Defining Specific Text Colours about word_cloud

[markklick206]

@irn-bru @amueller
Here is the code for what @irn-bru is looking for:

def setListOfcolor_func(word=None, font_size=None,  
                     position=None, orientation=None,  
                     font_path=None, random_state=None):  
    #define the list of set colors  
    color_list = ["red", "#FFFFFF", "#000000", "skyblue", "yellow"]  

    #return a random color in the list  
    return np.random.choice(color_list)  

# visulaize wordcloud
wordcloud = WordCloud(color_func=setListOfcolor_func).generate(YOUR_TEXT)

Thanks for writing such a helpful package @amueller
Cheers,

• @markklick206

[amueller]

Look at the example here:
http://amueller.github.io/word_cloud/auto_examples/a_new_hope.html
And the references here:
http://amueller.github.io/word_cloud/generated/wordcloud.WordCloud.html#wordcloud.WordCloud

You can either pass a color_func to the constructor or recolor an existing word cloud.

The function must look like this:

def red_color_func(word, font_size, position, orientation, random_state=None):
    return "hsl(10, 100%%, %d%%)" % random.randint(40, 100)

(this one will give you different lightnesses of a particular red hue)

[amueller]

Closing, let me know if you have additional questions.

[amueller]

[example output]

[irn-bru]

Thanks for the reply, amueller.

Yes, that was the template I was working from and had successfully managed to adjust the colours like you described. What I wanted to do though, was be able to choose the exact individual colours (either via RGB, HEX values or another way) and your script would then use ONLY the colours I chose when rendering the final image instead of just applying a hue.

There is a word cloud generator over at http://www.wordle.net/create which lets you pick the colour(s) you wish to use and renders the image with those but the reason I prefer your script is because I can use a mask which for me is a must.

[amueller]

You can just define a list, and let the function return a random item of the list.

[irn-bru]

That sounds about right. Next question is how would I go about it?! (Sorry, I said I was new to Python)

[amueller]

Can you be more specific in your question?

[irn-bru]

Apologies, I was wondering what code I would use to define a list of specific colours I'd like the script to use. Like if I only wanted #FFFFFF and #000000 colours to be used in the text.

[amueller]

Not sure what color format PIL (which I use for drawing) accepts, but possibly just
np.random.choice(["red", "black"])

[irn-bru]

Thanks but I haven't a clue on how to get it working to test it. I added the 'import numpy as np' line and added 'np.random.choice(["red", "black"])' but nothing changed. This is in the 'a new hope' example script... am I right in thinking the code has to get added to there?!

Again, apologies for my severe lack of knowledge.

[abubelinha]

It could be interesting that wordcloud directly accepts combined structured data (pandas, dictionaries or whatever) including word+frequency+color all at once, without having to deal with custom functions.

word	frequency	color
USA	25	red
China	25	yellow
France	8	green
Kenia	8	black
Australia	6	blue
Japan	6	yellow
Brazil	3	red
Morocco	3	black
New Zealand	3	blue
Portugal	3	green
...	...	...

(i.e. a long list of all world countries grouped by just 5 colors: i.e. the Olympic ring colors for each continent)

Pandas or database sources let you easily make joins to set colors depending on whatever criteria.
So I think there might be many similarly-structured datasets needing to use an approach like this.
It would be great to feed this directly to wordcloud.

Are there any documented examples already doing something similar?

Thanks @amueller and colaborators!
@abubelinha

[amueller]

@abubelinha what's the issue with the function, you can write a function that takes that dict, right?

[abubelinha]

@abubelinha what's the issue with the function, you can write a function that takes that dict, right?

"you can try" is probably more appropriate here (I have low programming skills).
I didn't say there was an issue. But it was a bit difficult for me to understand the example I found.

I will probably come up with a non-efficient solution (wasting memory, using unnecessary loops, whatever).
That's why I suggested it to be a good built-in feature, so people can quickly get using your library from pandas datasets.

That said, I am already working on it.
I started from generate_from_frequencies() example and try to adapt it for dataframe-extracted frequencies.
Does generate_from_frequencies() accept any other parameters than frequencies (i.e. to set colors) or do I need to apply them separately?

Thanks

[abubelinha]

OK, I ended up with something that works (at least with my first tests).
Probably not a good way of coding it, so I paste it here just in case somebody else can correct it or make suggestions.
@amueller I got specially confused about how to deal with random_state and pass it to inner function calls
(I don't even know if that makes any sense).
The function should probably take many other parameters needed to configure wordcloud calls, so I would thank if anyone improves it as much as possible:

def df2wordcloud(df=None, colword=None, colfreq=None, colcolor='color', random_state=1, debug=False):
	""" This takes a pandas dataframe which contains at least two columns (you must pass their names):
		- one column for words, 
		- another for word frequencies 
		- and another (optional) for using fixed colors per word 
		- plus any other columns you need (i.e. to define -or not- other coloring functions)
		If the 3rd column name is not passed, then default random colors are used instead.  
		
		Sorry about the odd debugging stuff. Please improve this code as much as you can:  
		- I wanted to count and output every color_function calls  
		  (couldn't figure out how to make that counter without using global statements).  
		- colordict is created for nothing (I didn't know if I would need it for color_function)  
		- To-do: I couldn't figure out how to pass a random_state to this function and 
		  propagate its value to inner function calls.  
		"""
	from wordcloud import WordCloud,random_color_func
	if df is None or colword is None or colfreq is None:
		return
	if debug: 
		# print("\nSOURCE DATAFRAME as a dict:\n\n", df.to_dict())
		# print("\nSOURCE DATAFRAME as a list of lists:\n\n")
		# print("pd.DataFrame(\n",df.values.tolist(),", \n columns = ",df.columns.tolist(), " )")
		print("\nSOURCE DATAFRAME:\n\n", df)
		
	wc = WordCloud(background_color="white", max_words=1000, scale=2, random_state=random_state)
	freqdict = dict(zip(df[colword], df[colfreq].astype(int)))
	if debug: 
		print("\nSOURCE FREQUENCES DICTIONARY (length={}):\n\n".format(len(freqdict)), freqdict)
	if colcolor is not None: 
		colordict = dict(zip(df[colword], df[colcolor]))
		if debug: print("\nSOURCE COLORS DICTIONARY (length={}):\n\n".format(len(colordict)), colordict)
	if debug: 
		print("\n","- "*50)
	wc.generate_from_frequencies(freqdict)
	global counter
	counter=0
	if colcolor is not None: # use color function that takes color from dataframe color column
		def my_color_func(word, font_size, position, orientation, font_path, random_state=1):
			color = df[df[colword]==word][colcolor].item()
			N = df[df[colword]==word][colfreq].item()
			if debug:
				global counter
				counter+=1
				print("{}. '{}' [N={}] to wordcloud with color --> '{}'".format(counter,word,N, color))
			return(color)
		wc.recolor(color_func=my_color_func)
	else: # use default random colors
		def my_color_func_default(word, font_size, position, orientation, font_path, random_state=1):
			N = df[df[colword]==word][colfreq].item()
			if debug:
				global counter
				counter+=1
				print("{}. '{}' [N={}] to wordcloud with default color".format(counter,word,N))
			return random_color_func(word, font_size, position, orientation, font_path, random_state)
		wc.recolor(color_func=my_color_func_default)
		pass 
	if debug: 
		print("\n --- I colored {} words out of {} --- \n".format(counter,len(df)))
	import matplotlib.pyplot as plt
	plt.imshow(wc, interpolation="bilinear")
	plt.axis("off")
	plt.show()

[amueller]

At this point, you can probably ask gpt-4 to write the code for you. The problem with creating a built-in function is that it would solve exactly your use-case, for a particular way of formatting the dataframe etc, while making the code for wordcloud even more complicated.

[abubelinha]

I don't understand what you mean with "particular way of formatting the dataframe" (unless you mean non-pandas dataframes).
The function I posted was not to solve my use-case. Actually, I haven't used it for anything yet.
I tried to design it to take any pandas dataframe. You must specify which columns contain the information for your wordcloud and which colours you want.

As the idea was somehow related to this issue title I just posted it here so anyone can reuse it (not expecting it to be part of this package).
It was more an exercise I did trying to understand how your library worked, so I could have a function for future usage (a python way to quickly create wordclouds from spreadsheet data with predefined colors).

Again, sorry about my dirty bad coding and debugging skills.
Anyway, it kinda worked but I couldn't figure out how to deal with random_state and pass it to inner function calls.
Could you help? That and gpt-4 both go far beyond my understanding.

Thanks

[amueller]

I was commenting on this:

That's why I suggested it to be a good built-in feature, so people can quickly get using your library from pandas datasets.