cheng-li / pyramid Goto Github PK

[chengli@fiji11 pyramid-0.1.0]$ python visualization/visualizer.py /huge1/people/chengli/projects/pyramid/archives/app3/ohsumed_20000/8/reports/train_reports
Traceback (most recent call last):
File "visualization/visualizer.py", line 1624, in
main()
File "visualization/visualizer.py", line 1602, in main
f1 = open(directoryName + configName, 'r')
IOError: [Errno 2] No such file or directory: '/huge1/people/chengli/projects/pyramid/archives/app3/ohsumed_20000/8/reports/data_config.json'
[chengli@fiji11 pyramid-0.1.0]$ python visualization/visualizer.py /huge1/people/chengli/projects/pyramid/archives/app3/ohsumed_20000/8/reports/train_reports/
Json:/huge1/people/chengli/projects/pyramid/archives/app3/ohsumed_20000/8/reports/train_reports/report_1.json load successfully.

When I run CBM with 10K training samples, 10K features and 12 labels (70 label sets) , it works fine. But when I increase the number of training samples to 150K, it throws the following error.. Most of the features are real valued. The dataset is around 95% sparse. The code uses around 50GB memory out of 128GB.

Exception in thread "main" java.lang.reflect.InvocationTargetException
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:497)
at edu.neu.ccs.pyramid.application.AppLauncher.invokeMain(AppLauncher.java:72)
at edu.neu.ccs.pyramid.application.AppLauncher.launch(AppLauncher.java:39)
at edu.neu.ccs.pyramid.application.AppLauncher.main(AppLauncher.java:24)
Caused by: java.lang.OutOfMemoryError: GC overhead limit exceeded
at java.lang.String.split(String.java:2338)
at java.lang.String.split(String.java:2410)
at edu.neu.ccs.pyramid.dataset.TRECFormat.fillMultiLabelClfDataSet(TRECFormat.java:332)
at edu.neu.ccs.pyramid.dataset.TRECFormat.loadMultiLabelClfDataSet(TRECFormat.java:159)
at edu.neu.ccs.pyramid.dataset.TRECFormat.loadMultiLabelClfDataSet(TRECFormat.java:106)
at edu.neu.ccs.pyramid.application.App5.train(App5.java:58)
at edu.neu.ccs.pyramid.application.App5.main(App5.java:38)
... 7 more

From web, what I got to know is, this error is raised when most of the run time is consumed in Garbage collection and the progress become too slow. Java shows this error as it suspects the program may never finish. As per the paper, CBM is able to handle TMC2007 dataset which is relatively large. So I am hoping there is a solution to this issue.

Any idea how to fix this issue?

[INFO] pyramid 0.12.6 ..................................... SUCCESS [ 0.238 s]
[INFO] phrase-count-plugin 1.0 ............................ SUCCESS [ 4.893 s]
[INFO] pyramid 0.12.6 ..................................... FAILURE [ 15.589 s]
[INFO] ------------------------------------------------------------------------
[INFO] BUILD FAILURE
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 20.851 s
[INFO] Finished at: 2020-03-07T20:08:16-05:00
[INFO] ------------------------------------------------------------------------
[ERROR] Failed to execute goal org.apache.maven.plugins:maven-assembly-plugin:2.5.5:single (default) on project pyramid: Execution default of goal org.apache.maven.plugins:maven-assembly-plugin:2.5.5:single failed: user id '16779829' is too big ( > 2097151 ). -> [Help 1]

I have attached the error message. Do you know how to deal with it?

In the output folder, model files are being saved. What do the model files contain? How to use them?

How can we get the predictions for the test samples?

Hi @cheng-li ,
Is it possible to save the confidence scores of the predictions i.e. for a test data point when a label subset is predicted what is the confidence (or probability) of that label subset. This will help us to identify the hard-to-classify data points.

Thanks in advance.