Beginner questions

[Emmanuel]

Hello.

I am master student and i am supposed to use rnnlib for my master thesis. However i am encountering difficulties to understand how to use this.

I have created nc files. Now i would like to :
-train the neural network
-save the trained network
-operate a classification on new nc files with the saved network

I found some save and display command but i don't understand how to use them.

Thanks for your help.

Best ,
Emmanuel

[Rania MAALEJ]

Help please!

[Adnan]

Hello, I'm a new user and have been using rnnlib for a year now. The reply may be helpful for new comers. Thanks Emmanuel for your question.
To run rnnlib, you can issue this command on linux terminal:
rnnlib YourConfigFile.config

If you want to save log files, you can write:
rnnlib --autosave=true YourConfigFile.config

Sample config files can be found in the example folder of rnnlib installation.

This command will save four files:
for transcription taks:
1. xyz.best_ctcError.save (your best network w.r.t. ctc Error)
2. xyz.best_labelError.save (your best network w.r.t. label Error)
3. xyz.last.save (Last saved network)
4. xyz.log (training log file)

Similar files will be created when working with other tasks.

Testing: There is a file called error_test.sh which one can use for testing your network. The command is:
./error_test [options] xyz.best_ctc/labelError.save test_ncFile.nc
This command will give a summary at the end about the accuracy of the network.

Other commands in the utils directory can be explored by simply writing them on terminal. They will show the syntax automatically.

[Darkwind]

Hi Adnan,

Just the like the rest of the users who asked, I am beginner at rnnlib.
In this answer, can you elaborate how the test_ncFile.nc looks? and how is it created? If it depends on case to case, can you giv me an example?
Excuse me for asking so many questions but the answer will really help beginners like me.

Thanks a lot

[Darkwind]

I figure it out.
For others, using the setup already provided and reducing the training set into two differen sets, I ran the python files(for eg. arabic_online.py) twice to create two diff netcdf files and used one to train and ./error_test, as you mentioned, above on the second one for testing.

[Bernard]

Adnan,
Do you also have an answer for the question "operate a classification on new nc files with the saved network"?
The documentation tells a lot about how to generate a network with rnnlib, but not how to make use of such a network, once generated.
Regards,
Bernard

[Simon]

Hi Bernard,

There are several scripts you can use in the "utils" folder.

For example, with dump_sequence_variables.sh you can use an already trained network to classify new nc files.

The command is :

path/to/dump_sequence_variables.sh bestTrainedNetwork.save dataFile.nc X

X is the number of the sequence you want to classify.

I think there are some other options, but you have to look into the script to guess what they are.

Regards,
Simon

[Magic Yang]

Hello,

I am a student and study RNN by myself.Like Simon said, we can use "utils" folder and more detils in README.
-dump_sequence_variables.sh:writes to file all the internal variables(vativations, delta terms etc. )of the network while processing a single sequence

[Rania MAALEJ]

[Rania MAALEJ]

[sana khamekhem]

Hello,

I'm a PhD student using the rnnlib, i have istalled rnnlib on ubunto, the installation was successful, but when i'm runnig 'rnnlib transcription.config' on terminal of Ubuntu, i have got an errror :
Inconsistency detected by ld.so: dl-version.c: 224: _dl_check_map_versions: Assertion `needed != ((void *)0)' failed!

Please help me: how can i run experiments?