How to Implement Autoencoders in Python and Keras || The Encoder
Join The Sound Of AI Slack community:
kzhead.info?even...
Learn how to build autoencoders with Python, Tensorflow, and Keras. In particular, in this video you’ll learn how to build the encoder component of an autoencoder.
Code:
github.com/musikalkemist/gene...
===============================
Interested in hiring me as a consultant/freelancer?
valeriovelardo.com/
Follow Valerio on Facebook:
/ thesoundofai
Connect with Valerio on Linkedin:
/ valeriovelardo
Follow Valerio on Twitter:
/ musikalkemist
===============================
Content
0:00 Intro
2:04 Autoencoder class + constructor
10:45 Implementation of _build and _build_encoder methods
17:21 Adding encoder input
19:26 Adding convolutional layers
31:09 Adding a bottleneck
37:12 Recap of how _build_encoder works
38:26 Implementation of summary method
39:38 Instantiate an (auto)encoder object
43:25 Next up + outro
Was eagerly waiting for this
I hope the wait was worth the while :)
This was really helpful for me to watch, Had to follow the tensorflow functional API tutorial before and learn that as I'm only used to using Sequential but this was very good thanks
I was looking for something just like this 💯💯🔥
I LOOOOVE IT! Beautiful explanation, thanks so much Valerio! Maybe you are Bach himself after all & we all lay in your wake! x
Valerio, I'm right now like entering in the amazing world of AudioAI, I didn't know that there was a difference in the way that software engineers program a model and the way that data Scientist do. Thank you very much for your coding suggestion, and all of these incredible tutorials.
I'm happy you find the tutorials useful!
Great vid as always. Btw are you planning on creating autoencoder based on LSTM?
amazing video!! Thank you so much!
You are my fav teacher, I have great respect for you and your enthusiasm for teaching
Thank you!
@@ValerioVelardoTheSoundofAI You must drink water after certain interval while doing such long lectures
merci beaucoup tu es le meilleur
thank you for your hard work. could you give a comment on what you think: 1) instead of linear layers in the bottle neck, should you put in some lstm ? 2) very interesting how you would do preprocessing data to feed into lstm (I mean reshape, unfold). 3) if you find it useful to experiment and compare with normal linear layers in bottle neck, to insert lstm in the architecture in the following videos, I would be very grateful.
Hey, great video! I have a question- How and where did you decide the number of convolution blocks in the code?
Grandissimo Valerio, hai creato una bellissima playlist. Comunque sarebbe interessante unire molti concetti teorici alla base (anche presenti in articoli) per poi applicarli in pratica. Ad esempio, sarebbe interessante il modo di interpretare delle notazioni relative a livelli di rete neurale e funzioni di loss particolari. Buon lavoro :)
Grazie Davide per il consiglio :)
Man you re amazing🎉
Excellent !
Hello, your channel looks good, I already subscribed
Your videos are very very interesting sir. Some videos related to Speech enhancement using CNN, please...........
It could be useful to use a slide with the auto encoder architecture model in order to understand better which part are you building. Anyway great job!
That's a nice tip - thanks! I'll see if I'll have time to implement it for the video about the decoder.
great video
Yeah, in Python the underscore is a convention to indicate the method is private.
how can we do this for audio music instead of images as input like lyrics(text) to song(audio) with mohammad rafi's voice
A Novel Fast and Secure Approach for Voice Encryption Based on DNA Computing I am a graduate student. Can you help with the deep learning method of voice coding?
Can we use the same code to implement for images
More or less yes.
what does the notation mean if you start the name of a method or function with "_" or even two "__" like in "__init__" or _build_"?
There's a convention in Python to start the name of a method with "_" / "__" for protected and private methods.
@@ValerioVelardoTheSoundofAI ah okay thank you, forgot about that
Hi. Thanks for all the videos. I'm having an issue with my execution of this one. When I run the code I get only one encoder_conv_layer with a total of 50,626 params. I've gone through and can't find my error. It is difficult to debug because the code is running ok. Does anyone have an idea of where I should look?
Have you tried to get the code from the repo and see if it works?
@@ValerioVelardoTheSoundofAI Wow, quick response! Yes. I've read through them side by side multiple times but can't see the problem. Is there a way to search differences between files?
And thanks again for all your work !
@@ValerioVelardoTheSoundofAI I just posted my file to the slack #random
@@fxyz3487 to check the difference between two files you can use the "diff" command in bash (www.geeksforgeeks.org/diff-command-linux-examples/)
Can you please upload an video for face swapping in images
I mainly focus on sound / music in this channel.
White IDE?
Yep, it's easier to read in a video ;)
Why are you using classes and functions in machine learning? Can't you do without them? Very hard to understand the code for me
there are lot of great libraries built by developers for deep learning. thats why. coding your own functions and model will be hectic and will require a lot of mathematics background.