1) Following popular demand, the raw datasets were now added to the repository. These are denoted by “_RAW”. Make sure you predict the right test dataset (test_dataset_RAW if you used train_dataset_RAW for training)! The spectra in the initial datasets (without the “_RAW” were normalized to unit maxima).

2) The predictions will be evaluated in terms of RMSE. Each element will be evaluated separately. The overall ranking will be determined as the weighted average of the achieved ranking across the 4 distinct elements. For example, if one provides the best prediction on 3 elements and the second-best prediction on the 4^th element, their final ranking will be 5/4=1.25. Potential ties will be resolved by taking into account the relative performance (to the next-best prediction) of the contestant.

3) While there are 50 spectra available for each test target, I expect a single prediction / target / element. For 15 test targets and 4 elements, this means 60 predicted values. Further clarifications regarding the formatting of the submissions are available in the submission forms.

4) Uncertainty refers to the 95 % prediction confidence. If you are not able to estimate your model’s uncertainty, provide “NA” values when submitting your results.

5) Considering the changes above (and due to multiple reports of the launch email ending up in spam folders), the deadline is prolonged till August 19.

To avoid spamming the whole community with further updates regarding the contest, I would kindly request anyone interested in participating to sign up to an email list and add kepes{at}vutbr.cz to their list of “non-spam” email addresses.

Your task is to predict the Cr, Mn, Mo, and Ni content of 15 metal alloys. More information is provided in the README file. The datasets are available here.

The results will be announced at the LIBS 2022 conference in Bari (IT) and communicated with the participants who provide a valid email address. The three best-performing groups will co-author a paper, describing the methodology used and the results obtained. The paper will be published in the Virtual Special Issue that Spectrochimica Acta B will devote to the Conference.

You can submit your predictions by uploading a file here (requires a Google account), by filling out a Google form (or an alternative), or by simply sending them via email to kepes{at}vutbr.cz as an attachment. The deadline is 23:59:59, July 31, 2022.

Note that you can submit your answer repeatedly. However, only your latest answer will be considered, and you will receive no feedback for preliminary submissions.

Looking forward to your contributions.