Hi @YyzHarry, I want to use your code of LDS to solve my problem,now I have a question to ask you：whether the input data format of LDS must be csv？Is npz format data OK？can it be applied to high-dimensional data？ I would appreciate it if you could give me some guidance.

Hi Team, I liked the ideas in your paper, but from reading the paper and provided code it sounds like the provided FDS and LDS code can be applied to any dataset/model? Is it really true?

• It looks like you are using only integers(as you are predicting age) to make a dictionary of histogram bins in both FDS and LDS. In the paper you say : "We use a minimum bin size of 1, i.e., yb+1 − yb = 1, and group features with the same target value in the same bin." I imagine this makes a lot of things easier but if you are facing imbalanced regression problem and your labels are float between 0 and 5 this version of code won't help you. Do you by any chance have code with general approach?
  • see following parts of code with usecase specific histogram bins:
    • https://github.com/YyzHarry/imbalanced-regression/blob/055a7b3804bbaf903ed25a55c11ab8acc6e142e1/agedb-dir/datasets.py#L60
    • https://github.com/YyzHarry/imbalanced-regression/blob/055a7b3804bbaf903ed25a55c11ab8acc6e142e1/agedb-dir/fds.py#L120
• I did not find an explanation for this clipping(maybe empirically it gave better results?): https://github.com/YyzHarry/imbalanced-regression/blob/055a7b3804bbaf903ed25a55c11ab8acc6e142e1/agedb-dir/datasets.py#L67
• there is also another clipping here(I guess again better empirical results?): https://github.com/YyzHarry/imbalanced-regression/blob/055a7b3804bbaf903ed25a55c11ab8acc6e142e1/agedb-dir/utils.py#L102

Note: I like the ideas in the paper, but due to lack of documentation/explanation I am right now spending a lot of time on generalizing the code and trying to figure out why you made some of the operations(eg. clippings)

hello, I read your paper interestingly and want to ask you a question about the prediction result processing. I would like to ask how to limit the last prediction y^ to be between 0 and 99, or to get it directly from the regression function without any processing?

Hi, Amazing for your great job in the imblanced regression problem. But I notice that this work discusses more on the 1D regression problems. What if the output is more than 1D (like Batch_Size x 10) ? Any suggestion will be helpful. Thanks.

Hello, I have some questions about the error pdf. Can I know how to get the right error pdf? Each labels have different numbers of samples, so should I apply a mean method for each label error or make them have the same amount?

Hi @YyzHarry ，The reproduced benchmark and model seem to be damaged。I use this link（ https://drive.google.com/file/d/1CPDlcRCQ1EC4E3x9w955cmILSaVOlkyz/view ）The downloaded model cannot be opened. It indicates that the data has been damaged. Can you update the model? Thank you！

Hi! This work is really fantastic! However, I found it hard to apply LDS/FDS to classic machine learning models like random forest. For example, after getting the effective label density with LDS, how should I use this?

I'm dealing my unbalanced data with your study; A little confusion: how to compute the Feature statistics similarity between the many-shot region and few-shot region? Cuz the features in the few-shot region bins are less than that in many-shot region.

Thanks to your attention!

Hi authors,

page 6 from your paper: Precisely, Focal-R loss based on L1 distance can be written as 1/n∑n i=1 σ(|βei|)γ ei, where ei is the L1 error for i-th sample, σ(·)

• QUESTION 1 : in the focal_mse loss: https://github.com/YyzHarry/imbalanced-regression/blob/055a7b3804bbaf903ed25a55c11ab8acc6e142e1/agedb-dir/loss.py#L24 should be torch.abs((inputs - targets)**2) and not only torch.abs(inputs - targets), am I correct?
• QUESTION 2 : why is there 2*torch.abs(...)-1 ? you do not have and -1 or 2* in the function in your paper?

Hi, I'm wondering whether I can use this method as a preprocessing step for non-DNN models, namely simple linear regression or elastic net regression? If so, how should I adopt this method?

Thank you so much!

Thanks a lot for your contribution, your works are really awesome. I am very interested in your work. However, during the code reading, I did not find the SHHS-DIR dataset. Could you publish the SHHS-DIR dataset or its sampling method, thank you!

Hello, in your paper on the problem of deep imbalance regression, I have the privilege of learning about the smoothing methods of LDS and FDS. In one of my machine learning projects predicting convective cloud precipitation, I wanted to use FDS to play a role in it because of the imbalance between non-precipitation samples and precipitation samples. I wonder how to smooth the feature statistic without knowing its label in the test set, in my data, my data is very unbalanced (70% of the data without precipitation), which results in the characteristic statistics of each label interval are particularly similar (about 98%), so that the smoothing effect is still not significant, and if there are some important points to pay attention to if using FDS in machine learning?

hello, I read your paper interestingly and want to apply it to my custom data. In this regard, two problems occurred.

first, when i evaluate my custom data, i can suffer this error do you know what is?

And can i extract each prediction results, not average prediction result?.

thank you

Hi @YyzHarry,

I am trying to adapt the example from here https://github.com/YyzHarry/imbalanced-regression/tree/main/agedb-dir with my custom model and data. Thus, I would like to ask you whether this would be feasible and if yes if there are any example showing explicitly how to do that.

Thanks.