Competition Leaderboard

Submission Format

This year, we will evaluate your model(s) on a private test set, which will take place on our servers and be bound by certain constraints. Your submission should be either a single file or a zip file containing an executable file named run (no extension) as well as any auxiliary files (such as a model etc.). Please also include a shebang on the first line of your executable, for example for a python script the first line should be #!/env/bin/python.

Your submission will have acess to the test set inputs located at location test, which will contain files of the form part-xxxx which will be in the same format as the training data (without the labels). Your submission should read in these files and output a single file results.csv, and provide a line for each row in the test set consisting of:

<Tweet_Id>,<User_Id>,<prediction_reply>,<prediction_retweet>,<prediction_quote>,<prediction_fav>

Please see the code samples for an example.

Runtime Constraints

Your submissions will be run inside a container of an Docker image which you can provide. This image will be run on the e2-highmem-16 instances on GCP (see here for details). You submission will have a 64 GB memory limit and 24 hours time limit to score the entire test set (~10 million rows).

We provide the recsyschallenge2021/submission_env image based on the gcr.io/google-appengine/python Docker image, as a default. The full list of installed packages in the image can be seen in the FAQs