relative_scaling=1 and repeat=True results in large space between words

[changbowen]

Description

New to word cloud... I want to generate one from a few strings and being able to customize the weight of each string. Following the documents I used generate_from_frequencies, relative_scaling=1 and repeat=True. However it results in large space between words. Removing relative_scaling=1 works fine but the font sizes does not match the weight / frequency value.

Steps/Code to Reproduce

wc = wordcloud.WordCloud(
    relative_scaling=1,
    max_words=200,
    height=800,
    width=800,
    prefer_horizontal=1,
    repeat=True,
    ).generate_from_frequencies({
        'Web Hotel': 3,
        'NGINX': 1,
        'hosting': 1,
        'Apache': 1,
        'Docker': 1,
        'Kubernetes': 1,
        'IIS': 1,
        'Node.js': 1,
        'Tomcat': 1,
        'MySQL': 1,
        'MongoDB': 1,
        'SQL Server': 1,
        'PostgreSQL': 1,
    })

wc.to_file('test.png')

Expected Results

An image with words filled in tightly.

Actual Results

Versions

Linux-4.4.0-18362-Microsoft-x86_64-with-Ubuntu-18.04-bionic Python 3.6.9 (default, Nov 7 2019, 10:44:02) [GCC 8.3.0] NumPy 1.18.1 matplotlib 3.1.2 wordcoud 1.6.0.post14+g1fc6868

[amueller]

Can you try increasing max_font_size? Sorry for the slow reply.

[changbowen]

Tried with max_font_size=2000, fonts are larger but still there is the gap between words. Here is the test image generated:

[amueller]

Thanks, I understand the issue now. I guess the current algorithm for repeat is a bit strange. I'm honestly not sure if repeat and relative_scaling=1 make sense together. Would you like all appearances of Web Hotel to have the same size?

Right now what happens is that it renders everything without repetition while respecting the relative frequencies. If there's space left, it then decreases the font size and starts over. That way it's impossible to fill the space with relative_scaling=1. Honestly I'm not sure there is a way to fill the whole space with these requirements. How would you do it?

[changbowen]

What I was trying to do is to "manually" specify some words to have higher "weight" (so they can appear larger) because I don't have a long source text for it to generate from. Is there any way to achieve that?

[amueller]

Well the question is what the size of the word with the highest weight should be when it's plotted the second time.

Let's say you only have two words, say 'two' and 'words', and you want 'two' to be twice as big as 'words'. If the original size of 'two' is 40, the size of 'words' will be 20. Do you now want 'two' to appear several times with size 40 and 'words' with size 20?

I guess that would be possible, though it would be quite different from what wordcloud is doing now. And you'll still end up with a bunch of holes, and the bigger the initial font is, the more holes you'd get.

I assumed you'd want the words to get progressively smaller, mostly because that's what wordcloud does right now. In that case it's unclear to me how to attach sizes to the first repeated word.

[amueller]

That would look like this:

[amueller]

Or like this:

I'm not sure if that's the desired outcome? But I feel like we should probably allow passing lists of frequencies again, which was supported at some point :-/

[changbowen]

Thanks for taking time to test these. Your assumption is right and I do think the fonts should be progressively smaller and thus filling the holes. Something like this:

The image is generated from wordart.com, powerful tool but only when you pay :)

[amueller]

This doesn't preserve the ratios as you can see, so you can already get this ;)