Markovify always outputs "None" with russian corpus about markovify HOT 12 OPEN

@asigalov61

Hey, the collab link is unavailable, could you fix it? I really want to fix the problem for me and your draft might help me.

from markovify.

@markelovstyle I was gonna make a small implementation Colab based on markovify for the Russian corpus so gimme a week or so and I will post a working example.

from markovify.

Thanks for your interest in this library, @wooferclaw. I'm not familiar enough with Russian to be able to diagnose the issue. One step, however, that would help help me and/or a more Russian-familiar community member resolve this: Could you attach a minimal Python script that demonstrates the problem?

from markovify.

@wooferclaw

I would check your text/char encoding first...then I would try well_formed = True and reject_reg = ' '(space for example) to disable rejection of non-standard chars.

@jsvine If you want/interested, I can assist with the Russian and Hebrew languages. Also, wanted to thank you for markovify! Great job. I love it.

from markovify.

Hi @asigalov61, and thank you for the offer and kind words. If you would like to provide examples of, and improvements to, using Markovify in those languages, that'd be great. Feel free to open an issue or PR, or to email me directly ([email protected]).

from markovify.

Hello, how are things going with the Russian-language integration? I really need Markovify in my project, but unfortunately I work with russian corpus

from markovify.

@markelovstyle Hey, I can help with Russian stuff.

Have you tried this?

f.write(TXT_String.encode('utf-8', 'replace'))

markov_text_model = markovify.NewlineText(text, well_formed=False, state_size=markov_chain_state_size)

And then try to use smaller state sizes (i.e 2) and also higher overlaps. Other settings are important too.

Let me know.

from markovify.

@markelovstyle Take a look at my implementation of markovify:
https://github.com/asigalov61/Markovify-Piano/blob/main/Markovify_Piano.ipynb

You are welcome to use the code as it supports a full range of uni-8 chars, which should work fine for Russian as well.

from markovify.

Thanks, it works, but not as well as English. In most cases I get None.

from markovify.

@markelovstyle If it works sometimes it means that you need to adjust the settings of the generator.

Also, the corpus must be properly formatted. This implementation requires sufficiently long sentences and a sufficiently long corpus.

from markovify.

Guys, i appreciate your help very much. @asigalov61 i have tried may encoding variations, none of them worked.

from markovify.

@wooferclaw @markelovstyle

Here, guys. I made a draft version. Works great on my end. Try it out and let me know.
https://colab.research.google.com/drive/1OLagaj21zjV5kxjR5DIU4kx7jHQG8ggt?usp=sharing

from markovify.