Inference

Scoring code

Challenge

The challenge will ask participants to measure the ionization fraction for each redshift x(z) and the corresponding  uncertainity P(x|z).

Scoring

The submission file is the full posterior 3D posterior which includes the cross-correlation between the three redshift bins. The score associated to each submission is the value of the posterior for the point corresponding to the true reionization values at the three redshifts, x’HI_z1, x’HI_z2, x’HI_z3. An example submission file with the proper header and data format can be found here.
We expect the submission file to strictly follow the properties as mentioned in the header (e.g., pixel size, number of pixels in each axis, etc).
Note that we also recommend the submission file to be normalized to 1, wherein the raw pixel values in the cube are simply summed and each pixel value is divided by this sum.  In case the submission file is not normalized in this way our scoring and validation script will automatically re-normalize the submission file following the above convention for uniformity.
The different datasets, PS1 and PS2, along with the optional PS3 and IM1 will all be scored separately. Scores corresponding to the same team and different datasets will not be combined, and no global challenge leader board will be produced.  The scores however will be useful in order to quantify and compare modelling- and foreground-induced errors.  During the challenge, only a format validation service for the submission will be returned. This will provide feedback on whether a submission has been received and it can be read correctly by the scoring routine. Scores will only be returned at the end of the challenge on the last submitted results.