HYDROSAVER22 Jan - 20 Feb
The Newcrest Crowd
This competition has finished.
The goal is to predict the % solids of underflow 3 hours from now. This value is in the column labelled "target". The competition is scored using the Root Mean Squared Error (RMSE) metric.
The winner will be the entry that has the best RMSE as measured on the private test set and has not violated any competition rules.
A sample submission is provided in the sample_submission.csv. Your .csv submission must have 439140 rows plus a header row, labelled target. Each row must contain numeric values. Note that some of the values of the target variable in the test data set are missing, these rows will not contribute to the final score, you still need to predict a numeric value for these timesteps, but these values will not contribute to the RMSE error. The number of submissions is limited to 10 per day, per team.
Extra Rule - respect causality:
To be eligible for a prize your solution must not require knowledge of the values in future rows. This means that your solution should not rely on the use of future processing plant data observations when attempting to predict the target variable at the current timestep. An unshared holdout dataset will be used to determine if the winning solution violates causality. In this final round, your submitted code will be run multiple times, and will only have access to historic data after it’s prediction for that time step has already been recorded. If your model’s accuracy performs more than 10% worse in this setting, your model will be disqualified.
Each team can submit a one-page slide with their top ideas, data or process insights they found during the challenge. This can be submitted in bullet points, text or graphics, but keep it to one A4 page. There is no right or wrong for this part of the competition. A team of processing specialists at Newcrest Mining will determine the top submission and award the prize. We will enable a feature to submit a Final Submission shortly. This Final Submission should include your top scoring csv file, your Insights Document and some optional free text.
Scoring Update (30 Jan 2018):
We have identified periods in the data where the target values (% underflow solids) rapidly drop to single digit values and then increase to over 65%. This is due to upstream mill shutdowns (trips), which cannot be reliably predicted with the available data.
As such we are going to exclude from the test set any value where the target is < 30 and > 65. This means that these periods will not count towards your leaderboard and final score, and for most people this should result in an improvement in your RMSE score.
The new scoring system went live on 30 Jan 2018, 11 am (UTC +10). From this point forward we will be using the new scoring method. We have reprocessed submissions that were submitted after 29 Jan 2018, 11 am (UTC+10). All submissions prior to that need to be re-uploaded.
Scoring Update 2 (3 Feb 2018)
We have extended the rows that are being excluded from the scoring process to include any rows where DIC88023.PV is < 30 and > 65. This has further improved the results across the leaderboard. We expect this will to be the last scoring modification.
Final Submission update (14 Feb 18)
To make a successful and valid submission to the HydroSaver competition, each team is required to:
- Have at least one valid .csv score on the leaderboard and
- Finalise their submission via the "Finalise Submission" link next to their leaderboard score
- Attach the following files
- The team's top scoring .csv file (mandatory)
- The team's Insights Document (optional, but required to qualify for the Insights Prize)
You can edit and update your submission as often as you like but it has to be finalised and fulfil the above criteria prior to the Event End Date of 19th February 18, 11:59 pm (UTC +10).
Unearthed will contact the teams with the top private score to request their executable code for testing. After verifying the Winners' submission, the results will be announced on 26th February 18, 2:00 pm (UTC +10)
You have accepted the T&C as part of the registration process for this competition. If you would like to review, you can find them here.