Author of the book Hands-On #MachineLearning with #ScikitLearn, #Keras and #TensorFlow. Former PM of #YouTube video classification. Founder of telco operator.
Auckland, New Zealand
- Tweets 2,184
- Following 361
- Followers 29,836
- Likes 3,782
Pinned Tweet
My new book just came out! 🎉😊
Kindle & e-book available now, print within 1–2 weeks.
You can get it at: homl.info/
(you'll also find free online content there)
Play with the notebooks at:
github.com/ageron/handson-ml…
Hope you'll find it useful!
70
259
2,255
237,893
Fake demos, CGI vaporware, teleoperated robots barely capable of walking, cars that help you crash, traffic jams under malls (=Hyperloop), buses so low they would crash into speed bumbs... Remove the hype and subsidies and all that's left is lies and dreams of oligarchy.
Optimus is your personal R2D2 / C3PO, but better
It will also transform physical labor in industrial settings
201
379
3,277
256,242
My favorite Python 3.5 feature: the matrix multiplication operator @
👇Python features thread👇
26
390
2,782
I wrote a Colab notebook that introduces differential calculus: derivatives, partial derivatives and gradients: colab.research.google.com/gi…
27
513
2,308
I noticed that DistilBERT loves movies filmed in India, but not in Iraq, so I plotted the result for each country: the resulting map is scary. #aibias
54
317
1,777
This is 100% #JuliaLang code. It trains a dense neural network using the #Keras Python library. If you replace the first 3 lines with "import tensorflow as tf; from tensorflow import keras", you can run the exact same code in Python. Talk about excellent interoperability! 🤝
20
216
1,513
I wrote a Matplotlib tutorial notebook:
colab.research.google.com/gi…
You'll learn how to make beautiful plots and animations, including, most importantly, XKCD-style plots:
21
254
1,246
Scikit-Learn ❤️ Pandas
I wrote a quick tutorial notebook on the new feature-name tracking features in Scikit-Learn 1.0:
homl.info/sklearn-pandas
They're a first step towards full DataFrame support
(Pandas in ⟶ Pandas out).
12
179
1,192
Hurray!!! 🥳 The 2nd edition of my book is available on Kindle (homl.info/amazon2) and eBooks (homl.info/ebook2)! The paper edition should be available within ~2 weeks. 🦎
75
192
1,172
Look what I just received! 🎉 Awesome book by @fchollet, he's put a huge amount of work into this 2nd edition. Excellent clear and intuitive explanations, just like in the 1st edition, plus plenty of great new Deep Learning content. Well done!
16
66
1,090
The salamander on the cover was cute in black & white, but it looks so much cooler in color. I think I might be scared to touch the book. 🥴
37
114
947
If you'd like to learn the @JuliaLanguage and you know Python, then this tutorial Colab notebook is for you. I just updated it to the latest Julia 1.7.1 version:
colab.research.google.com/gi…
Enjoy! 🙂
9
180
913
Cornell's excellent Machine Learning course is entirely online, for free, including all the lectures on YouTube and all the course notes:
cs.cornell.edu/courses/cs478…
That's the bright side of the Internet: so much knowledge freely available! 🙌😀
3
273
938
Just received the first copy of my book (3rd edition), woohoo! 🦎🎉😁
You can get it at:
homl.info/amazon3
Play with the notebooks at:
github.com/ageron/handson-ml…
See what's new at:
homl.info/changes3
Huge tx to everyone who made it happen!!! 🙏
Hope you'll find it useful!
44
111
886
I 💕 #JuliaLang !
It looks & feels like Python, with the speed of C++, the parallelism of Go, batteries included (especially for #DataScience), metaprog like Lisp, great libraries and a welcoming community.
I wrote a Julia tutorial for Python programmers: come & join the party!
A really in-depth #JuliaLang for Pythonistas Colab notebook by @aureliengeron. The notebook will install the Julia environment and @GoogleColab provides free GPUs!
colab.research.google.com/gi…
24
184
849
Is there still an active ML community on X? If there is, the platform is not showing it to me anymore.
The ratio of interesting ML posts in my feed has dropped from perhaps 25% to 1% or less since Musk took over.
I'm muting all the memes and propaganda, but it just won't stop. 💩
96
30
761
75,395
Two more chapters available in the early release! homl.info/er2
* 16: Processing sequences (eg. time series) using RNNs (eg. LSTM, GRU) and CNNs (eg. Wavenet).
* 17: Natural Language Processing using RNNs (Encoder-Decoder) and Attention models (Transformer).
Enjoy! 🦎
18
145
737
Inside a Jupyter notebook:
BAD:
!pip install somelib # may install in the wrong environment!
or
!conda install somelib # same problem
GOOD:
import sys
!{sys.executable} -m pip install somelib
or
!conda install --yes --prefix {sys.prefix} somelib
10
109
702
Two more chapters have just been added to the Early Release of the second edition of my book:
Chapter 10: Intro to Neural Nets with Keras
Chapter 11: Deep Learning Techniques (batch norm, dropout, optimizers, self-normalizing nets, and much more!)
homl.info/er2 🦎
17
129
675
Woohoo! The last 3 chapters are now in the early release:
17. Representation Learning and Generative Learning Using Autoencoders and GANs
18. Reinforcement Learning (includes TF Agents)
19. Training and Deploying TensorFlow Models at Scale
🦎
homl.info/er2
13
109
626
Just finished updating all the notebooks for my book to Scikit-Learn 0.24 and TensorFlow 2.4. Phew... 😅
github.com/ageron/handson-ml…
14
80
612
In @TensorFlow 1.9, it is much easier to use Keras with the Data API: just pass data iterators, specify the number of steps per epoch, and you're good to go! Plus it works in both graph mode and eager mode, kudos to the TF team!
7
147
582
Just pushed 2 more #TensorFlow 2 notebooks! The 1st covers how to process sequences (e.g., time series) using RNNs & CNNs. The 2nd covers NLP using RNNs, Encoder–Decoders & Attention (including Transformers). See below for the full menus. Bon appétit ! 👨🍳 github.com/ageron/handson-ml…
9
101
537
Just finished rewriting the ConvNet chapter! 😅
Now includes building ResNet-34 in #TensorFlow 2 (see image), fine-tuning a pretrained model, object detection and image segmentation.
I pushed the notebook: github.com/ageron/handson-ml…
8
97
539
@TensorFlow 2.0 preview is here! 🥳
I wrote a few Jupyter notebooks to teach #DeepLearning using tf.keras and TF 2.0, I hope you'll find them useful! Teachers, you are more than welcome to use them in your own courses.
github.com/ageron/tf2_course
7
175
528
Just pushed a notebook that covers loading and preprocessing data efficiently using #TensorFlow 2. It includes code examples for the data API, TFRecords & the Features API (+ short examples of TF Transform, TF datasets and TF Hub). Enjoy!
👉 Notebook 13 in github.com/ageron/handson-ml…
8
125
487
If you know Python and you'd like to learn Julia, check out this Colab notebook, I just updated it to Julia 1.8.2:
colab.research.google.com/gi…
7
82
461
Just finished the #TensorFlow 2 chapter for the upcoming 2nd edition of my #HandsOnML book. As a picture is worth a thousand words, here's a peek at one of the figures about @TensorFlow Functions. Looking forward to the #TFDevSummit ! 😀
10
72
468
I just received the final proofs of the 2nd edition! Double-checked everything, it's now ready to print. Exciting times!
😍🦎
21
40
471
Woo-hoo! Here's the full list of translations available so far for the 1st and 2nd edition of my book: homl.info/get/
The fish isn't sure what it's doing here.
43
49
442
Online conferences are the best:
* attend in pyjamas, pause, rewind, listen x1.5, never miss a talk
* no travel costs, anyone in the world can join, rich or poor
* talks can reach thousands of people, not just a room-full
* and waaaaaaaaay less CO2
6
42
427
I just had the chance to review "Natural Language Processing with Transformers" by @_lewtun, @lvwerra and @Thom_Wolf from 🤗@huggingface: it's a fantastic book! Very clear, hands-on, and packed with helpful tips & tricks.
Check out the early release:
learning.oreilly.com/library…
4
61
444
My cousin is an anesthetist doctor in France. She caught the virus, got better, and now she's back on the front line in one of the busiest hospitals in Paris. She's my hero.👩⚕️👏
3
15
421
The Early Release of the 3rd edition of my book is now available! homl.info/er3 🎉
It currently includes the first 9 chapters, raw & unedited.
It's the same overall structure as the 2nd edition, but with many improvements:
30
51
398
Einops simplifies and clarifies array/tensor manipulation. 👍
You really have to try it out, you'll love it: github.com/arogozhnikov/eino…
Plus it supports NumPy, TensorFlow, PyTorch, and more.
2
65
391
I used to prefer Jupyter Notebook over Jupyter Lab: simpler & cleaner UI.
But I recently gave Jupyter Lab another shot, and I ❤️ it:
✓ it has a debugger
✓ it's customizable & one click gives you a simplified UI.
👇
pip install jupyterlab # or conda install
jupyter lab
16
28
346
I just uploaded a short video to explain 3 concepts from Shannon's Information Theory that are frequently used in Machine Learning: entropy, cross-entropy and KL divergence. I hope you'll enjoy it!
piped.video/ErfnhcEV1O8
6
125
345
Want to learn how to build a sequence-to-sequence Encoder-Decoder using Keras, or using TF-Addons' seq2seq API? If so, check out the solution to exercise 9 in this notebook:
colab.research.google.com/gi…
16_nlp_with_rnns_and_attention.ipynb
Run, share, and edit Python notebooks
colab.research.google.com 5
64
332
Did you know you can start a server inside Colab and access it remotely?
E.g., the following code starts a TensorBoard server, and displays the UI inside Jupyter (as usual), but it also adds a link at the top: if you click it, you'll access the TensorBoard server in another tab.
6
44
320
I just uploaded a new video! It's a brief overview of the many improvements to come in @TensorFlow 2.0: eager execution by default, easy to combine with graph mode for performance and portability, a more object-oriented design, and a big cleanup. Enjoy!
piped.video/watch?v=WTNH0tcs…
6
109
324
My son's learning programming and he's shocked that a[1:3] means the *second* and third element in most programming languages.
Who decided this was a good idea?
Indexing in Julia (and other math-oriented languages) feels so much more natural—once you've unlearned 0-indexing.
38
22
321
Awesome book, with clear and engaging explanations, tons of useful practical tips and simple code examples to build state-of-the-art models. Congrats @jeremyhoward and @GuggerSylvain ! 👍
We've made a draft of our deep learning book available as @ProjectJupyter notebooks.
Important: This is for personal use, and not licensed for any redistribution or change of format or medium. Eg. don't make a PDF and share it, please!
github.com/fastai/fastbook
3
46
321
It's 2040. A team of researchers formulates a comprehensive mathematical theory of intelligence. Due to information-flow constraints, it turns out there's an upper limit to intelligence, dubbed Einstein's Razor, and humans are near optimal. The singularity is thus impossible. 1/6
11
44
316
Btw, if you'd like to know how to create an interactive world map like this one, it's actually very simple using plotly's choropleth() function.
Here's the Colab notebook with the whole code, it's just 8 simple cells:
colab.research.google.com/gi…
4
32
305
Have you ever used Scikit-Learn's TransformedTargetRegressor? It's pretty convenient when your targets need to be preprocessed.
For example, if they have widely different scales, it's probably best to predict log(targets), then compute exp(predictions). Like this:👇
4
43
315
LeNet-5, AlexNet, GoogLeNet, ResNet, SENet, Xception, ResNeXt, DenseNet, MobileNet, CSPNet, and EfficientNet.
Can you think of any other CNN-only image classification model that really should be introduced (briefly or not) in the 3rd edition of my book? Thanks for your help! 🙏
50
24
289
My new book on Machine Learning with TensorFlow is now available (Early Release)!
#MachineLearning #AI #TensorFlow shop.oreilly.com/product/063…
35
83
274
In the Transformer architecture (arxiv.org/abs/1706.03762), a Positional Embedding (PE) is added to each word embedding. This allows the model to know the word positions, both absolute and relative. Here's a diagram to understand how it works. 🦎
3
58
275
The Python installer does not install the root SSL certificates automatically on macOS. As a result, many people get a SSL: CERTIFICATE_VERIFY_FAILED error, and proceed to just deactivate SSL certificate verification. It's a common "answer" found online. Please don't do this! 🙏
7
31
253
A few years ago, I founded an A.I. consulting company which I named Kiwisoft. When I later moved to New Zealand, I found out that it's the name of a popular brand of toilet paper here. 🧻
9
14
235
Two more chapters in the early release (homl.info/er2)! 🦎
13=loading and processing data using tf.data, the features API, TFRecords, and more.
14=deep computer vision (image classification, object detection, semantic segmentation, etc.).
Enjoy!
10
29
238
The 3rd edition of my book is being printed, it will be released within the next few weeks! 📙🦎🎉😁
You can preorder it at homl.info/amazon3
You can play with the notebooks at github.com/ageron/handson-ml…
It covers many new topics, including diffusion models:
22
28
225
Yes, sadly I can confirm that I've had several months with almost no royalties despite my book being in Amazon's best seller list. Plus many negative reviews from readers complaining about poor quality print. Amazon benefits from counterfeit, so they turn a blind eye. 😰
Q: Does this affect any other book?
A: Absolutely. It affects nearly all high sale volume technical books on Amazon. If you've bought a technical book on Amazon recently there's a >50% chance it's a fake.
E.g.
@aureliengeron's book is also heavily affected, etc.
18
32
209
I greatly enjoyed this paper! 👍 It provides conceptual and technical tools to progress towards AGI. I wrote a TFDS dataset so you can load it as a TF dataset simply with tfds.load("arc"). You can play with it in this Colab notebook:
colab.research.google.com/dr…
I've just released a fairly lengthy paper on defining & measuring intelligence, as well as a new AI evaluation dataset, the "Abstraction and Reasoning Corpus". I've been working on this for the past 2 years, on & off.
Paper: arxiv.org/abs/1911.01547
ARC: github.com/fchollet/ARC
1
44
224
Congrats François! 🥳
This is a fantastic book, highly recommended reading for anyone who wants to learn ML. Super clear, with excellent code, explanations, analogies and insights.
Announcement: my book Deep Learning with Python (2nd edition) has been released.
500 pages of code examples, theory, context, practical tips... If you want to really understand how deep learning works, why it matters, and how to use it, this is your book!
manning.com/books/deep-learn…
1
11
206
Here's a simple #TensorFlow 2.0 example using TFDS, tf.data, and a custom layer with the same API as the upcoming Keras preprocessing layers. Hope it helps! 🦎
6
42
211
TensorFlow 2 is out, woohoo!!! 🥳
Congrats to the @TensorFlow team and to everyone who worked on this huge project! 👏
🗣 Attention #TensorFlow community!
We are excited to announce TensorFlow 2.0. Inspired by all of your feedback, this powerful and easy-to-use framework includes tight Keras integration, expanded TF datasets, and more.
Watch @lmoroney present here → goo.gle/2nTRZCp
1
23
193
O'Reilly just confirmed to me that they underestimated the number of copies to print in the 1st batch for Amazon. 😢The 2nd batch is much bigger and it should be available on Amazon around Nov 4th. In the meantime, bookstores like Barnes & Nobles seem to have copies available.
34
13
186
Wow, thanks a lot for your kind words, Tim! 😊
It's been a wild and exciting ride, and I feel incredibly grateful to everyone who helped me along the way. The field is moving so fast, I'm curious to see what this new decade will bring! 🦎
A nice review of what is also my own favorite book on machine learning. @aureliengeron has created a masterpiece. It is one of the best books @OReillyMedia has published in our entire history. kdnuggets.com/2020/01/book-s…
7
30
186
Ouch, my first accidental `rm -rf`! 😱
Fortunately, no serious data loss thanks to backups (just lost ~1 hour of work).
Any good protection against the evil rm -rf (other than growing a brain🧠)?
26
3
181
Just added code examples and exercises on #autodiff in @TensorFlow 2.0. Covers autodiff basics, computing 2nd order derivatives, how to write a custom training loop to train a Keras model, and how to use running metrics. See notebook #2 in github.com/ageron/tf2_course . Enjoy!
1
41
185
I'm starting a YouTube channel! Check out my very first video on Capsule Networks, I hope you'll like it: piped.video/pPN8d0E3900
6
65
174
You might also be interested in my notebook about Auto-Differentiation: github.com/ageron/handson-ml…
It explains forward/reverse mode autodiff, it shows how to use TensorFlow's autodiff functionality, and explains how it works by implementing a tiny autodiff framework:
3
30
173
Awesome video for an amazing paper! 😍
New Video 🔥 Deep Learning is very good at fitting functions numerically, but what about deriving symbolic expressions? How Graph Networks can learn Newtonian Physics and Dark Matter!
piped.video/LMb5tvW-UoQ
@MilesCranmer @PeterWBattaglia @KyleCranmer @DavidSpergel @cosmo_shirley
1
35
173
Great library! I'm currently updating the RL chapter in my book and adding some examples using TF-Agents. 🦎
The TF-Agents library is a well-tested, open-source resource for your reinforcement learning projects. Check out this #io19 to learn how you can mix and match TF-Agent components to create all new RL algorithms.
See how it's done now → goo.gle/2vPMB41
3
28
172
No one looks up sin(x) on a sin table anymore, so why do statisticians still look up distributions on tables? 📜
If they don't, why do stats courses still use tables rather than, say, scipy.stats? 🤔
14
13
163
A simple linear model trained on a person's past decisions can actually outperform that person! For example, a linear model of a bail judge is better than that judge (on average) at predicting which suspects will skip bail or commit another crime. How can that be? 1/7
4
29
158
One more chapter added to the Early Release of the 2nd edition:
Chapter 12: Building custom models and training algorithms using #TensorFlow 2.0. Also includes autodiff, tf.function(), graphs in TF2, and more. Unlimited flexibility, unlimited POWER!
homl.info/er2 🦎
8
35
158
Bye bye Facebook. Bye bye Instagram.
I am going off Facebook and Instagram & trying to delete my accounts. I cannot support an organization that pretends to support free speech, but actually enables hate speech. Neutrality in the midst of violence is violence. Bye bye Facebook. Bye bye Instagram. No more posts.
6
15
146
Make sure you set PYTHONHASHSEED=0 *before* you start Python or Jupyter, as Python only reads it upon startup. Note that Google Colab sets it for you.
3
23
158
Just reached 10,000 subscribers on my YouTube channel, woohoo! 🥳
Thanks so much to all of you, I'm really thrilled!
As soon as I finish the 2nd edition of my book, I'll make more videos, promise! What topics would you like me to talk about?
#10kcreator piped.video/c/AurelienGeron
25
16
159
TensorFlow 2.0 will be released later this year! It will encourage eager execution and high level APIs (e.g., tf.keras), it will support more platforms & languages, and there will be a lot of cleanup (less duplication, tf.contrib moved to core or other projects, etc.).
TensorFlow v2.0 is coming - with a focus on ease of use!
groups.google.com/a/tensorfl…
2
43
152
I just pushed two Docker images to run the handson-ml2 project, with all the latest libs (incl. TensorFlow 2.4.1). docker pull ageron/handson-ml2 for CPU-only, and handson-ml2:latest-gpu for GPU support. 😅
See github.com/ageron/handson-ml… for more details.
7
27
150
StackOverflow is currently down, you can take the rest of your day off.
Or... you can still view the cached pages in Google search (see screenshot) or use wayback.archive.org.
1
12
145
Keras now supports JAX and PyTorch, as well as TensorFlow, woohoo! 🎉😊
I've tried running my notebooks on Keras Core, it worked like a charm, mostly a drop-in replacement for tf.keras.
We're launching Keras Core, a new library that brings the Keras API to JAX and PyTorch in addition to TensorFlow.
It enables you to write cross-framework deep learning components and to benefit from the best that each framework has to offer.
Read more: keras.io/keras_core/announce…
1
18
154
31,882
Variable scopes and the whole tf.get_variable() mess is going away! To share variables, just reuse #Keras layers, it's so much cleaner and intuitive. I can't wait for @TensorFlow 2.0!
github.com/tensorflow/commun…
4
29
146
I'm sorry if I sounded too aggressive, it's hard to be subtle on Twitter, unless you just censor yourself and never speak your mind.
I'm impressed by self-landing rockets and (partly) self-driving cars, of course. But Musk lies way too much.
I'm not impressed by Optimus at all.
10
5
135
7,308
A few people have reported poor-quality counterfeit copies of my book floating around. 😢
If you receive one of these, please return it to Amazon, report the seller, and write a bad review for them – be sure you're reviewing the seller and not the book! 🙏
18
16
137
In Python 3, identifiers can contain Unicode letters & digits. This can make equations easier to read:
6
43
136
Notebooks updated: added clustering, anomaly detection and mixture models in notebook 08, updated the Estimator API example in notebook 10, added a short example for TF's Data API in notebook 12, and updated notebooks to the latest libraries. Enjoy! github.com/ageron/handson-ml
3
25
139
Yet another great video by @veritasium. Matrix multiplications are at the core of Deep Learning. Since we generally don't need perfect precision, using specialized analog chips could be much faster, smaller, and energy efficient than GPUs or TPUs.
New video: a brief history of AI and why analog computers may be the future of deep neural networks.
piped.video/GVsUOuSjvcg
Let me know if you prefer any of these alternative thumbnails:
2
17
140
Btw, the 2nd edition will be fully in color! 🥳
I tried to keep color-blind readers in mind, though (e.g., I wrote "the dashed line on the diagram" rather than "the green line"). If you're color-blind, I'd love to get your feedback if you find some figures hard to read in color.
4
5
133
TIL that %matplotlib inline is now optional when you import matplotlib.pyplot in a Jupyter notebook.
How long have I been using it for nothing!? 😬
6
9
131
Just added some simple code examples (+an exercise) about reading/writing TFRecords in my TF 2 course Jupyter notebooks, in case you're interested (the .tfrecords format is @TensorFlow's preferred binary representation for large datasets).
github.com/ageron/tf2_course…
2
32
126
I just uploaded a new video! How to implement CapsNets using TensorFlow. Includes a few tricks that you may find useful for other models as well: efficiently multiplying many matrices in one shot, writing dynamic loops & conditional operations, and more. piped.video/2Kawrd5szHE
3
65
126
I'm delighted to renew this crash-course on deep learning with #TensorFlow 2.0! Thanks @CriteoAILab for organizing this. See you all in June! 🤗
Back by popular demand, the Crash-course on deep learning 2, with Aurélien Géron, will take place on June 20, 24 & 26. Call for application deadline is June 7th. #AI #TensorFlow. Apply here: ow.ly/XIDh50uj4xN
2
20
118
If you want to understand automatic differentiation, grasp the difference between forward mode and reverse mode #autodiff, know how they impact performance and RAM usage, and play with simple toy implementations of each, check out my new notebook: github.com/ageron/handson-ml…
1
38
124
It has much more content than the 1st edition (×1.5 number of pages), but it doesn't look much thicker:
9
8
118
Wow! There's another guy called Géron who published *3* ML books about a month ago. All of them already have 70 super positive reviews on Amazon. Impressive! But I'm even more impressed by the reviewers: each of them rated ~10 books/day, every day, over the last month. Mmmh... 🕵️♂️
7
14
115
