#️⃣Software Engineering: Programming (Part XVI: The Software Quality Perspective and AI)

Software Engineering Series

Organizations tend to complain about poor software quality developed in-house, by consultancy companies or third parties, without doing much in this direction. Unfortunately, this agrees with the bigger picture reflected by the quality standards adopted by organizations - people talk and complain about them, though they aren’t that eager to include them in the various strategies, or even if they are considered, they are seldom enforced adequately!

Moreover, even if quality standards are adopted, and a lot of effort may be spent in this direction (as everybody has strong opinions and there are many exceptions), as projects progress, all the good intentions come to an end, the rules fading on the way either because are too strict, too general, aren’t adequately prioritized or communicated, or there’s no time to implement (all of) them. This applies in general to programming and to the domains that revolve around data – Business Intelligence, Data Analytics or Data Science.

The volume of good quality code and deliverables is not only a reflection of an organization’s maturity in dealing with best practices but also of its maturity in handling technical debt, Project Management, software and data quality challenges. All these aspects are strongly related to each other and therefore require a systemic approach rather than focusing on the issues locally. The systemic approach allows organizations to bridge the gaps between business areas, teams, projects and any other areas of focus.

There are many questionable studies on the effect of methodologies on software quality and data issues, proclaiming that one methodology is better than the other in addressing the multifold aspects of software quality. Besides methodologies, some studies attempt to correlate quality with organizations’ size, management or programmers’ experience, the size of software, or whatever characteristic might seem to affect quality.

Bad code is written independently of companies’ size or programmer's experience, management or organization’s maturity. Bad code doesn’t necessarily happen all at once, but it can depend on circumstances, repetitive team, requirements and code changes. There are decisions and actions that sooner or later can affect the overall outcome negatively.

Rewriting the code from scratch might look like an approachable measure though it’s seldom the cost-effective solution. Allocating resources for refactoring is usually a better approach, though this tends to increase considerably the cost of projects, and organizations might be tempted to face the risks, whatever they might be. Independently of the approaches used, sooner or later the complexity of projects, requirements or code tends to kick back.

There are many voices arguing that AI will help in addressing the problems of software development, quality assurance and probably other areas. It’s questionable how much AI will help to address the gaps, non-concordances and other mistakes in requirements, and how it will develop quality code when it has basic "understanding" issues. Even if step by step all current issues revolving around AI will be fixed, it will take time and multiple iterations until meaningful progress will be made.

At least for now, AI tools like Copilot or ChatGPT can be used for learning a programming language or framework through predefined or ad-hoc prompts. Probably, it can be used also to identify deviations from best practices or other norms in scope. This doesn’t mean that AI will replace for now code reviews, testing and other practices used in assuring the quality of software, but it can be used as an additional method to check for what was eventually missed in the other methods.

AI may also have hidden gems that when discovered, polished and sized, may have a qualitative impact on software development and software. Only time will tell what’s possible and achievable.

Previous Post <<||>> Next Post

📊Graphical Representation: Graphics We Live By (Part VIII: List of Items in Power BI)

Graphical Representation Series

Introduction

There are situations in which one needs to visualize only the rating, other values, or ranking of a list of items (e.g. shopping cart, survey items) on a scale (e.g. 1 to 100, 1 to 10) for a given dimension (e.g. country, department). Besides tables, in Power BI there are 3 main visuals that can be used for this purpose: the clustered bar chart, the line chart (aka line graph), respectively the slopegraph:

Main Display Methods

Main Display Methods

For a small list of items and dimension values probably the best choice would be to use a clustered bar chart (see A). If the chart is big enough, one can display also the values as above. However, the more items in the list, respectively values in the dimension, the more space is needed. One can maybe focus then only on a subset of items from the list (e.g. by grouping several items under a category), respectively choose which dimension values to consider. Another important downside of this method is that one needs to remember the color encodings. 

This downside applies also to the next method - the use of a line chart (see B) with categorical data, however applying labels to each line simplifies its navigation and decoding. With line charts the audience can directly see the order of the items, the local and general trends. Moreover, a line chart can better scale with the number of items and dimension values.

The third option (see C), the slopegraph, looks like a line chart though it focuses only on two dimension values (points) and categorizes the line as "down" (downward slope), "neutral" (no change) and "up" (upward slope). For this purpose, one can use parameters fields with measures. Unfortunately, the slopegraph implementation is pretty basic and the labels overlap which makes the graph more difficult to read. Probably, with the new set of changes planned by Microsoft, the use of conditional formatting of lines would allow to implement slope graphs with line charts, creating thus a mix between (B) and (C).

This is one of the cases in which the Y-axis (see B and C) could be broken and start with the meaningful values. 

Table Based Displays

Especially when combined with color encodings (see C & G) to create heatmap-like displays or sparklines (see E), tables can provide an alternative navigation of the same data. The color encodings allow to identify the areas of focus (low, average, or high values), while the sparklines allow to show inline the trends. Ideally, it should be possible to combine the two displays.  

Table Displays and the Aster Plot

One can vary the use of tables. For example, one can display only the deviations from one of the data series (see F), where the values for the other countries are based on AUS. In (G), with the help of visual calculations one can also display values' ranking. 

Pie Charts

Pie charts and their variations appear nowadays almost everywhere. The Aster plot is a variation of the pie charts in which the values are encoded in the height of the pieces. This method was considered because the data used above were encoded in 4 similar plots. Unfortunately, the settings available in Power BI are quite basic - it's not possible to use gradient colors or link the labels as below:

Source Data as Aster Plots

Sankey Diagram

A Sankey diagram is a data visualization method that emphasizes the flow or change from one state (the source) to another (the destination). In theory it could be used to map the items to the dimensions and encode the values in the width of the lines (see I). Unfortunately, the diagram becomes challenging to read because all the lines and most of the labels intersect. Probably this could be solved with more flexible formatting and a rework of the algorithm used for the display of the labels (e.g. align the labels for AUS to the left, while the ones for CAN to the right).

Sankey Diagram

Data Preparation

A variation of the above image with the Aster Plots which contains only the plots was used in ChatGPT to generate the basis data as a table via the following prompts:

• retrieve the labels from the four charts by country and value in a table
• consolidate the values in a matrix table by label country and value

The first step generated 4 tables, which were consolidated in a matrix table in the second step. Frankly, the data generated in the first step should have been enough because using the matrix table required an additional step in DAX.

Here is the data imported in Power BI as the Industries query:

let
    Source = #table({"Label","Australia","Canada","U.S.","Japan"}
, {
 {"Credit card","67","64","66","68"}
, {"Online retail","55","57","48","53"}
, {"Banking","58","53","57","48"}
, {"Mobile phone","62","55","44","48"}
, {"Social media","74","72","62","47"}
, {"Search engine","66","64","56","42"}
, {"Government","52","52","58","39"}
, {"Health insurance","44","48","50","36"}
, {"Media","52","50","39","23"}
, {"Retail store","44","40","33","23"}
, {"Car manufacturing","29","29","26","20"}
, {"Airline/hotel","35","37","29","16"}
, {"Branded manufacturing","36","33","25","16"}
, {"Loyalty program","45","41","32","12"}
, {"Cable","40","39","29","9"}
}
),
    #"Changed Types" = Table.TransformColumnTypes(Source,{{"Australia", Int64.Type}, {"Canada", Int64.Type}, {"U.S.", Number.Type}, {"Japan", Number.Type}})
in
    #"Changed Types"

Transforming (unpivoting) the matrix to a table with the values by country:

IndustriesT = UNION (
    SUMMARIZECOLUMNS(
     Industries[Label]
     , Industries[Australia]
     , "Country", "Australia"
    )
    , SUMMARIZECOLUMNS(
     Industries[Label]
     , Industries[Canada]
     , "Country", "Canada"
    )
    , SUMMARIZECOLUMNS(
     Industries[Label]
     , Industries[U.S.]
     , "Country", "U.S."
    )
    ,  SUMMARIZECOLUMNS(
     Industries[Label]
     , Industries[Japan]
     , "Country", "Japan"
    )
)

Notes:

The slopechart from MAQ Software requires several R language libraries to be installed (see how to install the R language and optionally the RStudio). Run the following scripts, then reopen Power BI Desktop and enable running visual's scripts.

install.packages("XML")
install.packages("htmlwidgets")
install.packages("ggplot2")
install.packages("plotly")

Happy (de)coding!

Previous Post <<||>> Next Post

📊Graphical Representation: Graphics We Live By (Part VII: Reading a Conversion Rates Chart with ChatGPT and Copilot)

Graphical Representation Series

One of the areas where ChatGPT, Copilot and other similar AI-based chatbots can help is in summarizing a chart saved as image. Ideally, the chatbots should be able also to approximate the points from the chart as well (an image is made of pixels and thus areas should be easy to delimit). So, I was wondering how far the chatbots can be used for these purposes. I used first an image copied from the web, though I realized that not all visual elements could be read (e.g. Copilot had issues retrieving the values for some months) and I had no basis data for comparisons to identify how big the deviations are. 

So, I created a chart in Power BI based on the below chart (see original data):

Conversion Rates Dual Axes Chart

Here's the output based on Copilot over several attempts:

	Original data			First attempt		Second attempt		Third attempt		Fourth attempt
Sorting	Month	Conv.	Conv. Rate	Conv.	Conv. Rate	Conv.	Conv. Rate	Conv.	Conv. Rate	Conv.	Conv. Rate
1	Jul	8	4	10	1	10	1	8	4	8	4
2	Aug	280	16	275	15	275	15	275	18	275	18
3	Sep	100	13	225	12	225	10	225	12	225	12
4	Oct	280	14	275	12	275	11	275	11	275	11
5	Nov	90	4	75	5	75	6	75	6	75	6
6	Dec	85	3.5	100	5	100	5	100	5	100	5
7	Jan	70	4.5	50	3	50	3	50	4	50	4
8	Feb	30	1.5	50	3	25	2	50	2.5	50	2.5
9	Mar	70	4	25	1	50	2.5	25	1.5	25	1.5
10	Apr	185	11	200	10	200	10	200	10	200	10
11	May	25	3.5	50	4	50	4	50	3.5	50	3.5
12	Jun	195	4	225	10	225	10	225	11	195	4

Please note that same values were repeated to identify whether the chatbot is able to identify the resemblance correctly. 

Based on the data several tests were made with ChatGPT4o, where the following prompts were given:

1. generate the data behind the chart as excel list (see 'First attempt' in the table)
2. generate the data behind the chart as excel list where blue is represented on the left axis and orange on the right axis (see 'Second attempt');
3. reevaluate the values from the previous output considering that the values for Jul are 8, respectively 4% (see 'Third attempt');
4. reevaluate the values from the previous output considering that the values for Jun are 195, respectively 4% (see 'Fourth attempt');
5. generate the data behind the dual axis chart as list from the last conversation;
6. summarize the chart by revealing trends and other statistics.

From attempt to attempt, the values get closer to the original data, corrections being made, though there are points in the final output that diverge considerably. ChatGPT does generate a table which can be easily copied to MS Excel, and even gives some instructions on how to generate a dual axis chart. 

It's great that one can use data between sessions and improve maybe the output. ChatGPT's summary is better than expected and it leverages the latest corrections. However, the estimations need to improve so that the inferences can be correct. 

Here's the output based on the same prompts in Copilot:

	Original data			First attempt		Second attempt		Third attempt		Fourth attempt
Sorting	Month	Conv.	Conv. Rate	Conv.	Conv. Rate	Conv.	Conv. Rate	Conv.	Conv. Rate	Conv.	Conv. Rate
1	Jul	8	4	300	18	300	18	8	4	8	4
2	Aug	280	16	250	16	250	16	250	16	250	16
3	Sep	100	13	200	14	200	14	200	14	200	14
4	Oct	280	14	150	12	150	12	150	12	150	12
5	Nov	90	4	100	10	100	10	100	10	100	10
6	Dec	85	3.5	50	8	50	8	50	8	50	8
7	Jan	70	4.5	300	18	300	18	300	18	300	18
8	Feb	30	1.5	250	16	250	16	250	16	250	16
9	Mar	70	4	200	14	200	14	200	14	200	14
10	Apr	185	11	150	12	150	12	150	12	150	12
11	May	25	3.5	100	10	100	10	100	10	100	10
12	Jun	195	4	50	8	50	8	50	8	195	4

Copilot's estimations are higher than the ones made by ChatGPT and deviate more from the original data. No reevaluations are done between prompts for the other values. The summary provides information that can be used to complement ChatGPT's output. 

Overall, ChatGPT seems to perform better than Copilot, at least for this example (though we might talk here about different "generations"). Unfortunately, given that the estimations provided by both chatbots deviate considerably from the expectation, the output needs to be revised and corrected, which decreases the usability of such chatbots. In fact, one can use them to generate an initial set of data and correct then the deviations.

The outputs of other chatbots like Google's Gemini or Claude-3-Haiku (via Poe) can't be compared with the ones from ChatGPT or Copilot yet. Claude-3-Haiku does provide estimated values (even with comma), though they deviate considerably from the original data. 

It would be interesting to test how other charts and plots are processed by chatbots, respectively whether the various visual elements (e.g. gridlines, ticks, markers) make a difference.

Previous Post <<||>> Next Post