[PRE REVIEW]: Explaining Black-Box Models through Counterfactuals

[whedon]

Submitting author: @pat-alt (Patrick Altmeyer) Repository: https://github.com/pat-alt/CounterfactualExplanations.jl Version: Editor: @matbesancon Reviewers: jmaces, @kartikeyrinwa Managing EiC: Mathieu Besançon

Author instructions

Thanks for submitting your paper to JCON @pat-alt. Currently, there isn't an JCON editor assigned to your paper.

The author's suggestion for the handling editor is @matbesancon.

@pat-alt if you have any suggestions for potential reviewers then please mention them here in this thread (without tagging them with an @). In addition, this list of people have already agreed to review for JCON and may be suitable for this submission (please start at the bottom of the list).

Editor instructions

The JCON submission bot @whedon is here to help you find and assign reviewers and start the main review. To find out what @whedon can do for you type:

@whedon commands

[whedon]

Hello human, I'm @whedon, a robot that can help you with some common editorial tasks.

:warning: JOSS reduced service mode :warning:

Due to the challenges of the COVID-19 pandemic, JOSS is currently operating in a "reduced service mode". You can read more about what that means in our blog post.

For a list of things I can do to help you, just type:

@whedon commands

For example, to regenerate the paper pdf after making changes in the paper's md or bib files, type:

@whedon generate pdf

[whedon]

Failed to discover a Statement of need section in paper

[whedon]

Wordcount for paper.tex is 6252

[whedon]

Software report (experimental):

github.com/AlDanial/cloc v 1.88  T=0.15 s (579.4 files/s, 108410.9 lines/s)
-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
TeX                             10            752            191           5581
HTML                             1            257              2           3044
Julia                           42            676            264           2877
Markdown                        19            582              0           1734
TOML                             6             30              0            204
YAML                             6              3              2            125
R                                2             10              6             66
Ruby                             1              8              4             45
SVG                              1              0              0              3
-------------------------------------------------------------------------------
SUM:                            88           2318            469          13679
-------------------------------------------------------------------------------

Statistical information for the repository '897a019721aade7ac36dc356' was
gathered on 2022/10/27.
The following historical commit information, by author, was found:

Author                     Commits    Insertions      Deletions    % of changes
pat-alt                         13          4129           4186          100.00

Below are the number of rows from each author that have survived and are still
intact in the current revision:

Author                     Rows      Stability          Age       % in comments
pat-alt                      57            1.4          0.1                7.02

[whedon]

:point_right::page_facing_up: Download article proof :page_facing_up: View article proof on GitHub :page_facing_up: :point_left:

[whedon]

Reference check summary (note 'MISSING' DOIs are suggestions that need verification):

OK DOIs

- doi.org/10.1038/s41597-020-0448-0 is OK

MISSING DOIs

- 10.2139/ssrn.195733 may be a valid DOI for title: Instrumental Variables Estimates of the Effect of Subsidized Training on the Quantiles of Trainee Earnings
- 10.1109/deeptest52559.2021.00007 may be a valid DOI for title: Machine Learning Model Drift Detection Via Weak Data Slices
- 10.3386/w20343 may be a valid DOI for title: Reference-Dependent Preferences: Evidence from Marathon Runners
- 10.1257/aer.99.1.486 may be a valid DOI for title: Indirect Effects of an Aid Program: How Do Cash Transfers Affect Ineligibles’ Consumption?
- 10.1017/cbo9780511618031.014 may be a valid DOI for title: “Coherent Arbitrariness”: Stable Demand Curves without Stable Preferences
- 10.1017/cbo9780511618031.015 may be a valid DOI for title: Tom Sawyer and the Construction of Value
- 10.1016/j.inffus.2019.12.012 may be a valid DOI for title: Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI
- 10.7551/mitpress/3077.003.0044 may be a valid DOI for title: Deciding Advantageously before Knowing the Advantageous Strategy
- 10.1007/978-1-4419-9096-9 may be a valid DOI for title: Reproducing Kernel Hilbert Spaces in Probability and Statistics
- 10.3386/w3487 may be a valid DOI for title: The Federal Funds Rate and the Channels of Monetary Transnission
- 10.1016/j.techsoc.2021.101852 may be a valid DOI for title: Machine Learning, Knowledge Risk, and Principal-Agent Problems in Automated Trading
- 10.2139/ssrn.965818 may be a valid DOI for title: Identification of Peer Effects through Social Networks
- 10.2139/ssrn.3534495 may be a valid DOI for title: Peer Effects in Networks: A Survey
- 10.2307/2234554 may be a valid DOI for title: Nonlinear Dynamics, Chaos, and Instability: Statistical Theory and Economic Evidence
- 10.3386/w4509 may be a valid DOI for title: Minimum Wages and Employment: A Case Study of the Fast Food Industry in New Jersey and Pennsylvania
- 10.1109/sp.2017.49 may be a valid DOI for title: Towards Evaluating the Robustness of Neural Networks
- 10.3386/w14032 may be a valid DOI for title: Does Your Cohort Matter? Measuring Peer Effects in College Achievement
- 10.3982/ecta10168 may be a valid DOI for title: From Natural Variation to Optimal Policy? The Importance of Endogenous Peer Group Formation
- 10.2139/ssrn.4090707 may be a valid DOI for title: Explainable Artificial Intelligence: Interpreting Default Forecasting Models Based on Machine Learning
- 10.3386/w15617 may be a valid DOI for title: Adjustment Costs, Firm Responses, and Micro vs. Macro Labor Supply Elasticities: Evidence from Danish Tax Records
- 10.1214/18-aoas1222 may be a valid DOI for title: Variable Prioritization in Nonlinear Black Box Methods: A Genetic Association Case Study
- 10.1145/3514094.3534144 may be a valid DOI for title: Counterfactual Explanations for Prediction and Diagnosis in Xai
- 10.1080/01621459.1999.10473858 may be a valid DOI for title: Causal Effects in Nonexperimental Studies: Reevaluating the Evaluation of Training Programs
- 10.2139/ssrn.1596425 may be a valid DOI for title: The Persistent Effects of Peru’s Mining Mita
- 10.3233/ds-200028 may be a valid DOI for title: Reinforcement Learning for Personalization: A Systematic Literature Review
- 10.3390/app11167274 may be a valid DOI for title: A Framework and Benchmarking Study for Counterfactual Generating Methods on Tabular Data
- 10.1037/0022-3514.37.7.1097 may be a valid DOI for title: The Stability of Behavior: I. On Predicting Most of the People Much of the Time.
- 10.1086/497818 may be a valid DOI for title: Clean Evidence on Peer Effects
- 10.2139/ssrn.203194 may be a valid DOI for title: Cooperation and Punishment in Public Goods Experiments
- 10.1287/mnsc.2017.2908 may be a valid DOI for title: On the External Validity of Social Preference Games: A Systematic Lab-Field Study
- 10.1037/0022-3514.75.3.617 may be a valid DOI for title: Immune Neglect: A Source of Durability Bias in Affective Forecasting.
- 10.1037/e722842011-016 may be a valid DOI for title: The Uncertainty Effect: When a Risky Prospect Is Valued Less than Its Worst Possible Outcome
- 10.1007/978-3-030-42553-1_3 may be a valid DOI for title: Bayesian Neural Networks: An Introduction and Survey
- 10.3386/w11562 may be a valid DOI for title: The Incredible Volcker Disinflation
- 10.1920/wp.cem.2017.0817 may be a valid DOI for title: An Econometric Model of Network Formation with Degree Heterogeneity
- 10.1017/cbo9780511618031.006 may be a valid DOI for title: Economic Theory of Choice and the Preference Reversal Phenomenon
- 10.1086/717935 may be a valid DOI for title: Name Your Friends, but Only Five? The Importance of Censoring in Peer Effects Estimates Using Social Network Data
- 10.1109/icmla.2011.144 may be a valid DOI for title: Thompson Sampling for Dynamic Multi-Armed Bandits
- 10.1016/j.ins.2008.07.024 may be a valid DOI for title: Improving Artificial Neural Networks’ Performance in Seasonal Time Series Forecasting
- 10.1145/1273496.1273541 may be a valid DOI for title: A Bound on the Label Complexity of Agnostic Active Learning
- 10.1177/2053951720926558 may be a valid DOI for title: The Virtue of Simplicity: On Machine Learning Models in Algorithmic Trading
- 10.1509/jmkr.48.spl.s23 may be a valid DOI for title: Increasing Saving Behavior through Age-Progressed Renderings of the Future Self
- 10.1006/obhd.1996.0077 may be a valid DOI for title: The Evaluability Hypothesis: An Explanation for Preference Reversals between Joint and Separate Evaluations of Alternatives
- 10.1017/cbo9780511618031.033 may be a valid DOI for title: Music, Pandas, and Muggers: On the Affective Psychology of Value.
- 10.1002/jae.2426 may be a valid DOI for title: A Social Interactions Model with Endogenous Friendship Formation and Selectivity
- 10.21105/joss.00602 may be a valid DOI for title: Flux: Elegant Machine Learning with Julia
- 10.1257/aer.97.3.890 may be a valid DOI for title: Meeting Strangers and Friends of Friends: How Random Are Social Networks?
- 10.1126/science.1111709 may be a valid DOI for title: Failure to Detect Mismatches between Intention and Outcome in a Simple Decision Task
- 10.2139/ssrn.3043404 may be a valid DOI for title: Estimation of Peer Effects in Endogenous Social Networks: Control Function Approach
- 10.1198/1061860032148 may be a valid DOI for title: A Modified Principal Component Technique Based on the LASSO
- 10.1109/mci.2022.3155327 may be a valid DOI for title: Hands-on Bayesian Neural Networks–a Tutorial for Deep Learning Users
- 10.1017/cbo9781139175197.009 may be a valid DOI for title: Experimental Tests of the Endowment Effect and the Coase Theorem
- 10.1016/0749-5978(92)90015-y may be a valid DOI for title: Reference Points, Anchors, Norms, and Mixed Feelings
- 10.1145/3313831.3376219 may be a valid DOI for title: Interpreting Interpretability: Understanding Data Scientists’ Use of Interpretability Tools for Machine Learning
- 10.4314/ajst.v5i2.15330 may be a valid DOI for title: Seasonal Time Series Forecasting: A Comparative Study of ARIMA and ANN Models
- 10.1007/978-3-030-93842-0_6 may be a valid DOI for title: Exploring Explainable AI in the Financial Sector: Perspectives of Banks and Supervisory Authorities
- 10.1081/etc-120025891 may be a valid DOI for title: Best Spatial Two-Stage Least Squares Estimators for a Spatial Autoregressive Model with Autoregressive Disturbances
- 10.1177/0956797612450302 may be a valid DOI for title: The Financial Costs of Sadness
- 10.3386/w9736 may be a valid DOI for title: Neoclassical Theory versus Prospect Theory: Evidence from the Marketplace
- 10.3386/w7682 may be a valid DOI for title: The Power of Suggestion: Inertia in 401 (k) Participation and Savings Behavior
- 10.2307/2298123 may be a valid DOI for title: Identification of Endogenous Social Effects: The Reflection Problem
- 10.2139/ssrn.2523510 may be a valid DOI for title: Goals as Reference Points in Marathon Running: A Novel Test of Reference Dependence
- 10.20955/wp.2015.012 may be a valid DOI for title: FRED-MD: A Monthly Database for Macroeconomic Research
- 10.1007/s10618-013-0342-x may be a valid DOI for title: Visualizing Multi-Dimensional Decision Boundaries in 2D
- 10.1016/j.artint.2018.07.007 may be a valid DOI for title: Explanation in Artificial Intelligence: Insights from the Social Sciences
- 10.1037/0022-3514.54.4.687 may be a valid DOI for title: The Nature of Adolescent Competencies Predicted by Preschool Delay of Gratification.
- 10.1145/3287560.3287574 may be a valid DOI for title: Explaining Explanations in AI
- 10.1145/3351095.3372850 may be a valid DOI for title: Explaining Machine Learning Classifiers through Diverse Counterfactual Explanations
- 10.1109/cisda.2015.7208643 may be a valid DOI for title: Evaluating Model Drift in Machine Learning Algorithms
- 10.1016/s0167-7152(96)00140-x may be a valid DOI for title: Sparse Spatial Autoregressions
- 10.1145/3241036 may be a valid DOI for title: The Seven Tools of Causal Inference, with Reflections on Machine Learning
- 10.5860/choice.38-4002 may be a valid DOI for title: Economic Events, Ideas, and Policies: The 1960s and After
- 10.1016/j.jeconom.2014.08.008 may be a valid DOI for title: Estimating a Spatial Autoregressive Model with an Endogenous Spatial Weight Matrix
- 10.3386/w2966 may be a valid DOI for title: Does Monetary Policy Matter? A New Test in the Spirit of Friedman and Schwartz
- 10.1038/s42256-019-0048-x may be a valid DOI for title: Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead
- 10.3386/w7469 may be a valid DOI for title: Peer Effects with Random Assignment: Results for Dartmouth Roommates
- 10.1080/01621459.2017.1395341 may be a valid DOI for title: Least Ambiguous Set-Valued Classifiers with Bounded Error Levels
- 10.1109/icdcsw.2011.20 may be a valid DOI for title: Finding a" Kneedle" in a Haystack: Detecting Knee Points in System Behavior
- 10.1086/209205 may be a valid DOI for title: Choice Based on Reasons: The Case of Attraction and Compromise Effects
- 10.21034/qr.1011 may be a valid DOI for title: Are Forecasting Models Usable for Policy Analysis?
- 10.1037/e301532005-001 may be a valid DOI for title: Who Accepts Savage’s Axiom?
- 10.1109/tmm.2014.2330697 may be a valid DOI for title: A Simple Method to Determine If a Music Information Retrieval System Is a “Horse”
- 10.1017/cbo9780511618031.039 may be a valid DOI for title: Libertarian Paternalism Is Not an Oxymoron
- 10.1016/0165-1765(81)90067-7 may be a valid DOI for title: Some Empirical Evidence on Dynamic Inconsistency
- 10.1086/380085 may be a valid DOI for title: Save More Tomorrow™: Using Behavioral Economics to Increase Employee Saving
- 10.1007/978-1-4613-2391-4_2 may be a valid DOI for title: The Framing of Decisions and the Psychology of Choice
- 10.1177/0956797610396225 may be a valid DOI for title: How Incidental Values from the Environment Affect Decisions about Money, Risk, and Delay
- 10.1145/3287560.3287566 may be a valid DOI for title: Actionable Recourse in Linear Classification
- 10.1037/0022-3514.79.1.66 may be a valid DOI for title: Egocentric Empathy Gaps between Owners and Buyers: Misperceptions of the Endowment Effect.
- 10.2139/ssrn.3589337 may be a valid DOI for title: Modeling Multivariate Time Series in Economics: From Auto-Regressions to Recurrent Neural Networks
- 10.2139/ssrn.3063289 may be a valid DOI for title: Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR
- 10.1007/bf00116900 may be a valid DOI for title: Learning in the Presence of Concept Drift and Hidden Contexts
- 10.1093/biostatistics/kxp008 may be a valid DOI for title: A Penalized Matrix Decomposition, with Applications to Sparse Principal Components and Canonical Correlation Analysis
- 10.1016/j.eswa.2007.12.020 may be a valid DOI for title: The Comparisons of Data Mining Techniques for the Predictive Accuracy of Probability of Default of Credit Card Clients
- 10.1016/s0925-2312(01)00702-0 may be a valid DOI for title: Time Series Forecasting Using a Hybrid ARIMA and Neural Network Model

INVALID DOIs

- None

[matbesancon]

@whedon assign @matbesancon as editor

[whedon]

OK, the editor is @matbesancon

[carstenbauer]

@whedon assign @matbesancon as editor

[whedon]

OK, the editor is @matbesancon

[pat-alt]

Just to flag that I have been adding updates to the package functionality and will likely add more in the coming days. Should not affect the draft paper in any substantial ways, but where necessary I will revise as soon as possible.

[matbesancon]

@whedon add jmaces as reviewer

[whedon]

OK, jmaces is now a reviewer

[matbesancon]

@whedon add @jmaces as reviewer

[whedon]

OK, @jmaces is now a reviewer

[matbesancon]

ok @pat-alt let us know when this is ready

[pat-alt]

Will do, thanks. Need to finish some work on a (related) research paper before the holidays and would then turn to this first thing in the new year. Does that work?

[matbesancon]

yes it does :)

[pat-alt]

Sorry, working on it 😅 will get there this week

[pat-alt]

hi @matbesancon it's taken me longer than expected, but everything is updated now in the PR linked above. Sorry for the delay and thanks for your patience. I'll update the docs and push a new package release later this week, but the paper is ready.

Also, as I had mentioned in the original submission, there is some issue with listing captions. I have simply refrained from using captions for now, but this messes up the cross-references, so that definitely needs editing.

[pat-alt]

@whedon generate pdf

[whedon]

:point_right::page_facing_up: Download article proof :page_facing_up: View article proof on GitHub :page_facing_up: :point_left:

[matbesancon]

sorry about the delays on our side now, I will send this back to the reviewers

[pat-alt]

No worries!

Just to flag that I've developed the package quite a bit since Jan. Some major updates that may be worth mentioning in the paper include:

• [ ] Improved benchmarking (0.1.8)
• [ ] Full composability of counterfactual generators (0.1.9)

Unfortunately, I won't really have time to make further edits to the paper until mid-May, but thought I'd still mention it here so reviewers are aware.

Thanks!

[pat-alt]

@matbesancon with JuliaCon 2023 approaching fast, is there anything I can do to accelerate this? 😄 Happy to ask for interest in reviewing this on discourse, slack and socials if that's acceptable.

We've had some major updates in meantime and now have a group of TU Delft students working on this package full time for a couple of weeks, so at this point it may well be worth updating the paper as well.

Let me know how if I can help in any way, would be great to see this move forward

[kartikeyrinwa]

Introduction

In this project the authors have developed a Julia library to construct counterfactual explanations for Machine Learning models. The goal is to develop a library that can easily be extended to new ML models and counterfactual generators. The authors achieve this by allowing for easy extensibility of the allowable input model and of the generating functions. The present illustrative examples on the Give Me Some Credit and MNIST data sets.

Review

Overall, I believe that the provided library is a key contribution to Julia's machine learning ecosystem. It comes along with extensive documentation and will be a key resource for researchers and practitioners. I recommend acceptance with minor revision.

Minor Revisions

• In section 2.2 you mention different desiderata for counterfactual explanations. It would be a good idea to provide functions to evaluate the different generated counterfactuals on some standard metrics for these properties.
• Figure 1 is difficult to read. I recommend increasing the font size of the text in the boxes.
• It might be a good idea to give a description of counterfactual search outside gradient descent so that people unfamiliar with the topic understand what approaches there can be besides gradient descent.
• In the paper you refer to the code in various locations as Listings. However, the code blocks have no numbering. I recommend provide such numbering to the blocks to make them easy to find.
• Section 7.1 has a missing reference for the PROBE method.
• Section 2.2: realistic plausible -> realistic and plausible

Reviewer Checklist

Conflict of interest

• [x] As the reviewer I confirm that I have read the JuliaCon conflict of interest policy and that there are no conflicts of interest for me to review this work.

Code of Conduct

• [x] I confirm that I read and will adhere to the JuliaCon code of conduct.

General checks

• [x] Repository: Is the source code for this software available at the repository url?
• [x] License: Does the repository contain a plain-text LICENSE file with the contents of an OSI approved software license?
• [x] Authorship: Has the submitting author (@pat-alt) made major contributions to the software? Does the full list of paper authors seem appropriate and complete?

Functionality

• [x] Installation: Does installation proceed as outlined in the documentation?
• [x] Functionality: Have the functional claims of the software been confirmed?
• [x] Performance: If there are any performance claims of the software, have they been confirmed? (If there are no claims, please check off this item.)

Documentation

• [x] A statement of need: Do the authors clearly state what problems the software is designed to solve and who the target audience is?
• [x] Installation instructions: Is there a clearly-stated list of dependencies? Ideally these should be handled with an automated package management solution.
• [x] Example usage: Do the authors include examples of how to use the software (ideally to solve real-world analysis problems).
• [x] Functionality documentation: Is the core functionality of the software documented to a satisfactory level (e.g., API method documentation)?
• [ ] Automated tests: Are there automated tests or manual steps described so that the function of the software can be verified?
• [ ] Community guidelines: Are there clear guidelines for third parties wishing to 1) Contribute to the software 2) Report issues or problems with the software 3) Seek support

Paper format

• [x] Authors: Does the paper.tex file include a list of authors with their affiliations?
• [x] A statement of need: Do the authors clearly state what problems the software is designed to solve and who the target audience is?
• [x] References: Do all archival references that should have a DOI list one (e.g., papers, datasets, software)?
• [x] Page limit: Is the page limit for full papers respected by the submitted document?

Content

• [x] Context: is the scientific context motivating the work correctly presented?
• [x] Methodology: is the approach taken in the work justified, presented with enough details and reference to reproduce it?
• [ ] Results: are the results presented and compared to approaches with similar goals?

[matbesancon]

@whedon remove @jmaces as reviewer

[whedon]

OK, @jmaces is no longer a reviewer

[matbesancon]

@whedon add @kartikeyrinwa as reviewer

[whedon]

OK, @kartikeyrinwa is now a reviewer

[matbesancon]

@whedon start review

[whedon]

OK, I've started the review over in https://github.com/JuliaCon/proceedings-review/issues/130.

[pat-alt]

Thanks for this very helpful list of comments @kartikeyrinwa - I plan to address them all this week. Tracking/comments in this issue.