Compare Two Histograms in a Single Plot

[brichard1638]

This function proposes to visually compare two distinct Histograms in a single plot. The idea for this function originated from the following article, found on the R-bloggers site:

https://www.r-bloggers.com/2023/09/histograms-with-two-or-more-variables-in-r/

However, the original function provided in the article has been slightly modified, improving its visual effect. The following information is provided in support of this function:

PROPOSED FUNCTION NAME: compHist

FUNCTION STRUCTURE: x1 = numeric vector 1 x2 = numeric vector 2 main = "Plot Title Name"

FUNCTION OUTPUT: A Single Plot Consisting of Two Colorized Histogram Plots with an Overlap Color that Joins Them

Ideally, there should be three additional arguments controlling for Histogram color including the overlap color. However, due to the structure of the function, the rgb function used in the col argument ONLY allows for the creation of a single set of colors. Preliminary testing revealed that changing the rgb values completely compromised the ability to control the overlap color that joins the Histograms.

There may be another means by which to control for histogram colorization including the overlap color. The modified skeleton code used in the construction of this function is provided as follows:

set.seed(249) x1 = rnorm(1000, mean = 0) x2 = rnorm(1000, mean = 2)

compHist(x1, x2, main = "Histogram of rnorm Distributions With Means 0 & 2")

minx = min(x1, x2) maxx = max(x1, x2)

hist(x1, main = main, xlab = "", ylab = "", col = rgb(0, 0, 1, alpha = 0.6), xlim = c(minx, maxx))

hist(x2, xlab = "", ylab = "", col = rgb(1, 0, 0, alpha = 0.6), add = TRUE)

legend("topright", legend = c("Mean: 0", "Mean: 2", "Overlap"), fill = c("lightslateblue", "salmon","mediumvioletred"))

[oobianom]

Working on development. A bit busy these days, but we will get there.

[oobianom]

I worked on this. Then I was thinking if it would be good to also give the user the ability to separate the plots initially before comparing. What do you think? You can test out the function and let me know. Maybe it defeats the purpose of the function in the first place.

[brichard1638]

I tested and reviewed the compHist function which I understand is still under development. Adding the separate argument as a binary value really improves the utility of the function. It makes it easier for the user to use and apply in terms of capturing either separate or compared Histograms. Moving forward, I believe this functionality should remain within the function.

[oobianom]

Thanks Brice, alright we will leave it this way. I will work further on the documentation for it and you may take a look afterwards.

[oobianom]

This function is all set, Brice! You may take a look before I close the issue

[brichard1638]

The compHist function has been tested in quickcode version .6

The compHist function passed the plot test referenced by its corresponding documentation.

However, the compHist function failed the plot test when color variables in the color argument are modified. When the colors are changed, only plot legend colors change. The colors defining the Histogram do not change to match the plot's legend colors.

The following code is provided to reproduce the error: x1 = rnorm(1000, mean = 0) x2 = rnorm(1000, mean = 2)

library(quickcode) compHist( x1 = x1, x2 = x2, title = "Histogram of rnorm Distributions with Means 0 & 2", color = c("red", "blue", "green") )

RECOMMENDATION: It may be easier for the user to discern the difference between combined Histogram colorization and the overlap color by differentiating them within the function. For example, breaking out the color argument to hcol1, hcol2, and overlapCol or a similarly defined term would better clarify these color distinctions. It would also control the use of color within the function as providing more than three colors in the original colors argument should crash the algorithm.

In addition, providing only one color for each itemized argument provides better configuration guidance for the user.

[oobianom]

I have partially addressed this! The overlap color would not be defined by the user, it will be determined based on the overlap between the colors. Take a look when you get a chance.

For the other part, I am leaning towards the recommendation to declare hcol1 and hcol2. However, it may seem like too many argument for the user to enter.

[brichard1638]

I have concluded the tests on the compHist function. This function has passed the unit tests against which it was tested. The following feedback is provided:

• It is strongly recommended that the color argument be broken out into two separate arguments. The problem with a single color argument is that there is nothing preventing the user from entering more than two colors, which distorts the output. When possible, programmatic safeguards should be considered to help guide the user on how to effectively use a function. While it may be obvious to you and I that only two colors are the acceptable configuration for this argument, that is certainly not the case for every user.
• Concerning the idea that breaking out the color argument would represent too many args comprising a function, you really are only adding one more arg to the function. Eight args does not constitute an excessive number supporting a function in R provided they are easy to understand in context to the function in which they are applied. Before making the assumption that eight args is too many args for a function, consider the parameter arguments used in the nnet function found in the nnet package. Depending on how the function is configured, there are between 20-24 args to consider, not including args passed to and from other methods. (Ref: https://cran.r-project.org/web/packages/nnet/nnet.pdf). While it is not the rule, I have personally observed R functions that contain upwards of 80 args!
• The separate argument is a really nice compliment to the compHist function! This arg increases the value of the function by improving its flexibility for showcasing various plot configurations. This is achieved by encoding a single binary value.
• If it's one criticism to be observed in the compHist function, it's that it should be reconsidered as to whether to allow the user to customize the ylab arg. This value should always be labeled "Frequency." An alternative option would be to remove the ability to encode a ylab arg altogether, by adding a default value called "Frequency." This would reduce the number of args for this function by one, which is consistent with a previous concern expressed regarding excessive functional args. While the original function contained a ylab arg, upon closer analysis, it probably should be reconsidered.
• The default value of the ylab arg when applying the hist function is "Frequency."
• Finally, it would be nice to have one to three recommended color-pair options outlined in the documentation from which to choose for encoding the color of the compared Histograms. It is not clear which color combination is the best to use given the nature of this plot. This type of directed information would save the user time by facilitating plot development. It is believed that the user would appreciate being able to quickly select a good color combination without having to take the time to experiment with several color pairings.

[oobianom]

Hey Brice, I separated the color argument. For now, I have left the ylab argument in it, but the default is "Frequency" as you've suggested above. For your final point about one to three recommended color-pair, I agree very much. Do you have suggestion for color-pair that we should add?

[oobianom]

I added the follow now -

Recommended color pairs col1 = #00539C (and) col2 = #EEA47F col1 = brown (and) col2 = beige col1 = pink (and) col2 = #2F3C7E col1 = red (and) col2 = yellow col1 = limegreen (and) col2 = blue col1 = #990011 (and) col2 = #317773

[brichard1638]

I successfully retested the compHist plot based on the latest changes made to the quickcode package. The following feedback is provided:

• When executing the compHist plot in RStudio, any previous plot(s) that exist in the plot tab, are overwritten with the latest plot. The correct behavior is that the plot should be added as another plot within the plot tab. In its current form, each execution of the compHist plot overwrites all other plots contained in the plot tab, if they exist, To reproduce this error, apply the following code: x1 = rnorm(1000, mean = 3) x2 = rnorm(1000, mean = 5) hist(x1) compHist( x1 = x1, x2 = x2, title = "Histogram Comparison", col1 = "darkred", col2 = "aquamarine4", separate = FALSE )
• The overlap color provided in the legend does not match the overlap color provided in the plot. A review of the compHist code revealed that a function referenced in the quickcode package titled, mix.color is used to generate the overlap color. It is strongly recommended that another algorithm be used to generate the overlap color.
• Hex colors used in the Recommended Color Pairs section of the function's documentation are inconsistent as to form. Hex colors should be converted to color strings for ease of reading and use. To convert hex colors to the closest color string provided in R, use the following code: library(DescTools) HexToCol("#00539C")

Using the HexToCol function, hex colors referenced in the documentation have been converted to their closest string color in R:

00539C = dodgerblue4

EEA47F = darksalmon

2F3C7E = royalblue4

990011 = darkred

317773 = aquamarine4

[oobianom]

Thanks for the review. To your point #1: Yes I did this on purpose previously, but I can have it not overwrite. To your point #2: Let me see if I can refine that mix.color function further. To your point #3: This is great and I agree, so I will include this changes

[oobianom]

Updated, Brice!

[brichard1638]

I tested the compHist function in the latest version of quickcode version .6. Each of the three bullet points recently cited as issues have been successfully addressed in this latest package version.

The only thing I would add to the documentation for this function, perhaps under the section Some Recommended Color Pairs would be to add the following statements:

• Hexadecimal values can also be passed
• "For best visual results, col1 should be a dark color and col2 should be passed as a light color. For example, col1 = "black", col2 = "yellow"."

While it's not in your recommended color pair list, the color combination of purple-yellow can also be used to generate a nice plot visualization.

[brichard1638]

Obi:::

Can you publish now? I am looking forward to the .6 version of quickcode!!! Would it be possible to include the archive function assessment in the .7 version of quickcode? I also already have a few new functions ready to be presented for your review in the .7 version.

After .6 is published, I will present them to you.

[oobianom]

Great, Brice! I will add that note. No worries, I will also add purple-yellow.

[oobianom]

Regarding having the archive function assessment in 0.7, that's fine. I eagerly anticipate hearing your ideas! Again, no rush. We will get 0.6 out first sometime next weekend.

Before that, I want to let you know that I also added two functions 'in.range' and 'seq3' functions for 0.6. If you have time before next weekend, you can take a look at them both and review. If you have feedback, you may open new issues so we can address them together.