Kernel Learning by Spectral Representation and Gaussian Mixtures

Round 1

Reviewer 1 Report

Llamas et.al. presented a new approach for classification and regression. Overall, the paper is both well-written and organized. Some strengths related to scientific content are:

1. An overview of related works, some from recent years, is presented. It contains studies from specific literature about the subject in discussion.

2. The authors describe their original contribution related to the topic in discussion.

3. The methodology is well-defined and justified from a scientific point of view. It provides a detailed analysis of the results.

5. The results are well illustrated and discussed.

6. The conclusions are clearly defined.

 

I have some broad remarks:

1. [Line 182] It is not clear why the authors chose N=6000

2. [Line 183] Why did authors choose to present accuracy as a metric when tables mention AUC score

3. [Line 194] Probably it is a typo: should be Tables 3,1,2

4. [Line 225] Typo: As it can "bee"

5. [Line 227-228] It was not clear why authors are pointing to one data point to talk about prediction vs actual value

I think the draft is almost ready for publication with few updates as suggested above.

Author Response

Dear reviewer

 

We appreciate the review and comments you made to our article, we respond to comments below

 

Llamas et.al. presented a new approach for classification and regression. Overall, the paper is both well-written and organized. Some strengths related to scientific content are:

1. An overview of related works, some from recent years, is presented. It contains studies from specific literature about the subject in discussion.

 

Thanks for your comments

 

2. The authors describe their original contribution related to the topic in discussion.

 

Thanks for your comments

 

3. The methodology is well-defined and justified from a scientific point of view. It provides a detailed analysis of the results.

 

Thanks for your comments

 

4. The results are well illustrated and discussed.

 

Thanks for your comments

5. The conclusions are clearly defined.

Thanks for your comments

I have some broad remarks:

1. [Line 182] It is not clear why the authors chose N=6000

We chose N=6000 given the slowness of the algorithm. The use of Gibbs sampler and Python where the main problems for the algorithm. In the case of the Gibbs sampler, it is well know that parallelization of the runs has not been achieved.

2. [Line 183] Why did authors choose to present accuracy as a metric when tables mention AUC score

Thanks for pointing out our error. It has been corrected.

3. [Line 194] Probably it is a typo: should be Tables 3,1,2

Thanks for pointing out our error. It has been corrected.

4. [Line 225] Typo: As it can "bee"

Thanks for pointing out our error. It has been corrected.

5. [Line 227-228] It was not clear why authors are pointing to one data point to talk about prediction vs actual value

We expanded the explanation to involve how the MSE error points to good performance on regression prediction. The sample was an example of that.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper is well written and very theoretical. There are some minor issues to be addressed before it can be accepted.

 

1. In my understanding, Eq 12 is not proper. I think a correct way to present Eq 12 is that a summation of some variables follow a new distribution.

 

2. The figures in the paper have too small font sizes to read.

 

3. The references are not cited properly in format.

 

4. It is more convincing to provide p-value rather than providing R2 value only.

Author Response

Dear reviewer

We appreciate the review and comments you made to our article, we respond to comments below

 

1. In my understanding, Eq 12 is not proper. I think a correct way to present Eq 12 is that a summation of some variables follows a new distribution.

 

Thanks for your comment. Regarding Eq 12, we rewrote in the correct form given that is a Mixture of Gaussians

 

2. The figures in the paper have too small font sizes to read.

 

The font sizes were resized

 

3. The references are not cited properly in format.

 

We have analyzed and correct our paper according to “the cited format” in the general document provided by MDPI.

 

 

4. It is more convincing to provide p-value rather than providing R2 value only.

In our case, we want to know how much data variation is explained by our model. So, for this paper our target is to see how the model shows the variations in the classification and in the regression.

     

Author Response File: Author Response.pdf

Reviewer 3 Report

We here summarize these points:

1.    The abstract includes an important point of work.

2.    The introduction already shows the novelty in the article, is clear and relatively easy to read.

3.    The steps are clear. However, the results in Figure 3 and Figure 4 need to be better clearly because they are still unclear or look blurry (related to the description of each legend in the figure).

4.    The results sound right and rather complete. The methods are generally appropriate, although clarification of a few details and provision of a rationale for the use of particular formula should be provided. For example, it is better to write down the reason why the distribution shown in Equation (12) is chosen in more detail.

5.    The experiment had good accuracy, in most cases an accuracy above 0.8. For example in the dataset for  breast cancer an accuracy of 0.89, and in the credit-g dataset an accuracy of 0.91 using only 500 frequencies in 5 swaps. It is better to write down the procedure to perform with accuracy.

Comments for author File: Comments.pdf

Author Response

Dear reviewer

We appreciate the review and comments you made to our article, we respond to comments below

 

1. The abstract includes an important point of work.

 

Thanks for your comments

2. The introduction already shows the novelty in the article, is clear and relatively easy to read.

    Thanks for your comments

3. The steps are clear. However, the results in Figure 3 and Figure 4 need to be better clearly because they are still unclear or look blurry (related to the description of each legend in the figure).

The description of each legend in the figures were fixed

4. The results sound right and rather complete. The methods are generally appropriate, although clarification of a few details and provision of a rationale for the use of particular formula should be provided. For example, it is better to write down the reason why the distribution shown in Equation (12) is chosen in more detail.

      Thanks for your comment. Regarding Eq 12, we rewrote in the correct form given that is a Mixture of Gaussians

5. The experiment had good accuracy, in most cases an accuracy above 0.8. For example, in the dataset for breast cancer an accuracy of 0.89, and in the credit-g dataset an accuracy of 0.91 using only 500 frequencies in 5 swaps. It is better to write down the procedure to perform with accuracy.

      Thanks for your comments, we consider only showing all results because the procedure is a little extensive, around two sheets.

Author Response File: Author Response.pdf