Effective TensorFlow for Non-Experts (Google I/O '17)
ฝัง
- เผยแพร่เมื่อ 18 ต.ค. 2024
- TensorFlow is Google's machine learning framework. In this talk, you will learn how to use TensorFlow effectively. TensorFlow offers high level interfaces like Keras and Estimators, which can be used without being an expert. This talk will show how to implement complex machine learning models and deploy them on any platform that supports TensorFlow.
See all the talks from Google I/O '17 here: goo.gl/D0D4VE
Watch more Android talks at I/O '17 here: goo.gl/c0LWYl
Watch more Chrome talks at I/O '17 here: goo.gl/Q1bFGY
Watch more Firebase talks at I/O '17 here: goo.gl/pmO4Dr
Subscribe to the Google Developers channel: goo.gl/mQyv5L
#io17 #GoogleIO #GoogleIO2017 event: Google I/O 2017; re_ty: Publish;
Treat this like a lecture at school. Take notes, star the words you need to look up, practice. I am certainly a non-expert but this had enormous value for an amateur programmer! Not sure what the comments are complaining about
I don't know about you guys but im a noob when it comes to machine learning. I know absolutely zero, nada, nil and yet Francois explained Keras so beautifully and easily that I finally feel confident in using it. I have been trying to understand Keras all day and nothing online helped more than this guy! Thanks Francois!
For those who complain about this video, the title doesn't say "Introduction to Machine Learning 101". The speakers assume that the audience has a basic understanding of machine learning but not TensorFlow or Keras. I have a basic understanding of machine learning and I can honestly say that this video is helpful.
do you recommend any video to start with. I think you're right
Thanks! You may have just jump started my 3rd career. I owe you one brah.
The high abstraction level by Keras seems like a really good idea to me
We need to begin to develop user friendly web apps for everyday people to take advantage of tensorflow's capabilities with no programming skills but with the extended capability of python scripts in the same software. Data may come from csv files or database. This will empower everyone and not just programers and will speed up the way we work at home or office.
Was thinking of creating something exactly like that.
i agree. that will make ML matter much more!
Sounds cool! Has anyone started with this?
Sounds sweet! Would this growth be too quick? Would we bottleneck ourselves with a lack of processing power?
A lot of the times a phone or a tablet is pretty idle. We like it that way cause the battery isn't drained that fast, on the other hand it is a lot of free processing power.
I think google will control the whole world one day...such an amazing company
Will definitely try this out! Looks really simple! Guess it´s a good start to get some results and dig down from there to understand the 2-3 layers below! We need to go deeper ;)
Soo can you single out the specific element that you're looking for in an image? For example, can it single it where the cherry blossoms are, or does it just know that they are there?
This is an excellent video! It a great overview :D
it's for non experts but it's definitely not for layman
this is a tech conference, of course at least you need to have some basic in tech.
WHich version are you using for tensorflow.keras? i tried to do the import like the example with 1.2.0 rc0 and got
ImportError: cannot import name 'keras'
Could please someone explain why "embedding_column" and "indicator_column" are needed? My google searches were futile.
Hi, panzach..What I understand it is simply different way of changing non-numeric data to numeric.
So for example, embedding will be using a matrix(for eg, shape of 10x1) with random number for each label.
While one hot, will be using 0 or 1 of combination, so is kind of for example, I have three animal names(cat, dog and fish).Most simplest one hot vector will be cat is[0,0] dog will be [0,1] and fish will be [1,1]..
Ultimately, it is always depend on your input shape of your model
Thank you, Sing Lee!
for (cat, dog, fish), one-hot code for cat is (1,0,0), dog is (0,1,0) and therefore fish is (0,0,1). Each class wired to its neuron cell so 3 classes needs 3.
Functions do not exist, example: tf.contrib.train - so i asume, that all this is deprecated. Is there an updated version somewhere? I cannot find a single example on this "experiment" thingy anywhere!
thx
Didn't knew Faramir worked in google ! :)
But if, preset of answers in learning objects are used in this library, and you don't preset somthing, how it works. For examle you have a vacabulary with 100,1000 or may by more preset vacabulary with dependences of what rhis example realy means, as an array elements and you dont realy train your model by answer what is this in vacabulary. How it could works correctly?
Some body can explain to me why python can look like this? layers.LSTM(256)(frame_features)?
def my_function():
def another_function():
...
return another_function # note we are not calling it, we are returning a function itself
out = my_function() # out == another_function
out() # we can call it
my_function()() # we can also remove the variable and just call whatever myfunction returns
thanks, really new to python :(
Wonder how much money Francois makes with this integration of Keras into TF, his thought leadership to practical DL community has been incredible.
Cannot install tensorflow as wrapt is not getting installed
HELP!!!!!!!!!!!!!!!!!!
Hey, I am a 17 years old Student from Germany. Does somebody know a good Tutorial for start learning Tensorflow?
balls nicht viel habe mich ein bisschen mir Tensorflow beschäftigt. Das war es dann aber auch
Hey I have mentioned that in one of other replies:
decent free course on ML prepared by Vincent Vanhoucke, one of the engineers at Google Brain project:
www.udacity.com/course/deep-learning--ud730
Also very good resource: neuralnetworksanddeeplearning.com/
More about ML than tensorflow itself.
Hope this will help ^^
Has anyone been able to wrap a Keras model in a custom Estimator and wrap that in an Experiment and run it? The slide at 36:57 indicates that it can be done, but parts of the code is elided. @fchollet is the code for this available?
Actually, I think it turns out that the Keras model is an Estimator (or at least at the same level of abstraction, since it has its own fit and evaluate). However, still not able to do model.get_estimator() in the experiment_fn function. Perhaps the get_estimator() exists only in the Google internal version so far?
"So what hairstyle were you thinking of? Short...? Medium...?"
"Umm... i'll have both, please."
Let's focus more on the content and less on the looks ;-)
Cristi Neagu 😂👍👍👍👍
'TensorFlow for Non-Experts'.
Just blows through major concepts with ML jargon without explaining anything or what the called TF methods do. Not helpful.
What do you have questions about? There's a lot in there but it's mostly pretty basic terms in your first class or two learning about ML. I enjoyed that he didn't get into the depths of the math involved in the models which causes many talks to be very long.
I think this talk is for people who already know Machine learning but are switching to TensorFlow. That being said it would be nice if he touched a bit on the jargon because nearly everyone I've seen who talks about ML seems to go straight into a paragraph full of hundreds of long complicated words.
Fidars Andersons
Thanks for having mentioned it. saved 40 minutes of my life
non-expert doesn't mean clueless...
He actually is dumbing it down. I completely understand your comment but the simple fact is this stuff is not easy to dumb down. If he had to explain nuance concepts at every turn, his talk would be much longer.
If anybody is trying to play with Tensorflow API r1.3 and Keras API one important thing to mention: code in this presentation is a bit incorrect. First 3 lines should go like this:
from tensorflow.contrib.keras.api import keras
from tensorflow.contrib.keras.api.keras import layers
from tensorflow.contrib.keras.api.keras.applications import InceptionV3
Not sure what branch guys used for this presentation but it must be some cutting edge one Very confusing... ;-/
Filip Dąbrowski that's correct... This is the right way to do it. Broke my head trying to figure what was wrong with my bloody conda setup...
even this dude is nervous
the uncle of Zuckerberg?
its all about how to call methods.... not how those things are working :D but since this is for nonexperts kind of okay
tf's code is very light nowadays. i think google should find a way to simply the concepts. As an extreme example, explaining everything in 5 mins.
Please update the title appropriately
Effective Tensorflow for Non-Experts:
1.) pip uninstall tensorflow
2.) pip install pytorch
what's the requirement for a developer to learn TF? all my life just deal with IT ELSE, FOR loop, SQL and this non expert video is an alien to me
I'd say machine learning and deep learning. I don't use TF yet, but I imagine it as just a practical implementation of ML algorithms. If you know ML and Deep learning, you should be good to go.
I would say familiarise your self with machine learning concepts and algorithsm(not all of them just the popular ones used in productions, unless you have the time lol) and learn calculus 1&2, linear algebra, data analysis and statistics
'And of course, LEGO bricks are very accessible. They're accessible to any human being ages five and above.
And that's really the idea, the ideal, I would say, that we had in mind when we designed Kera API.
We wanted to be the LEGO for deep learning.'
Well this comparison is very bad... LEGO is mostly available in western countries and as well is neither accessible nor affordable at all!
Regarding the cheapest sets start at around 10$ for a single car (!!!!) up to open end you can't express yourself and/or experiment unless you/your parents bought plenty of crazyly expensive packs...
So better don't use this comparison.
Tbh it was so for 'non-experts' that I quit halfway
For newbie I would also recommend to try TFlearn github.com/tflearn/tflearn
I wouldn't. Not only does Keras provide all of the functionality and conciseness of TFlearn, but it has better documentation and a far larger community.
спасибо
hahaha, tensorflow for... Non expert? Really?
Good
10:33 scikit learn
David arquette with Justin bieber hair
not helpful
github.com/tensorflow/ecosystem for anyone that wants that
link to notebook referenced in talk goo.gl/0OgXiL
Nice one google.
s
Not helpful. Just spitting out all the Jargons and APIs.. no details provided like why, what are the options. In fact scaring hell out of audience I believe.
arey sir, yaha kya kr rhe ho?
Aagya tubhi :) there are much better videos go for it .. especially by Martin on Tensorflow without PHD
You still gotta work in this life. Google--fu and TH-cam-fu will get you to the promised land provided you put the work in on the consuming end (you)
Just saying.. I am defiantly the layman in question and this is helpful, maybe a a little daunting , but scary is not the word i would use.
No
The speaker looks really disinterested!
u mean tensor and neuronal...
Not a good explanation...its for those whose already knows how to use Tensor Flow
I did not understand anything, the talk is not structured, jumping from A to Z the back to B, weird crap. I stopped watching after 10mn, waste of time