This is our implementation of a Skip-Gram model with negative sampling.
NB: We have a mistake in the test-format.sh.
Concerning the test script we have an issue running it. The condition where the script verifies if the similarity values are numbers is an error but the values are numbers. So, I think it will be better to use your own test_format.sh
To generate the tar.gz skipGram file use this command :
tar cvf skipGram.tar.gz skipGram.py
We used Docker as our algorithm didn’t work on Windows. Therefore, we created a container under Linux in order to execute the proper command lines and have the proper environment.
This is the docker pull command:
docker pull amirmahmoudi/nlp
docker run -it <image-name> /bin/bash
bash ./test_format.sh skipGram.tar.gz #error will be raised but the output is valid for the `results.txt`
The docker image is really heavy as all the environment is downloaded on the docker pull command, including all the packages, libraries and most importantly the billion words .tar.gz file.
You should run the skipGram.py in this zip file using your own shell script, the docker is here to show the extra work we have done on this project.
Chloé Daems
[email protected]
Anne-Claire Laisney
[email protected]
Amir Mahmoudi
[email protected]