When you're writing code, it sounds in the background like airplane cabin white sound and it's relaxing. Nice tutorial! I'm a PyTorch Scholar and since the resources in Udacity aren't good enough for me, I'm watching your very helpful tutorials. Thanks.
Great videos. Best content. Nice animations. The fact that you make your videos like a meme is exceptional and make watching them enjoyable. Keep up the good work! I am your fan :)
This was very helpful. I came here looking for what the squeeze operation does. Was happy to find answers to some other questions in the back of my mind. (what does the -1 mean in resize operations) I will say though, that sometimes the visuals can be a little distracting. (Mainly the pancake one. I was having a hard time listening to what you were saying.) Other than that, this was very helpful. I might watch more of your videos in the future!
Hi there , i followed your tutorials and they seem great!! I implemented in the following manner (Squeeze function in reshape) def flatten_by_reshaping(): rdata = data.reshape(1,-1) rd = rdata.reshape(rdata.shape[1] , ) return rd print(flatten_by_reshaping() , len(flatten_by_reshaping()))
If you only reshape there is a bracket inside the bracket so that it is not a real one Dimensional Tensor but it stays the two-dimensional Tensor that has only the first Zero Index filled?
I couldn't concentrate on what he was talking about, I was thinking about how delicious these pancakes look. I guess I should go get something to eat before I continue.
Hey Akash - It's good to use flatten directly. The method also comes with the start_dim parameter. It is important to understand that flatten is a special case of the more general idea of reshaping. The presentation here was meant to explore the various concepts.
@@MohammedNoureldin a little late answer but 7:06 basically tells it. I think reshape() determines the number of individual elements in the tensor and reshapes it as one axis with 12 elements in it. Sorry if i'm wrong.
When you have a sentence like this: "The primary ingredient we use to produce our product, a function that maps inputs to correct outputs, is data.", it's better to write it as: "The primary ingredient we use to produce our product - a function that maps inputs to correct outputs - is data." (ie. using dashes instead of commas). Otherwise it's confusing and difficult to make out that it's a 'parenthetical statement'.
def flatten(t): t = t.reshape(1, t.size()[0] * t.size()[1]) t = t.squeeze() return t I know this is terribly ineffective but it's something different then all the other ones out there. ;)
Hey Ulm - Nice! I didn't think of passing t.numel(). Passing -1 was on my mind though. :) Also, thanks for introducing the view() alternative. I didn't mention it in the video, but I did include it in the blog post.
@@deeplizard Yes! Thanks, corrected :) It is strange that they'd pick that word 'rank' though, pretty confusing in my opinion. Nevertheless, great videos!
Nice tutorial! It would be really great if you change the backsound while you write the code. Because it is not that comfortable to hear, especially when you hear with earphone.
Thanks the video for the t.reshape(1,-1) is It this the same explanation for numpy using reshape ? By Julu Ahamed : stackoverflow.com/questions/18691084/what-does-1-mean-in-numpy-reshape
Great series but wow that sped up keyboard typing makes it very annoying to go back and go through the video again. Not sure if silence would be any better but it sounds like a rat on amphetamines.
i do not like the comment and the tone "Bakers flat to make goods instead we do it to make intelligence". The snobbish tone of programmers who think they know it all but they are scared because they do not. The analogy is good the way it is said is not
Check out the corresponding blog and other resources for this video at: deeplizard.com/learn/video/fCVuiW9AFzY
thank you so much
thanks deeplizard for making pytorch so easy to understand , an excellent series!!!
Hey Prakhar - You are welcome!
Gratulations on your 100.000 followers here! :)
🥳🥳
Really love the analogy ! And I'm really enjoying the course so far.
Check out the corresponding blog and other resources for this video at: deeplizard.com/learn/video/fCVuiW9AFzY
def flatten(t):
return t.reshape(-1)
When you're writing code, it sounds in the background like airplane cabin white sound and it's relaxing. Nice tutorial! I'm a PyTorch Scholar and since the resources in Udacity aren't good enough for me, I'm watching your very helpful tutorials. Thanks.
Hey Felipe - You are welcome! Definitely appreciate your feedback. It helps a lot to hear weather subtle things make a difference. 🙏
@@deeplizard I'm a Pytorch Scholar too! And share the same feelings as Felipe. Thanks a ton deeplizard :)
Great video! I’m halfway thru the playlist!
best work u guys are doing, lot of hard work u both did ,to get this knowledge , love from India
Great videos. Best content. Nice animations.
The fact that you make your videos like a meme is exceptional and make watching them enjoyable.
Keep up the good work! I am your fan :)
Thank you Hossein!
This was very helpful. I came here looking for what the squeeze operation does. Was happy to find answers to some other questions in the back of my mind. (what does the -1 mean in resize operations) I will say though, that sometimes the visuals can be a little distracting. (Mainly the pancake one. I was having a hard time listening to what you were saying.) Other than that, this was very helpful. I might watch more of your videos in the future!
awesome content and the way you deliver hats off!
Thank you! 🎩
Hi there ,
i followed your tutorials and they seem great!!
I implemented in the following manner (Squeeze function in reshape)
def flatten_by_reshaping():
rdata = data.reshape(1,-1)
rd = rdata.reshape(rdata.shape[1] , )
return rd
print(flatten_by_reshaping() , len(flatten_by_reshaping()))
Make sure you add the "data" parameter to the function signature.
I haven’t installed PyTorch yet, but assuming that it works the same as a list, would the answer be: reshape(1,-1)[0]?
Hey Mick - Nice! I didn't think of this one. It works!
I did the same and it worked =B
It works but it wouldn't be the most straightforward approach. But it's good you came up with this, the more ways the merrier.
t.reshape(t.numel()) or t.reshape(-1)
@deeplizard thank you very much for this video and the full playlist. Can you make a series on NLP with Pytorch ?
Currently working out the details about which direction we'll go in terms of content. Thank you for the suggestion.
What to do if I want to append a tensor with another tensor?
We have cat and stack options. See this one: deeplizard.com/learn/video/kF2AlpykJGY
If you only reshape there is a bracket inside the bracket so that it is not a real one Dimensional Tensor but it stays the two-dimensional Tensor that has only the first Zero Index filled?
Hey Terragon - You are correct! 🚀
Thanks to your great tutorial series with best style of explanation i found on the web! cant wait for the next episodes ;-)
Very good.
i was waiting for the pancakes to be squished 'flat'...
Haha! 🤣
same here , hahahaha
me too .. lol
I couldn't concentrate on what he was talking about, I was thinking about how delicious these pancakes look. I guess I should go get something to eat before I continue.
link for the ted talk? Come on, its expected to have a link.
Agree. Added to the description. Link: th-cam.com/video/aR5N2Jl8k14/w-d-xo.html
6:00 why you are showing food, I am hungry 😅. But it look like delicious :)
Why not to use directly t.flatten() instead of t.reshape(-1) OR t.reshape(1,-1).squeeze() ?
Hey Akash - It's good to use flatten directly. The method also comes with the start_dim parameter. It is important to understand that flatten is a special case of the more general idea of reshaping. The presentation here was meant to explore the various concepts.
@@deeplizard Okay got it, thanks
Thanks a lot! Great video
Hey Daniel - You are welcome!
t=t.reshape(-1)
@Phil Ad why?
@Phil Ad I am flattened by it.
Please explain how does this work.
@@MohammedNoureldin a little late answer but 7:06 basically tells it.
I think reshape() determines the number of individual elements in the tensor and reshapes it as one axis with 12 elements in it. Sorry if i'm wrong.
What's your keyboard? loved the sound too much!
def flatten(t):
numel = t.numel()
return t.reshape(numel)
Hey Erik - I like. 🚀
When you have a sentence like this: "The primary ingredient we use to produce our product, a function that maps inputs to correct outputs, is data.", it's better to write it as: "The primary ingredient we use to produce our product - a function that maps inputs to correct outputs - is data." (ie. using dashes instead of commas). Otherwise it's confusing and difficult to make out that it's a 'parenthetical statement'.
Hey Philippe - I like your point here. The sentence is a bit hard to interpret. I'll keep this in mind going forward. Appreciate your input!
def flatten(t):
t = t.reshape(1, t.size()[0] * t.size()[1])
t = t.squeeze()
return t
I know this is terribly ineffective but it's something different then all the other ones out there. ;)
I like it. It's good practice for us to see different approaches and be able to reason about their validity. Thanks for adding it.
Created a couple of quiz questions that use your example:
deeplizard.com/learn/video/fCVuiW9AFzY
5:45 why you distracting me with food !
🥞
I don't know what the low rumbling sound is, but it is ridiculously soothing. Like doing coding on a star trek starship.
🌌🚀
ted talk www.ted.com/talks/maurice_conti_the_incredible_inventions_of_intuitive_ai?language=en
The other way to flatten the tensor is by using t = t.reshape( -1 ) in flatten(t) function
t.view(t.numel())
or
t.view(-1)
Hey Ulm - Nice! I didn't think of passing t.numel(). Passing -1 was on my mind though. :)
Also, thanks for introducing the view() alternative. I didn't mention it in the video, but I did include it in the blog post.
Keep up the great work! :)
t.reshape(-1)
or
t.reshape(t.numel())
your tutorials really worth than the paid courses. Huge thanks from my heart.
Thank you! Glad these videos are helpful. You are welcome!
t.reshape(12) or t.reshape(-1)
Answer for flattening with reshaping : t.reshape([-1])
t.reshape(t.numel()) or t.reshape(tensor(t.shape).prod().... Either of these should work
t.reshape(3,4)
you made me hungry around the 6th minute
😋🍴🥞
Oof, i think the bing chat can take that picture and be able to tell that the bridge is dangerous to cross right now meaning the future is now🎉
OMG I wish people wouldn't use "rank" in this way. Rank refers to the number of linearly independant columns in a matrix...
Do you have a typo?
@@deeplizard Yes! Thanks, corrected :) It is strange that they'd pick that word 'rank' though, pretty confusing in my opinion. Nevertheless, great videos!
cool
numel = t.numel()
nt = t.reshape(numel)
Nice! Moving fast! 🤖
t.reshape(12)
t.reshape(-1)
t = t.reshape(12) works fine but the other ppl's t = t.reshape(-1) seems way more elegant
Thanks MinJoong! Yes. Agree.
eazy peezy...
t.reshape(-1)
Nice tutorial! It would be really great if you change the backsound while you write the code. Because it is not that comfortable to hear, especially when you hear with earphone.
def flatten():
return t.reshape(-1)
t.reshape(1,-1)[0], better late than never :p
t = t.reshape(12)
print(t.reshape(t.numel()))
Thanks the video for the t.reshape(1,-1) is It this the same explanation for numpy using reshape ?
By Julu Ahamed : stackoverflow.com/questions/18691084/what-does-1-mean-in-numpy-reshape
You are welcome! Yes. That's right.
my answer (t.reshape(-1)). Without looking, I swear
t.reshape(-1)
t.reshape(12)
I'm sorry, zucchini on a pizza ????
reshape(-1,)
Great series but wow that sped up keyboard typing makes it very annoying to go back and go through the video again. Not sure if silence would be any better but it sounds like a rat on amphetamines.
t = t.reshape(t.numel()) :)
i do not like the comment and the tone "Bakers flat to make goods instead we do it to make intelligence". The snobbish tone of programmers who think they know it all but they are scared because they do not. The analogy is good the way it is said is not
Now do scatter()
r.reshape(1,torch.tensor(r.shape).prod()).squeeze() for a tensor r
t.reshape(t.numel())
Looks good Barry!
t.flattern() makes my life simpler😅😂
def flatten(t):
return t.reshape(-1)
t=t.reshape(-1)
t.reshape(-1)
t.reshape(-1)