03-12 16:33:46 INFO DistriOptimizer$:243 - [Epoch 15 59952/60000][Iteration 74997][Wall Clock 3175.96207633s] Train 12 in 0.021189853seconds. Throughput is 566.308789400285 records/second. Loss is 0.0012443860759958625. Current learning rate is 0.05. 2017-03-12 16:33:46 INFO DistriOptimizer$:243 - [Epoch 15 59964/60000][Iteration 74998][Wall Clock 3175.983266183s] Train 12 in 0.025899406seconds. Throughput is 463.33108952382923 records/second. Loss is 0.0018258094787597656. Current learning rate is 0.05. 2017-03-12 16:33:47 INFO DistriOptimizer$:243 - [Epoch 15 59976/60000][Iteration 74999][Wall Clock 3176.009165589s] Train 12 in 0.023225649seconds. Throughput is 516.6701692598557 records/second. Loss is 0.001348257064819336. Current learning rate is 0.05. 2017-03-12 16:33:47 INFO DistriOptimizer$:243 - [Epoch 15 59988/60000][Iteration 75000][Wall Clock 3176.032391238s] Train 12 in 0.023787667seconds. Throughput is 504.4630900541865 records/second. Loss is 5.4836273193359375E-5. Current learning rate is 0.05. 2017-03-12 16:33:47 INFO DistriOptimizer$:286 - [Epoch 15 59988/60000][Iteration 75000][Wall Clock 3176.032391238s] Epoch finished. Wall clock time is 3193550.324218ms 2017-03-12 16:33:47 INFO DistriOptimizer$:431 - [Wall Clock 3193.550324218s] Validate model... 2017-03-12 16:34:02 INFO DistriOptimizer$:471 - top1 accuracy is Accuracy(correct: 9889, count: 10000, accuracy: 0.9889) [Wall Clock 3193.550324218s] Save model to ./model
$ ./dist/bin/bigdl.sh -- spark-submit --master local[1] \ --class com.intel.analytics.bigdl.models.lenet.Test \ ./dist/lib/bigdl-0.1.0-SNAPSHOT-jar-with-dependencies.jar \ -f mnist/ --model ./model/model.75001 --nodeNumber 1 --core 1 --env spark -b 1 Set engine type to mkl_blas 2017-03-12 16:50:19 INFO ThreadPool$:87 - Set mkl threads to 1 on thread 1 2017-03-12 16:50:19 INFO ThreadPool$:87 - Set mkl threads to 1 on thread 1 2017-03-12 16:50:21 INFO ThreadPool$:87 - Set mkl threads to 1 on thread 1 2017-03-12 16:50:22 WARN Utils$:36 - Warning: for better training speed, total batch size(1) is recommended to be at least two times of node number(1) * core number(1), please tune your batch size accordingly 2017-03-12 16:50:22 INFO DistriValidator$:50 - model thread pool size is 1 top1 accuracy is Accuracy(correct: 9887, count: 10000, accuracy: 0.9887)