yeah, and he should show us the money he earned from the video "how to predict stock prices"... most of his videos are about non-feasible stuff or stuff that requires time and dedication and not a couple of evenings of hacking. other channels have a github account, here you see just chunks of non-working sample code, if anything at all.
This is the one channel I don't need to set the speed to 1.25x or 1.5x!! I really enjoy the enthusiasm, and I usually rewind to focus on just the code the second time.
Your videos are improving--this one explains more of the "why" than your usual focus on the "what" without much explanation. I would still recommend spending more time explaining why you're making various design decisions, otherwise you're not really teaching any of the core principles at hand. For example, why did you use an RBM or Gibbs sampling? What do they conceptually represent? What other options did you have? Also, it seems odd that you don't consistently share the actual generated results of your models. Don't leave us hanging :)
Hey! good point. but I like it how he throws just the topics. That gives a great head start. The rest of it can be explored in more detailed but dry videos. At least his videos never bore you.
Dude you are super awesome. And your sense of humor is beyond even the most popular of typical RUclips stars. It's awesome that you recognize the significance of WaveNet's achievement in music generation -- It got buried beneath the media hype of "Daddy's Car" which is much less of a breakthrough. Seems like the general public doesn't realize the potential of WaveNet for musical composition and/or synthesis simply because it wasn't branded that way.
Greetings from India. I am an machine learning enthusiast as well, working on getting better and your channel couldn't get any more inspiring for me sir. Great job!
wow tensorflow can be applied to so many things, it's too cool, it's clear that you're an expert in this area, I'd like to see you teach machine learning in a more leisurely way from the basics to the complex.
Hi, Siraj It's two months I started learning python and I am pursuing bachelors in Computer Science . I am totally into music production using Ableton Live. I found these things interesting, Can you guide me from where to begin with ? Thank you
awesome! now this is something i can rely upon to get into deep learning and can fit into my busy schedule. keep it coming siraj amazing videos. n bow to your sense of humor
its a script you need to put in the dir with it; its linked in the description: github.com/llSourcell/Music_Generator_Demo/blob/master/midi_manipulation.py
I've watched many programming videos, and usually I speed them up to 1.25 or 1.5 but with your videos sometimes I have to slow them down. Don't get me wrong I love the fast pace of it, not just watching you type, just explain the code. Also the content is awesome.
Your videos are great! Just a real quick suggestion. Can you create like a workshop stuff, so you can livestream and we code with you. Also installing Tensorflow is a big pain depending the OS. Way to go! Congrats for this amazing channel!
Hey Siraj, I've been trying to port this for a few days to python 3 and I can't. The only problem I have with learning from you is getting a set of running code. Most of the time I can't install the dependencies you use because they are unsupported via my version of python. Are there instructions on how to install your exact python environment? Thank you!
When I ran rbm_chords.py I got the following error : File "rbm_chords.py", line 30 print "{} songs processed".format(len(songs)) ^ SyntaxError: invalid syntax
ok so i tried running the code provided at the github link but i keep getting errors like there is no midi_manipulation or there is no module named midi how do i fix this
I know you are playing the "Short Attention Span" card here but that probably applies more to pop music and cat videos... if someone is geeking out on your video they're most likely not of that mindset, you could slow down a bit and explain more, without loosing the "SAP" theme, these are good videos and you should have more subs but when I watch them I feel like you're taking the piss a bit and that could be why your subs are so low
I wonder if you can use a CNN to come up with a compact parametric model for realistic sounding pianos (and other acoustic instruments) instead of depending on 50GB of wav samples as we do now.
Thank you Nicholas, these guys don't understand it is a video. Anyone can pause and replay any part, so it's BS that speed is an issue. Do not accuse me of not wanting more Siraj content. I've seen every video and liked each one. Siraj gives quality explanation on every topic, and doesn't need to cater to some people who can't figure out the RUclips video player.
Hi Siraj, have you thought about semiotics as it applies to music? I'm surprised you're working with wavs and actual sound when I think the semantic content could mostly be represented in MIDI values. Sure, different instruments or timbres convey a kind of meaning, but I think the heavy lifting is going to be in the melody. Of course, if you're more of a Zimmer guy than you are a John Williams guy, then meaning won't be in the melody as much. But keep in mind, Zimmer's music *fits* into the (limited) environment bounded by the fact that some timbres can't be generated electronically (trumpets, saxophones) as compared to others (drums). I'll check my wallet this weekend and see what I can do to help with your Patreon.
Is there any website which lists most of the state of the art approaches for different fields in artificial intelligence? A single website which would link to all the state of the art papers.
Thought I love the videos, I really wish you used Python 3 in your code examples. Everything is moving towards Python 3, and it's especially frustrating for us Windows users who can't use TF on Python 2 even if we tried. As a result it takes more time to convert things to work on Python 3 than to learn what is the actual algorithm in the example :(
Hi Siraj, Thanks for the awesome videos. I looked for the midi package and there were a few, so I wasn't sure which one to get. Could you let me know which package you used?
Any tips for running this in python3? I'm working on porting the library for the midi manipulation but that appears to be the only one that's incompatible right now.
Love your tutorials. So concise and elegant. Been binge watching and trying them out whole day. Would it be possible to have something like "Book generation" based on author profiling using DeepMind's CNN kinda approach or the current RBM one. Where one could feed in a few books by some author and the system would output maybe some paragraphs of text akin to the author's writing style??
There's super simple ways to do this in natural language by using n-grams. You should look into the nltk library and using n-gram (a sequence of words) to generate texts. stackoverflow.com/questions/1150144/generating-random-sentences-from-custom-text-in-pythons-nltk
Hi when i put the gibbs_sample i always got this message below. what's problem? please help Traceback (most recent call last): File "C:/Users/user/PycharmProjects/untitled3/test.py", line 24, in x_sample = gibbs_sample(1)
Hey Siraj could you help me out please? when i run the script i always get invalid syntax sometimes other errors on line 82 or 77 the parameters are missing
hey siraj i am trying to run the code i fixed most of the errors i faced windows and python 3.5 :( so.... but still not able to get python midi to work some error due to 3.5 python but tensorflow doesnt work on windows python 2.7 help!!!!!!!!
hey siraj i m not able to install midi on my mac i m geting following error print "No sequencer available for '%s' platform." % platform ^ SyntaxError: Missing parentheses in call to 'print' help me to figure it out
This is because that module is a python 2 package, while you are running python 3. This is a key difference between the two versions. Python2.X = print "statement" Python3.X = print("statemet")
I ran into the same problem. Try this repo: github.com/louisabraham/python3-midi. There was another issue that I had to fix locally but then it worked!
I just spent a whole day trying to get everything to work, and I realized that Siraj's script is for Python 2, not 3. However, Tensorflow does not work on Python 2, so wtf.....
you need github.com/llSourcell/Music_Generator_Demo/blob/master/midi_manipulation.py to resolve the midi_manipulation dependency ; awesome stuff @Siraj ( as always :p )
File "rbm_chords.py", line 82, in gibbs_sample [_, _, x_sample] = control_flow_ops.While(lambda count, num_iter, *args: count < num_iter, AttributeError: 'module' object has no attribute 'While' Im not seeing 'Where' in the control_flow_ops. Am I reading this wrong? www.tensorflow.org/api_docs/python/control_flow_ops/control_flow_operations
Marcus Dushshantha Chandradasa It was a dependency issue I was having within a Conda env. I gave up and set up tensorflow using pip. Thanks for the reply.
Mid-eye??? I think you might have meant to say that to be funny but just in case not, its Mid-eee....Having gone to Juilliard I can tell you that this approach to composition never really came into the curriculum-- the hard part is composing something that's pleasing to the ear.
Thanks Christina! I think that error has to do with my code being made for an older version of tensorflow. Try this repository instead github.com/dshieble/Music_RNN_RBM it will give you similar results and (hopefully) work
Hi Siraj, nice vid!! I tried your codes and dhieble's codes , looks like it get same error : "AttributeError: 'module' object has no attribute 'While'". Any suggestion?
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
![I.N "HALLUCINATION" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/n5B5q1Hwt_U/mqdefault.jpg)
Do a demo at the end of your videos god dammit !
i will def do this more thanks
pls. It would be so interesting to see the demo!
Keep up good work.
agree
yeah, and he should show us the money he earned from the video "how to predict stock prices"... most of his videos are about non-feasible stuff or stuff that requires time and dedication and not a couple of evenings of hacking. other channels have a github account, here you see just chunks of non-working sample code, if anything at all.
鵜里 let me guess... You actually invested in stocks by taking that AI as the base of your investment ?? LOL
This is the one channel I don't need to set the speed to 1.25x or 1.5x!! I really enjoy the enthusiasm, and I usually rewind to focus on just the code the second time.
thanks!! good to get that validation
Your videos are improving--this one explains more of the "why" than your usual focus on the "what" without much explanation. I would still recommend spending more time explaining why you're making various design decisions, otherwise you're not really teaching any of the core principles at hand. For example, why did you use an RBM or Gibbs sampling? What do they conceptually represent? What other options did you have?
Also, it seems odd that you don't consistently share the actual generated results of your models. Don't leave us hanging :)
very, very good feedback. Thank you Cameron. I will think hard about this
^This
+1!
Cameron Behar bang on target 🎯
Hey! good point. but I like it how he throws just the topics. That gives a great head start. The rest of it can be explored in more detailed but dry videos. At least his videos never bore you.
Dude finally someone who talks at the pace I enjoy learning :D Keep the tempo :D
Interesting... I don't understand anything
Dude you are super awesome. And your sense of humor is beyond even the most popular of typical RUclips stars.
It's awesome that you recognize the significance of WaveNet's achievement in music generation -- It got buried beneath the media hype of "Daddy's Car" which is much less of a breakthrough. Seems like the general public doesn't realize the potential of WaveNet for musical composition and/or synthesis simply because it wasn't branded that way.
thanks!! Wow! deepwave definitely needed better branding you're right
what exactly makes replicating wavenet expensive? using raw sound vs. midi?
Well... today was productive for me because I discovered your channel! Thank you.
woot welcome!
Greetings from India. I am an machine learning enthusiast as well, working on getting better and your channel couldn't get any more inspiring for me sir. Great job!
hey Ashwin, so glad to hear it :)
Can I contact you via email? I could probably use your inputs
the opening and motivation for the video is dope
Thanks Sandeep! I'm playing around with the video style a bit more these days
wow tensorflow can be applied to so many things, it's too cool, it's clear that you're an expert in this area, I'd like to see you teach machine learning in a more leisurely way from the basics to the complex.
glass of water guy with another fascinating video
You guys are trying to make us lost our job!! DAMNNN!! Y'all know I am majoring in music composition now.
dont worry it will help existing composers get better
Hi, Siraj It's two months I started learning python and I am pursuing bachelors in Computer Science . I am totally into music production using Ableton Live.
I found these things interesting, Can you guide me from where to begin with ?
Thank you
What a pace. You got a Tension Flow !
What dependency do you need to import the midi module in the midi_manipulation.py script?
same
up
i love the content you provide,
IMO 5 ish min is perfect
thanks Florian :)
Sirajology x
awesome! now this is something i can rely upon to get into deep learning and can fit into my busy schedule. keep it coming siraj amazing videos. n bow to your sense of humor
Thanks Swaroop. I bow to your busy schedule
what dependency should be installed for using the midi_manipulation class
i have the same doubt
anyone can help please
its a script you need to put in the dir with it; its linked in the description: github.com/llSourcell/Music_Generator_Demo/blob/master/midi_manipulation.py
What dependency do you need for the import midi module in the midi_manipulation.py script?
same problem
its always give ' no module name midi_manipulation'
I've watched many programming videos, and usually I speed them up to 1.25 or 1.5 but with your videos sometimes I have to slow them down. Don't get me wrong I love the fast pace of it, not just watching you type, just explain the code. Also the content is awesome.
Great video man! Just earned a sub
Thank god for creating subtitles. My deep learning algorithms of my brain is still not evolved to analyse and understand you video at that speed.
Your videos are great! Just a real quick suggestion. Can you create like a workshop stuff, so you can livestream and we code with you. Also installing Tensorflow is a big pain depending the OS.
Way to go! Congrats for this amazing channel!
i live stream every wednesday at 10 AM thanks!
Hey Siraj, I've been trying to port this for a few days to python 3 and I can't. The only problem I have with learning from you is getting a set of running code. Most of the time I can't install the dependencies you use because they are unsupported via my version of python. Are there instructions on how to install your exact python environment? Thank you!
You can use anaconda to create a different Python environment.
@@anuragsharma1953 I nearly lost my mind before I saw your comment. Thanks bro
Siraj: Am a fan of Hans Zimmer
Me: Were is the like button?
When I ran rbm_chords.py I got the following error :
File "rbm_chords.py", line 30
print "{} songs processed".format(len(songs))
^
SyntaxError: invalid syntax
ok so i tried running the code provided at the github link but i keep getting errors like there is no midi_manipulation or there is no module named midi how do i fix this
ok so i am still having this problem does any one know why?? the rest of the code seems to work (im just copying in the code to python IDLE 3.5)
The usual way is to install the Python libraries you need with pip install.
I know you are playing the "Short Attention Span" card here but that probably applies more to pop music and cat videos... if someone is geeking out on your video they're most likely not of that mindset, you could slow down a bit and explain more, without loosing the "SAP" theme, these are good videos and you should have more subs but when I watch them I feel like you're taking the piss a bit and that could be why your subs are so low
hmm. i will try to slow down more thank you :)
Well explain more for sure, but no slow down please :3
I wonder if you can use a CNN to come up with a compact parametric model for realistic sounding pianos (and other acoustic instruments) instead of depending on 50GB of wav samples as we do now.
Try pianoteq. It's not perfect, but it's really close to realistic and it's only a couple of mb in size.
THIS is the FUTURE !
Jesus, I love your videos but please slow down just a tad! There is no way I'm retaining half of this.
you can watch it at .25 speed. I personally like his speed as it is.
The audio doesn't even work at 0.25 speed.
Anthony wtf are you doing? Do you want more Sirajology content or less? Slowing down can mean further explanation...
I just watch the videos multiple times to improve comprehension.
Thank you Nicholas, these guys don't understand it is a video. Anyone can pause and replay any part, so it's BS that speed is an issue.
Do not accuse me of not wanting more Siraj content. I've seen every video and liked each one. Siraj gives quality explanation on every topic, and doesn't need to cater to some people who can't figure out the RUclips video player.
Hi Siraj, have you thought about semiotics as it applies to music? I'm surprised you're working with wavs and actual sound when I think the semantic content could mostly be represented in MIDI values. Sure, different instruments or timbres convey a kind of meaning, but I think the heavy lifting is going to be in the melody.
Of course, if you're more of a Zimmer guy than you are a John Williams guy, then meaning won't be in the melody as much. But keep in mind, Zimmer's music *fits* into the (limited) environment bounded by the fact that some timbres can't be generated electronically (trumpets, saxophones) as compared to others (drums).
I'll check my wallet this weekend and see what I can do to help with your Patreon.
Is there any website which lists most of the state of the art approaches for different fields in artificial intelligence? A single website which would link to all the state of the art papers.
Thought I love the videos, I really wish you used Python 3 in your code examples. Everything is moving towards Python 3, and it's especially frustrating for us Windows users who can't use TF on Python 2 even if we tried. As a result it takes more time to convert things to work on Python 3 than to learn what is the actual algorithm in the example :(
Also note that Python 2 officially ends in 2020, so in 2.5 years everyone is going to have to switch anyway.
Is it possible to implement an autotuner for tts so then you can have natural sounding speech?
Hi Siraj, Thanks for the awesome videos. I looked for the midi package and there were a few, so I wasn't sure which one to get. Could you let me know which package you used?
any of them work, they are all parts of the same song thanks Babak!
How can I combine music and lyrics to create a song this way?
Sirajology you are the coolest machine teacher (those who make the machine learn)
thanks so much!
Where does the gibbs_sample function come from?
github.com/llSourcell/Music_Generator_Demo/blob/7de383489cc609fe1d1ef5a029b2c0f44c9f73e6/rbm_chords.py
Hey Siraj, the channel is the nuts. Love the energy, ML focus, and gift to gab showcasing. Why no requirements.txt file for the dependencies?
@3:18 What's up with the chastity belt hahahaha
restricted!!
can you please tell why the size of visible layer is 2*note_range*num_timesteps ?
Is there a synth that mixes sounds through combining the sounds at the microscopic level? Not “n synth” through another
Now that was dope!
and so are you! thanks
Any tips for running this in python3? I'm working on porting the library for the midi manipulation but that appears to be the only one that's incompatible right now.
yea the midi library is the real crux, if you port it, post about it in the comments people will be very happy thanks
Did you port it to python3?
It creates 9 different short outputs. How can I make long single song, around one minute maybe?
Love your tutorials. So concise and elegant. Been binge watching and trying them out whole day.
Would it be possible to have something like "Book generation" based on author profiling using DeepMind's CNN kinda approach or the current RBM one. Where one could feed in a few books by some author and the system would output maybe some paragraphs of text akin to the author's writing style??
There's super simple ways to do this in natural language by using n-grams. You should look into the nltk library and using n-gram (a sequence of words) to generate texts. stackoverflow.com/questions/1150144/generating-random-sentences-from-custom-text-in-pythons-nltk
Thanks Jon. I'll give it a shot. :)
thanks so much!! yes absolutely
Hi, I have just joint as a member of your channel. Can you guide me how to access more resources for members? Thank you!
nice video! i am testing it now :D
May I suggest you build a neural network that picks clothing for you?
No. I don't wanna see cats on his shirt.
ha!!! that is a startup idea right there for sure
What's going on with audio quality of this video about audio?
Hey Siraj, Awesome video. One question how can I combine those generated midi files?
Hi when i put the gibbs_sample i always got this message below. what's problem? please help
Traceback (most recent call last):
File "C:/Users/user/PycharmProjects/untitled3/test.py", line 24, in
x_sample = gibbs_sample(1)
Hey siraj where do you get all those massive datasets from? do you have some site with data for every topic or what? :)
Btw nice videos
can you explain how to converted the midi files to msgpack. Thanks for this video
Happy , upbeat song? Oh boi, you pissed all the metalheads out there.
This is the error while running the code...
any patch to this ?
AttributeError: 'module' object has no attribute 'While'
Hey Siraj could you help me out please? when i run the script i always get invalid syntax sometimes other errors on line 82 or 77 the parameters are missing
make an issue in the repo someone will get to it soon thanks
hey siraj i am trying to run the code i fixed most of the errors i faced windows and python 3.5 :( so....
but still not able to get python midi to work some error due to 3.5 python but tensorflow doesnt work on windows python 2.7 help!!!!!!!!
that too is for 2.7 :(
yeah I'm having the same issue, not really sure what to install for the midi file handling
midi istall shows error for me :(
same
if resolve please help me
hey siraj i m not able to install midi on my mac i m geting following error
print "No sequencer available for '%s' platform." % platform
^
SyntaxError: Missing parentheses in call to 'print'
help me to figure it out
Use parenthesis with print function. print ("No sequencer available for '%s' platform." % platform)
This is because that module is a python 2 package, while you are running python 3. This is a key difference between the two versions.
Python2.X = print "statement"
Python3.X = print("statemet")
I ran into the same problem. Try this repo: github.com/louisabraham/python3-midi. There was another issue that I had to fix locally but then it worked!
I just spent a whole day trying to get everything to work, and I realized that Siraj's script is for Python 2, not 3. However, Tensorflow does not work on Python 2, so wtf.....
Just copy your code into a Python 3.X IDE. Run it and fix the handful of errors that will occur due to version differences.
How can i create a data if i can't find that data any where
Ahahahah you are so funny! Ending with the silver hair dye note was fabulous. I have a question, what's your favorite platform to write in python?
and this is awesome, thank you
That thing you're wearing is very distracting 😨
Love the content though 🤟
i want that shirt - so much swag
Great presentation. Is it fair to say that wavenet is a reccurrent CNN?
its a convolutional recurrent network. which is rare, you usually don't see those two combined. great results though
to answer your question, yes
Hey! I admire the content in your videos. Is there any chance for you to upload work on RNN's and video processing similar to CNN's?
yes i can put that in my queue :)
Great video!
thanks!
THAT RESTRICTED PART!! LMAO
lmaoooo
Wait so how do you train it?
Hey can you direct me to a paper/study material that explains the math behind the code
see this cs224d.stanford.edu/reports/NayebiAran.pdf
Does anybody who know who won the coding challenge for this? Can't find the code anywhere. It's to create an 'upbeat' midi track with this method
Is this the unofficial Youggle? YouKaggle?
Can anyone tell me how do I control the length of the generated MIDIs?
low level stuff see stackoverflow.com/questions/11059801/how-can-i-write-a-midi-file-with-python
you need github.com/llSourcell/Music_Generator_Demo/blob/master/midi_manipulation.py to resolve the midi_manipulation dependency ; awesome stuff @Siraj ( as always :p )
Now that you talk slower the videos aren't under 5 minutes anymore LOL. But still awesome video!
LOL thanks!
super awesome!!!
thanks u2
Why you dont show us as result what you have explained?
File "rbm_chords.py", line 82, in gibbs_sample
[_, _, x_sample] = control_flow_ops.While(lambda count, num_iter, *args: count < num_iter,
AttributeError: 'module' object has no attribute 'While'
Im not seeing 'Where' in the control_flow_ops. Am I reading this wrong?
www.tensorflow.org/api_docs/python/control_flow_ops/control_flow_operations
did you get this fixed? I think you have to use .while_loop instead of While. But the arguments are different
Marcus Dushshantha Chandradasa
It was a dependency issue I was having within a Conda env. I gave up and set up tensorflow using pip. Thanks for the reply.
What version of TF are you using if you don't mind telling me? I use 0.12 and looks like While is not in the version.
Mid-eye??? I think you might have meant to say that to be funny but just in case not, its Mid-eee....Having gone to Juilliard I can tell you that this approach to composition never really came into the curriculum-- the hard part is composing something that's pleasing to the ear.
ha you're right. much more work to be done in this field. it could use your expertise!
how do I get midi_manipulator?
its included in the code
Siraj Raval but you first need to install it right?
“In *just* 90 lines.”
Even with 10000 hours you wouldn't be as good as hans zimmer lmao
0:02 most of the "artists" today !! :D
So where is the music ?
What do I do after i get my rbm_chords.py to run
check the folder named "out" for the resultant generated music files
hi, where to buy your top?
mission district SF. tamo design SF
First of all i LOVE your videos
Thanks Christina! I think that error has to do with my code being made for an older version of tensorflow. Try this repository instead github.com/dshieble/Music_RNN_RBM it will give you similar results and (hopefully) work
Hi Siraj, nice vid!!
I tried your codes and dhieble's codes , looks like it get same error : "AttributeError: 'module' object has no attribute 'While'".
Any suggestion?
change the line in rbm_chords.py from "[_, _, x_sample] = control_flow_ops.While(lambda count, num_iter, *args: count < num_iter,gibbs_step, [ct, tf.constant(k), x], 1, False)" to "[_, _, x_sample] = control_flow_ops.while_loop(lambda count, num_iter, *args: count < num_iter, gibbs_step, [ct, tf.constant(k), x])
". This should fix the issue
Good vid!
thank you!
you're amazing.
thanks!
OMFG HIS FACE AT 00:04
lol
code ???
Hey Siraj, I emailed you my chatbot idea, please reply.
All his videos must have been sped up....i put the speed at 0.5. :)
i will slow down
you need more views
hey Andreas you are so right!! Thanks. Please re-share if you found it useful :)
Hello...where can I find song demos ?
www.ingenieriaacustica.cl
protip: watch at 2x speed
dope
#swag
you have that #swag too thanks
Did anyone pause at 3.18 😆😆😅😂
Instead of doing some hypertalk like a speedfreak I would like to hear the music you made with tensorflow, now it's only bla bla
You speak like a genie coming out of the bottle.