Hi Jianguo,

Great work! I'm interested to know if you have implemented the code for "Learning Optimal Conformal Classifiers"?

Hello,

Thank you for this helpful toolbox!

I saw that you have implemented weighted conformal predictor based on the paper "Conformal Prediction Under Covariate Shift".

However, is this the same as the non-exchangeable conformal prediction/risk control as described in the following papers?

• Conformal Prediction Beyond Exchangeability
• Non-Exchangeable Conformal Risk Control

Thanks!!
Tingyu Shi

Hello,

thanks for your very useful toolbox.
Since you seem to be very experienced with conformal prediction, could you give me an idea about the weakness of the THR score function, when combined with the SplitPredictor?

Across a range of models I tested, it seems to provide a vastly smaller average_size (Easily half of APS) for a given reached coverage_rate. Is it "just", as mentioned in the RAPS Paper (https://arxiv.org/abs/2009.14193), the theoretical guarantee of the coverage rate?

Additionally, I was wondering if the "naive" method of this paper (https://arxiv.org/pdf/2306.09335) is implemented here as well under a different name (Is it Margin?)?

Thanks again for your great work :)

Regards
Thomas

Hello,

and thanks again for your helpful toolbox.
I have a question regarding the implementation of "Margin()":

You mention
"
indices = torch.arange(num_labels).to(probs.device)
temp_probs[:, indices, indices] = -1

torch.max(temp_probs, dim=-1) are the second highest probabilities

"

Why would the comment be the case? From my perspective indices would need to something like
torch.max(probs, dim=-1).indices for this to make sense.
But I also dont understand why you unsqueeze anything here.
Could you maybe explain that?

Best Regards,

Thomas

想利用这个包对深度学习模型应用于文本分类的输出结果进行共形预测，但是不知道如何调整