Closed vishaal27 closed 4 days ago
Hi Vishaal,
Thank you for your interest in our work!
We will add some instructions to the notebooks as soon as we have some time. In the meantime, I can give you some instructions here. First, you would replace anchor_points['scenario_name']
with your stratified random sample indices for each scenario and anchor_weights['scenario_name']
with 1/number_item
, where number_item
is the number of seen examples (eg, 100 in our notebook). Second, you need to fit the IRT model and compute pirt_preds['scenario_name']
, which should be done in the same way as in the notebook. To get gpirt_preds['scenario_name']
("random++", in your case), you will need to compute lambds['scenario_name']
, which is done by using get_lambda(b, v/number_item))
where v
and b
should be computed as in the notebook.
Please let me know if you need further assistance or if anything does not make sense!
Thanks for the response, will try it and let you know if I face any issues
Hey, thanks for your great work and releasing your code publicly.
I am interested in the implementation of the random++ method. As far as I understand, this would just involve stratified sampling instead of clustering during the anchor point selection stage. However, I am a bit confused how to make it work with the IRT estimates. Could you please add that code into the notebook too (or if not provide some pointers how I could do it). Thanks again!