Database Reference
In-Depth Information
...
14/11/16 14:56:11 INFO SparkContext: Job finished: mean at
StreamingModel.scala:159, took 0.09122 s
-------------------------------------------
Time: 1416142570000 ms
-------------------------------------------
MSE current batch: Model 1: 97.9475827857361; Model 2:
97.9475827857361
RMSE current batch: Model 1: 9.896847113385965; Model 2:
9.896847113385965
...
Since both models start with the same initial weight vector, we see that they both make
the same predictions on this first batch and, therefore, have the same error.
If we leave the streaming program running for a few minutes, we should eventually see
that one of the models has started converging, leading to a lower and lower error, while
the other model has tended to diverge to a poorer model due to the overly high learning
rate:
...
14/11/16 14:57:30 INFO SparkContext: Job finished: mean at
StreamingModel.scala:159, took 0.069175 s
-------------------------------------------
Time: 1416142650000 ms
-------------------------------------------
MSE current batch: Model 1: 75.54543031658632; Model 2:
10318.213926882852
RMSE current batch: Model 1: 8.691687426304878; Model 2:
101.57860959317593
...
If you leave the program running for a number of minutes, you should eventually see the
first model's error rate getting quite small:
