-
This work was originally done in the 70's. Phd thesis of Thomas Owens. He identified several short melodic patterns in Bird solos, gave each pattern a code, listed how many times each pattern was repeated in each solo. It stands to be the best work on Charlie Parker. And the best Phd thesis ever in any field
Originally Posted by princeplanet
Here is a link:
http://www3.telus.net/sametz/charts/...ens%201974.pdf
-
02-23-2018 12:17 PM
-
Here are some computer generated jokes using machine learning. View the source code here. You can see the blogpost here.
I like my women like I like my coffee. I don't like coffee.
What do you call a political position? A condescending con descending.
What do you call a cow with no legs? Ground beef.
What's the difference between a snowman and a snow woman? Snowballs.
I think the ground beef and snowballs jokes are commonly known jokes. Interesting that a computer, like a human, sometimes steals jokes.
-
Haha, already funnier the most of the comedians in my town... But I'll only go see these robots perform if you stick me on the door and buy me drinks all night... Originally Posted by omphalopsychos
EDIT: ...after reading a few more on the blog post, I must confess I found the "non" jokes to be the most amusing.... sheeet, I could see myself actually getting into this kind of entertainment!
Your momma's so fat she threw his family up in the morning.
What do you call a short sense of humor? A charming saucer.
What do you call a woman who has no legs? Doesn't matter.What do you call a deer with no eyes? No idea.Last edited by princeplanet; 02-23-2018 at 12:47 PM.
-
Yep, got it, read it (most of it anyway) a while back. Norgaard would have been foolish to not have made full use of it. An amazing piece of work, Parker would have been shocked! (and probably non-plussed at the same time)... Originally Posted by Tal_175
This work was originally done in the 70's. Phd thesis of Thomas Owens. He identified several short melodic patterns in Bird solos, gave each pattern a code, listed how many times each pattern was repeated in each solo. It stands to be the best work on Charlie Parker. And the best Phd thesis ever in any field Here is a link:
http://www3.telus.net/sametz/charts/...ens%201974.pdf
-
If you're saying that chess is an easier domain to conquer than Jazz improvisation, that might be true. It probably also demands a very different approach. There are more chess positions than there are sub-atomic particles in the universe. That's infinite for all intents and purposes of designing a real-time game. Originally Posted by princeplanet
The complexity doesn't lie in the size of the space of possibilities but more the complexity of criteria for filtering these possibilities and evaluating the extend to which filtering is successful.
I am not an expert in machine learning nor am an expert in improvisation, BTW. I studied ML at a relatively high level at some point but it wasn't my research field.Last edited by Tal_175; 02-23-2018 at 10:04 PM.
-
Wow, if true, that's quite an astonishing fact, particularly given that the Universe itself is probably infinite! Originally Posted by Tal_175
I still say good Jazz improv has more variables. It's not just note choices. In Chess, it's not counted as a variation if the player moved his piece quickly, slowly, scraped it, knocked into the other pieces, coughed, or smiled while he made the move, etc etc etc etc etc etc etc etc
-
So, I found a midi database of Charlie Parker's solos from the omnibook. It lacks the chord changes but it has CP's lines. I have an interesting idea for a project. If i can join this to a database with the changes, then I could probably train a model on the omnibook to generate Parker-esque solos. Moreover, if I can incorporate the changes, then I would be a able to throw a new tune at the model (i.e. a tune that's not in the omnibook) and have the model produce a parker-esque solo on that too.
You'll have to convince me to prioritize this over just practicing my instrument, but it's something I've been wanting to try for a while. If I do it, I'll certainly share the results here.
-
That would have to be a figure for the observable universe, surely, which is huge but finite. Originally Posted by princeplanet
Wow, if true, that's quite an astonishing fact, particularly given that the Universe itself is probably infinite!
I still say good Jazz improv has more variables. It's not just note choices. In Chess, it's not counted as a variation if the player moved his piece quickly, slowly, scraped it, knocked into the other pieces, coughed, or smiled while he made the move, etc etc etc etc etc etc etc etc
The entire universe is of unknowable size, although we can make theoretical estimates.
The numbers in combinatorics get silly very quickly. That's where our biggest finite values come from that we have a use for. Graham's number is a famous example:
Large numbers - Wikipedia
You might think a googolplex is big but that's just peanuts compared to Graham's number
-
But you absolutely have to do this! Some of us here can help maybe... BTW, where can I find this midi database of the Omni book? Originally Posted by omphalopsychos
-
Man that would be a lot of work. I mean what you suggest is probably a decent first approximation. But your training set may not be large enough. You may need to think about a negative training set (which is easy, I can send you my solos Originally Posted by omphalopsychos
) Training parameters and what learning algorithm to use would involve a lot of trial and errors.
The only realistic way to tackle this is to quit your job, and try to get into a well funded Phd program with a CS Prof who's an amateur jazz musician
Than you'll have lots of time to practice as well.
-
I read somewhere that the estimated number is roughly 2^80. Originally Posted by christianm77
-
That's not even a proper big number lol. You can write it down in x to the y notation. Originally Posted by Tal_175
Chess is pretty measly too, but tops that one easily: Shannon number - Wikipedia
Astronomical numbers are pretty small in the grand scheme of things.
Now: the possible particle states in the observable universe? Now that might be a number I could respect.
You can express this is a the distance you need to travel in an infinite cosmos to find an exact repetition of the observable universe.
-
(2^80)! Originally Posted by christianm77
-
Originally Posted by Tal_175
Oh god, I find a mathematical expression funny. JGO has reached peak nerd.
-
Here's the parker database: ownCloud
Tal, I think it's somewhat feasible. I'm an obsessive hobbyist. I've been able to work on research project in my spare time and they were of (barely) sufficient quality to present at academic conferences. Since I use recurrent neural networks a lot for my work, the coding aspect wouldn't be excessive. I think the most burdensome bit would be in processing the data, but that could be made tolerable.
No way I'd leave my job for a PhD at this point, but I could see if any schools are interested in funding the research. That would certainly motivate me to spend some time on it.
-
Physicists Calculate Number of Parallel Universes
Turns out it's 10^10^16, effectively
or 1010,000,000,000,000,000
Seems kinda small. You don't need the silly arrows.
-
There's probably a hack to be had, check out the Bird soloist in BIAB, it's certainly not one of the better ones, not enough of the stock phrases, but if you put all the 232 phrases or whatever Owens found that crop up a lot, then you can split them mainly into 2 camps, Tonic and Dominant. Further, you divide them by which note the phrase starts with, and another list where the last note is categorised. . You analyse any chord progression and give each bar a T or D along with attendant key of the moment and randomly insert a T or D phrase to begin the solo. From here, according to how each last note moves chromatically to the nearest resolving note into the next chord's phrase, the subsequent phrase is chosen that has the required first note sought.
This is the "entry level" and will sound "shoe-horned". From here you introduce new "rules" to refine the phrase shapes and the way they resolve. Perhaps new short link material needs to be created (Parker used many 4 note groups for short cells), ways to vary how phrases start and end rhythmically etc. Idiomatic use of Blues lines will be an important part, careful analysis will show where Parker often defers to such usage...
Infact, the more you analyse things, the more you actually will be spending useful time "practicing", you're practicing the art of high level improv by dissecting the DNA of the greatest improvisor of all time. By being the programmer you become the Mind of Bird, or at least Robot Bird. This very well might inform your own playing far more than your current practice schedule...
-
Maybe I could use the forum to listen to and rate the generated solos and use that information to inform some loss function. However, I think a basic LSTM without reinforcement learning would work fairly well. Lately I've been doing some work with Generative Adversarial Networks, and that might also be useful to improve the solos. Originally Posted by Tal_175
-
What's the possible number of total single time 32 bar Alto Sax bebop solos allowable under the constructional outlines given on the first Barry Harris DVD?
Obviously a smaller number than the number of possible combinations of chromatic scale notes and rests... I'd be interested to know how much smaller.
But also, because BH, we must allow for triplets.
-
(Not enough to actually work it out.)
-
This is like peanuts compared to just A(4,3): Originally Posted by christianm77
Ackermann function - Wikipedia
PS: None of these are of course large enough numbers to describe the heights of nerdiness this thread is prepared to explore.
-
I think I got it, it's : Originally Posted by christianm77
472,981,913,231,938,122,628,863,640,496,872,110,32 9,81,483,958,402,554,982,109,487,904.0630194839290 37673762345091859840547637637401340673675670139458 91438609860137450430674 bazillion
+/- 3%
-
You ever hear the one about George Dubya where he's briefed that four Brazilan soldiers were sacrificed on the Afghan front one day. "Oh my goodness" says George before turning his head to one of his aids to ask "Remind me, how many in a Bazillion again?"...
(eat my dust, Robo-Comic!)...
-
As I understand it, machine learning involves the machine figuring out the pathway to success. Originally Posted by christianm77
But, you do have to give it an idea of what success is.
When Charlie Parker started playing bop, a lot of people thought it was noise. It took a while before his style was widely seen as a success. I guess that doesn't stop you from codifying it afterwards. But, it seems to me that it can't evolve like the human can.
Miles might be a better example, because he kept changing. How does a computer figure out how to break the rules and take a completely new direction?
-
I say it can't, because it doesn't feel jealous, bored, lofty, competitive etc etc Originally Posted by rpjazzguitar
Still, the ML guys will wanna tell you that at least the manifest semblance of these innately human traits can not only be coded to mimic behaviour based on known experiences from the past, but to mimic the way in which the human seems predisposed to choose any future path, based on the way that humans have chosen paths in the past... or something...
No point arguing that a human can't be replaced - how can it, we are organic and have a subconscious will to survive at a cellular level which a computer can't have -because AI is not interested in that. It seems it's interested mainly in the appearance of intelligence, or even a different kind of intelligence, some of which (worryingly) can be self programmed.
So, is AI the coming Dystopian nightmare? Or will it be humanity's saviour? Imagine a World Government where a central computer makes the only Fair decisions for us all on key international issues (because we finally realise no human can ever be dispassionate, rational or fair). It might even end trade disputes, War, hunger and disease. Or on the other hand.....
Reply With Quote
There will be another you changes
Today, 02:06 AM in Comping, Chords & Chord Progressions