My new concept was that my program would take a large text, generate a Markov matrix by analyzing it, create a new text, generate a new matrix by analyzing the new text, create a new text from the new matrix, and so on. The text I decided on was Donna Haraway's A Cyborg Manifesto. I found the full text online, cleaned it up (since it had a bunch of whitespace, page numbers and spelling errors), and saved it.
Using your conceptual research journal entry for context, describe the ofxaddon that will be incorporated in your final project.
I am using ofxMarkovChain. I initially had problems implementing it, but I got it to work by looking at this post and changing the name of the check() function to mycheck() instead. After that, I was able to make its two examples work.
I had some trouble writing to a text file, though. I tried it both with ofFile and with ofstream, but no matter what, for some reason it's writing Chinese characters instead of the string I specified. Even using information from the official C++ docs, it's still not working. Stackoverflow says that it's related to the fact that the string's binary values are being recorded and used to locate an address in memory, but I haven't found a fix for it yet, as of May 30, 6:15PM. It seems to me that I'm doing things the way that is being recommended, but for some reason it's not working the same way. Typical programmer troubles, I suppose.
This doesn't work either.
How does the addon support your conceptual experimentation?
The addon is a simple framework through which I can implement a previously created Markov transition matrix.
How does the addon work? What other research/libraries support it?
The addon reads a text file formatted in the style of a square matrix. Each row and column represents one possible state, and each value in each row represents the probability of transitioning to the corresponding state. From there, the addon can generate a sequence of states that can be used to draw to the screen. It uses ofRandom to randomly determine the next state.
It's based on the theory of Markov chains, which are random processes that transition from one state to another based on probabilities that depend solely on the current state. So the probability of, say, state B appearing in a Markov chain is dependent on the previous state and the likelihood of it transitioning to state B. With a transition matrix like:
| State | Probability of trans. to A | Probability of trans. to B | Probability of trans. to C |
|---|---|---|---|
| A | 10% | 60% | 30% |
| B | 50% | 30% | 20% |
| C | 20% | 10% | 70% |
So say you're in state A, you have a large likelihood of transitioning to state B (60%), and vice versa (50%). So whenever state B and A appear, it's very possible that they will alternate back and forth. But if state C appears, it's likely that it will simply stay in state C instead of transitioning to A or B, because the probability of it staying in state C is so high (70%).
Markov chains are named after Andrey Markov
Document a first integration of the addon in your app development.
At this stage, I still haven't gotten the app to write properly to the file, but it's definitely writing. So I implemented the Markov addon in order to generate the output according to a transition matrix that was provided with one of the examples.
The transition matrix only has 5 states at this point, so it'll generate only the letters a through e.
So I have my update() function calling the ofxMarkovChain's update() and getState(), which generates a new state and then returns that state, respectively.
From there I have a switch statement that matches each state to a letter. I then have the program write the generated letters to the file and also print them to the console. Hooray, it works!
And the written files look like this:
Searching "c++ writing chinese to file" led me here, but I don't think the problem is the same as mine. This thread also wasn't that helpful. This post seems more promising, though.
I tried writing to a text file in a non-openframeworks project, and it worked!
So it seems that the problem lies with something in openframeworks, not with my computer or with xcode. Hmm.
To create the transition matrix, I'm using a vector of vectors of floats - the same format as the matrix used in ofxMarkovChain. I will also be using a separate vector to store the characters that correspond to each state.
References:
Markov Models for Text Analysis - http://www.stat.purdue.edu/~mdw/CSOI/MarkovLab.html
A working Markov text generator - http://projects.haykranen.nl/markov/demo/
The source for the text generator - https://github.com/hay/markov
Markov Chain text generator - https://golang.org/doc/codewalk/markov/
Creating a Markov Twitter bot - http://sts10.github.io/blog/2014/12/23/guide-create-markov-twitter-bot/
https://www.cs.umd.edu/class/winter2012/cmsc389C/Projects/P1/P1.html
Stackoverflow Markov advice - http://stackoverflow.com/questions/12065503/predicting-next-char-in-random-text-generation-based-on-some-input-file
Stackoverflow checking for an element in a vector - http://stackoverflow.com/questions/6277646/in-c-check-if-stdvectorstring-contains-a-certain-value
Stackoverflow iterator help - http://stackoverflow.com/questions/15099707/how-to-get-position-of-a-certain-element-in-strings-vector-to-use-it-as-an-inde
Addons:
https://github.com/elaye/ofxMarkovChain
https://github.com/andrebaltaz/ofxHMM
Artists:
Bruce Ellis and Rob Pike - Mark V. Shaney
No comments:
Post a Comment