"org.bytedeco.javacv.FrameRecorder$Exception: No video output stream" while running ALE.java example

[anis016]

@saudet

Hello ! I am trying to run ALE.java in org.deeplearning4j.examples.rl4j package .. but, getting this error. Please check the gist !

https://gist.github.com/anis016/6e5a49cc40776a410bb15d7fa0d09bf4

Moreover, please share with me the correct ROM file for "pong.bin". Getting an warning given below.

Game console created:
ROM file: pong.bin
Cart Name: Video Olympics (1978) (Atari)
Cart MD5: 60e0ea3cbe0913d39803477945e9e5ec
Display Format: AUTO-DETECT ==> NTSC
ROM Size: 2048
Bankswitch Type: AUTO-DETECT ==> 2K

WARNING: Possibly unsupported ROM: mismatched MD5.
Cartridge_MD5: 60e0ea3cbe0913d39803477945e9e5ec
Cartridge_name: Video Olympics (1978) (Atari)

[saudet]

That looks like the right bin. What do you get running the sample code on the following page?
https://github.com/bytedeco/javacpp-presets/tree/master/ale#sample-usage

[anis016]

I ran the sample code in the link, and this is working fine ! I can see the GUI window (ALE Viz) where the "pong" is played and also I could see the printing of the episodes...

======================================================================
A.L.E: Arcade Learning Environment (version 0.6.0)
[Powered by Stella]
Use -help for help screen.
Warning: couldn't load settings file: ./ale.cfg
Game console created:
ROM file: /home/anis/Java_Workspace/SharedLibraryInterface/src/main/resources/pong.bin
Cart Name: Video Olympics (1978) (Atari)
Cart MD5: 60e0ea3cbe0913d39803477945e9e5ec
Display Format: AUTO-DETECT ==> NTSC
ROM Size: 2048
Bankswitch Type: AUTO-DETECT ==> 2K

Screen Display Active. [Manual Control Mode] 'm' [Slowdown] 'a' [Speedup] 's' [VolumeDown] '[' [VolumeUp] ']'.

WARNING: Possibly unsupported ROM: mismatched MD5.
Cartridge_MD5: 60e0ea3cbe0913d39803477945e9e5ec
Cartridge_name: Video Olympics (1978) (Atari)

Running ROM file...
Random seed is 123
Episode 0 ended with score: -21.0
Episode 1 ended with score: -21.0
Episode 2 ended with score: -20.0
Episode 3 ended with score: -21.0
.
.

[saudet]

Ok, good!

There's been a couple of things fixed since DL4J 0.9.1 though, so could you try with 0.9.2-SNAPSHOT and let me know if this still happens? Thanks!

[anis016]

Hello @saudet !
It took a while to build everything from scratch...

1. I built in this way for 0.9.2-SNAPSHOT: https://gist.github.com/anis016/0f9c759e57dd641cd18e67516475dbed

2. I created a new module "rl4jexample" in the "rl4j" and copied the ALE.java in here. Pom: https://gist.github.com/anis016/f70e5a3e61d2c7ba38f72fe602d0957e

3. The error: https://gist.github.com/anis016/81c216265526a7fea1f613741fdd2211

Please advise further !

[saudet]

Ah, you didn't have to build everything from scratch, binaries are available here:
https://deeplearning4j.org/snapshots

In any case, thanks for the log! It makes it clear the issue is this:

[libx264 @ 0x7f2f09132580] [Eval @ 0x7f2f11e71bf0] Invalid chars '.0' at the end of expression '30.0'
[libx264 @ 0x7f2f09132580] Unable to parse option value "30.0"
[libx264 @ 0x7f2f09132580] Error setting option crf to value 30.0.
org.bytedeco.javacv.FrameRecorder$Exception: avcodec_open2() error -22: Could not open video codec.

Looks like an issue with JavaCV. I should be able to fix this today...

[saudet]

Actually, I'm unable to reproduce this issue. "crf" is a float value, and setting something like "10.5" works just fine on my machine. Could you try to run the following test and let me know what happens?
https://github.com/bytedeco/javacv/blob/master/platform/src/test/java/org/bytedeco/javacv/FrameGrabberTest.java#L266

[anis016]

Hello @saudet
Sorry for the late reply. I am only able to run dl4j, when I have access to a faster PC. However,

1. I am not able to run the Testcase you mentioned. I have added the screenshot. There's several symbols it can't find. Please check and advice.
   

2. Previously you mentioned, "you didn't have to build everything from scratch, binaries are available here: https://deeplearning4j.org/snapshots" I was not able to figure it out properly to run the 0.9.2 snapshot. could you please share with me a pom file ? or maybe a small project where ALE.java is running ? as I actually need to run only "org.deeplearning4j.examples.rl4j" package "ALE.java" ! And, since, it has no problem in your machine, I guess somewhere I am doing a bit wrong.

Thank you

[saudet]

You can find a sample pom.xml file for 0.9.2-SNAPSHOT on this branch:
https://github.com/deeplearning4j/dl4j-examples/tree/sa_snapshot

It looks like your IDE has trouble loading artifacts from the snapshot server though, that's why JavaCV isn't working there. You'll need to figure out why that is...

[anis016]

hello @saudet

there's some update... now I am getting this error after trying in another PC... please have a look. we tried in the master branch.

12:55:18.204 [main] ERROR org.deeplearning4j.rl4j.learning.sync.SyncLearning - Training failed.
java.lang.RuntimeException: Non-existent compression algorithm requested: [FLOAT]
	at org.nd4j.linalg.compression.BasicNDArrayCompressor.decompressi(BasicNDArrayCompressor.java:250)
	at org.nd4j.linalg.api.ops.executioner.DefaultOpExecutioner.validateDataType(DefaultOpExecutioner.java:541)
	at org.nd4j.linalg.cpu.nativecpu.ops.NativeOpExecutioner.exec(NativeOpExecutioner.java:680)
	at org.nd4j.linalg.cpu.nativecpu.ops.NativeOpExecutioner.exec(NativeOpExecutioner.java:119)
	at org.nd4j.linalg.api.ndarray.BaseNDArray.assign(BaseNDArray.java:1274)
	at org.nd4j.linalg.api.ndarray.BaseNDArray.put(BaseNDArray.java:2297)
	at org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.QLearningDiscrete.setTarget(QLearningDiscrete.java:200)
	at org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.QLearningDiscrete.trainStep(QLearningDiscrete.java:162)
	at org.deeplearning4j.rl4j.learning.sync.qlearning.QLearning.trainEpoch(QLearning.java:93)
	at org.deeplearning4j.rl4j.learning.sync.SyncLearning.train(SyncLearning.java:38)
	at com.packt.JavaDL.ReinforcmentLearning.ALE.main(ALE.java:66)
java.lang.RuntimeException: Non-existent compression algorithm requested: [FLOAT]
	at org.nd4j.linalg.compression.BasicNDArrayCompressor.decompressi(BasicNDArrayCompressor.java:250)
	at org.nd4j.linalg.api.ops.executioner.DefaultOpExecutioner.validateDataType(DefaultOpExecutioner.java:541)
	at org.nd4j.linalg.cpu.nativecpu.ops.NativeOpExecutioner.exec(NativeOpExecutioner.java:680)
	at org.nd4j.linalg.cpu.nativecpu.ops.NativeOpExecutioner.exec(NativeOpExecutioner.java:119)
	at org.nd4j.linalg.api.ndarray.BaseNDArray.assign(BaseNDArray.java:1274)
	at org.nd4j.linalg.api.ndarray.BaseNDArray.put(BaseNDArray.java:2297)
	at org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.QLearningDiscrete.setTarget(QLearningDiscrete.java:200)
	at org.deeplearning4j.rl4j.learning.sync.qlearning.discrete.QLearningDiscrete.trainStep(QLearningDiscrete.java:162)
	at org.deeplearning4j.rl4j.learning.sync.qlearning.QLearning.trainEpoch(QLearning.java:93)
	at org.deeplearning4j.rl4j.learning.sync.SyncLearning.train(SyncLearning.java:38)
	at com.packt.JavaDL.ReinforcmentLearning.ALE.main(ALE.java:66)

gist: https://gist.github.com/rezacsedu/46281a884179c17123516e8114e058cd

[saudet]

@raver119 I'm also getting that stack trace on my machine. Do you know what changed with respect to the compression algorithms?

[raver119]

Nothing changed AFAIK, but i'm seeing the same issue in nd4j-tests, so i'm going to check & fix that once i'm done with cnn gradients problem.

[saudet]

ok, so it is an issue with compression, good to know!

[saudet]

Confirmed fixed, thanks to @raver119!

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.