A Software Toolbox for Realistic Dataset Generation for Testing Online and Offline 3D Bin Packing Algorithms

Round 1

Reviewer 1 Report

Dear author,

Thank you for submitting your article “A Software Toolbox for Realistic Dataset Generation for Testing On-line and Off-line 3D Bin Packing Algorithms”. I have carefully reviewed your work and I find it very interesting. However, I have some comments and suggestions that could enhance the quality of your article:

1. In Table 3, please specify the unit of each parameter for clarity.

2. Please explain how you calculated the entropy index and why you chose this measure.

3. It would be helpful to include a pseudocode of the software toolbox to illustrate its main steps and functions.

4. Please provide some information about the size and shape of the bin used in your experiments.

5. To demonstrate the validity and reliability of your software toolbox, please provide the benchmark results when packing in the provided datasets.

6. For the online 3-D bin packing problem, please indicate the order of arrival of each item in the datasets, so that other researchers can easily evaluate their approach performance in comparison with the other methods.

I think your article has a great potential and I recommend it for publication after you address these issues.

Best regards,

English language and style are fine/minor spell check required.

Author Response

Dear Reviewer, please find our answers to your comments below. For the sake of readability we highlight our answer here and the corrections in the paper in blue.

I have carefully reviewed your work and I find it very interesting. However, I have some comments and suggestions that could enhance the quality of your article:

We are very glad to hear that the reviewer has found our paper very interesting and we thank the reviewer for the collegiality.

1. In Table 3, please specify the unit of each parameter for clarity.

Surely, we have updated the table with the required information.

2. Please explain how you calculated the entropy index and why you chose this measure.

Thank you for pointing this out. We now clarify in section 2 that "The entropy of an order was calculated in the following way: the dataset was grouped by order and product, and the Quantity values were summed for each group. The Shannon index was then calculated and normalized for each order. The function entropy from the python scipy.stats._entropy package was used for calculating the index. Normalization of values between 0 and 1 was carried out by subtracting the minimum entropy value from the calculated index and then dividing by the difference between the maximum and minimum entropy values."

3. It would be helpful to include a pseudocode of the software toolbox to illustrate its main steps and functions.

We have include a new figure, figure 1, that shows a flowchart of the procedure. This is done in section 2 when the procedure is explained for the first time. We have made small modifications in table 4 to link it to the new figure.

4. Please provide some information about the size and shape of the bin used in your experiments.

The dataset does not make any assumptions about the the size and shape of the bin. The dataset contain synthetic order derived with high fidelity from a industry dataset. Currently the industry that provided us the original dataset is using Euro pallets with size (length, width height) = (1200x800x1400) all measurements in millimeters. However this is an important clarification and we have addressed it in the paper at the end of Section 2. We have addressed it together with comment number 5 below and we kindly ask the reviewer to consider our integrated answer to comments 4 and 5 below.

5. To demonstrate the validity and reliability of your software toolbox, please provide the benchmark results when packing in the provided datasets.

This is a vert an important question and although we have developed our own solution we argue that such results should not be put together with the datasets because they are independent. We clarify this positioning in the paper at the end of section 2 by writing: "There are a two final pertinent observations about the generated dataset and the process for generating them.}

The first is that the provided datasets do not make any assumption about the bin size. For reference, the base dataset used to create the published datasets considered standard euro pallets with length 1200 mm, width 800 mm and a maximum palletizing height of 1400 mm. However, companies may change their bin format while keeping package sizes constant and vice versa.

The second is that the provided datasets do not include any information about what constitutes an optimal palletization sequence and positioning for the products in the orders. The reason for this is manifold. Different companies will use specific parameters for evaluating the quality of a pallet. Classic scientific metrics such as: minimization of envelope volume, minimization of number of bins, minimization of bin volume, etc; actually have a relatively low importance for many practical applications. Other constrains like: products of the same type being packed together, expensive products being packed in the center, the creation of interlocking layers that improve the stability of the bin/pallet when being displaced, the need to include additional load carriers between certain products, etc; are much more important in real world applications. The goal with the provided datasets is that whomever uses them must clearly contextualize the application area and indicate which palletization quality metrics were considered. Any order can be palletized/packed but quality indicators will vary greatly. Dataset users will subsequently establish the benchmark for their specific application scenario."

In this context our palletization algorithms respect constrains that may be irrelevant for other user of the datasets and we feel that the two (our benchmarking results and the dataset used in the benchmarking) should not be  confused but rather referred two at a later moment when the algorithms and not the datasets are being discussed.

6. For the online 3-D bin packing problem, please indicate the order of arrival of each item in the datasets, so that other researchers can easily evaluate their approach performance in comparison with the other methods.

This is an interesting question and it is important to mention that the provided toolbox does not prescribe a specific usage, allowing user quite much flexibility. We have clarified this in the paper in the end of section 4 by writing: "The generated data allows an high degree of flexibility in the usage of the data. For example, in online mode, users can opt to use a generator function such as the one provided by the code in the code repository, and receive randomly sampled data points, or they may enforce a specific delivery order, by re-grouping or re-ordering the provided dataset according to specific criteria and retrieving orders sequentially. A combined strategy is also possible whereby the users train and benchmark their algorithms using a specific order sequence from the provided datasets and then validate on order from the generator. The tool box does not prescribe any specific usage."

I think your article has a great potential and I recommend it for publication after you address these issues.

Thank you again for the pertinent and accurate comments and the collegiality in conveying them.

Reviewer 2 Report

Review of article 2466094

A Software Toolbox for Realistic Dataset Generation for Testing On-line and Off-line 3D Bin Packing Algorithms

By Ribeiro and Ananno

 

This work is to provide a toolbox for generating dataset for testing 3D bin packing.  This work has been appropriate completed and may be of interest to the audience of the process journal.  I found some points as addressed below:

(1)   It seems to me the authors did not provide an algorithm flowchart, and only referred the detail of their work to web site. I think this is not standard for a journal paper.

(2)   The author should provide the downstream of this work.  This data set should be good for some optimal packing approaches or software. How does these dataset work in those approaches?

 

I would recommend this work for publication should the above points are well-treated.

Comments for author File: Comments.pdf

This work is presented in a logical and clear manner.

Author Response

Dear Reviewer, we have addressed you valuable comment and suggestions to the best of our understanding. We highlight both our replies here and in the revised manuscript in blue.

This work is to provide a toolbox for generating dataset for testing 3D bin packing. This work has been appropriate completed and may be of interest to the audience of the process journal. I found some points as addressed below:

• It seems to me the authors did not provide an algorithm flowchart, and only referred the detail of their work to web site. I think this is not standard for a journal paper.

Thank you for this pertinent comment. We have added a new figure, new figure 1, with a flow chart of the algorithm. We have also made changes to table 4 to better link it to the newly introduced figure.

• The author should provide the downstream of this work. This data set should be good for some optimal packing approaches or software. How does these dataset work in those approaches?

We have throughout the paper better clarified the use cases and philosophy supporting the design of the dataset. See in particular section 2 where we position the utilization of the work. There we write: 

"There are two final pertinent observations about the generated dataset and the process for generating them.

The first is that the provided datasets do not make any assumption about the bin size. For reference, the base dataset used to create the published datasets considered standard euro pallets with length 1200 mm, width 800 mm and a maximum palletizing height of 1400 mm. However, companies may change their bin format while keeping package sizes constant and vice versa

The second is that the provided datasets do not include any information about what constitutes an optimal palletization sequence and positioning for the products in the orders. The reason for this is manifold. Different companies will use specific parameters for evaluating the quality of a pallet. Classic scientific metrics such as: minimization of envelope volume, minimization of the number of bins, minimization of bin volume, etc.; actually have relatively low importance for many practical applications. Other constraints like, products of the same type being packed together, expensive products being packed in the center, the creation of interlocking layers that improve the stability of the bin/pallet when being displaced, the need to include additional load carriers between certain products, etc.; are much more important in real world applications. The goal with the provided datasets is that whomever uses them must clearly contextualize the application area and indicate which palletization quality metrics were considered. Any order can be palletized/packed but quality indicators will vary greatly. Dataset users will subsequently establish the benchmark for their specific application scenario."

I would recommend this work for publication should the above points are well-treated.

Thank you for your important comments and collegiality.

Round 2

Reviewer 1 Report

Dear Editors and Authors,

The article has been revised satisfactorily and meets the standards of the journal. I approve it for publication without further changes.

Best Regards,

Thanh-Hung Nguyen