retraining model problem about deeplearning4j-examples HOT 9 CLOSED

plus, the arguments are:
coefficients_test.bin conf_test.json 150 150
in both Train and Retrain

from deeplearning4j-examples.

So: I'm unable to reproduce this on 0.4-rc3.7 or current DL4J master, running directly in IntelliJ.
In both cases I get 20 iterations.
Can you confirm that you are running the latest version (0.4-rc3.7)?

Weird issue though is the difference between 0.4-rc3.7 and current master (see gist below). I will investigate this more tomorrow. I haven't seen any issues with any of our other examples, so I'm not sure what's going on there.
https://gist.github.com/AlexDBlack/cb8e74a37255ede96885

from deeplearning4j-examples.

OK i will update the jar file to the latest version .. lol and try it again ...
thx!

from deeplearning4j-examples.

hi @AlexDBlack
I've updated to:
deeplearning4j-core-0.4-rc3.7.jar
nd4j-api-0.4-rc3.6.jar
nd4j-bytebuddy-0.4-rc3.6.jar
nd4j-x86-0.4-rc3.6.jar

But the problem is that it seems like validation doesn't work
that is, I modify Retrain.scala so that it also validates the train data
the validation result turns out to be weird

Rerain.scala:
Retrain.txt

and here's the result:(run Train then Retrain)

I also investigated to the output ndarry of reloadmod.output() and found out that the outputs are all the same as: 0.416384, 0.333089, 0.250526
(but I didn't show in Retrain.scala and the images here)

Thx for any help!!

from deeplearning4j-examples.

Just a side note: please use the same rc3.7 for nd4j.

from deeplearning4j-examples.

I've tried to do this by adding unmanaged dependencies ... that seems a bad idea :)
Instead, it works fine just using sbt with build.sbt like:

libraryDependencies ++= Seq(
"org.deeplearning4j" % "deeplearning4j-core" % "0.4-rc3.6",
"org.jblas" % "jblas" % "1.2.4",
"org.nd4j" % "nd4j-jblas" % "0.4-rc3.5"
)

thx for any suggestion from this platform and gitter

from deeplearning4j-examples.

Any update on this?

from deeplearning4j-examples.

nope, everything works well on core3.6 with managed dependencies

thx for your help

from deeplearning4j-examples.

Cool. Make sure to stay up to date. We're on rc3.8 now. Thanks!

from deeplearning4j-examples.