Autoencoders in Python with Tensorflow/Keras

[yatharthshah]

This is why I keep up with YouTube. This scale of Quality information is what the makers of internet envisioned. Keep up the good work.

[francistembo]

Signed my first contract as a python programmer last Friday. I learnt it all here, Tensorflow, opencv, django, flask....Thanks man. I'll surely pay my gratitude in the future.

[EastSideGameGuy]

I just wanna say that, I started off with your tutorials to get into Machine Learning and boy have I come a far way from when I started your tutorials gave me just what I needed to study on my own and learn these things (also hey I ditched keras/tf for pytorch seems a lot more efficient honestly) but thanks and congratulations on 1 million

[ElksuGuitar]

Congrats sentdex for 1 mil! At around 7:15 you divided x_train by 255.0 for the second time btw, so the values were between 0 and 0.000015

[hemanthkotagiri]

I like how you're putting out really long videos with huge amount of information to follow, even though the rate at which you're uploading is slow. It suffices. Thank you so much, dude! You've been my inspiration since I started programming!

[hola-kxgn]

I really like that you explain every reasoning, for someone who is not that great at understanding things first time this is very helpful.

[hackercop]

7:22 its because x_train is a numpy array and /= isn't supported with numpy (it is supported with regular python btw).

[python]

Man, you find the perfect topics for tutorials!

[killer-whale]

It's actually amazing to see a tutorial maker making mistakes. That way you can learn what errors you might encounter and how to tackle them. Never thought of it that way.

[Fruchtkotzekiddy]

i had this video in my watch-later list since you uploaded it - because i didnt knew what autoencoders where and they sounded pretty hard to understand
but man are you good at explaining.

i learned python from your channel before went to university - and it everything super easy for me - since i already knew how to programm - thanks to you

[iantimmons]

Well done!

For the visualization, you need to use the squeeze() function in Colab

ae_out = autoencoder.predict([ X_test[0].reshape(-1, 28, 28, 1) ])
img = ae_out[0] # predict is done on a vector, and returns a vector, even if its just 1 element, so we still need to grab the 0th
plt.imshow(tf.squeeze(ae_out[0]), cmap="gray")

[kevinshen]

jaw droped when you add noise to the image. this is amazing

[mohammednasir]

Another awesome aspect of autoencoders is, if you take the decoder part and give it a vector of size that matches it's input, you now have a generative model which is called variational autoencoders that you can use to generate images just like GANs.

[fiNitEarth]

I'd be so excited if you were to start a series on chess engines. It would be fun to see how you would go about this, even if the engine wouldn't be that strong!

[CaptJeanPicard]

Thanks for this Video. I would love to see more of these where you explain concepts like autoencoders and transformers visually. Really helpful.

[MaxJr]

Amazing tutorial! Thank you very much for providing so inspiring and high-quality information. Indeed, your machine learning and python tutorial series helped me a lot when I changed my career towards data science.

I am wondering if you are planning to do a new video about variational autoencoders. That would be awesome to see!

[Techbin-ynwr]

14:28 Correction: "to map input to output"

[Alaska-mkok]

Congratulations on 1 million subscribers!

[DimaZheludko]

That division by 255 was a tricky guess. I could not understand what's wrong with your code.
However, losses in the ballpark of e-7 were a good hint. Thanks for that error. It's always nice to learn by other's errors. Especially when you can watch them at 2x speed.
Thanks for your work.

[adempc]

One Million - congratulations! You should get yourself a fancy mug.