Profile picture
, 3 tweets, 1 min read
My Authors
Read all threads
New content on keras.io: Keras debugging tips.

keras.io/examples/keras…
1 - Test the parts before you test the whole
2 - Use `model.summary()` and `plot_model()` to check output shapes and connectivity graph
3 - Use `run_eagerly=True` in `compile()` to debug training step-by-step
4 - Use the TF profiler (simple callback) to fix performance issues
Also, yes -- if you ever find yourself needing second-order gradients when writing a Keras training step, this is how to get them:
Missing some Tweet in this thread? You can try to force a refresh.

Keep Current with François Chollet

Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

More from @fchollet see all

Profile picture
François Chollet
@fchollet
In your opinion, will the US experience deliberate Internet blackouts (ordered by the state) before the end of the year?
Internet blackouts are a natural next step when authoritarian regimes are faced with widespread social unrest that official propaganda channels can't contain -- alongside sending in the troops. This is really asking "do you think the US will follow the remainder of the playbook"
I personally don't think it's likely, maybe 25% chance at most. But it's within the range of possible events one should consider
Read 3 tweets
Profile picture
François Chollet
@fchollet
Here's the thing about AI: you get what you optimize for. If you optimize for a specific skill, like chess or StarCraft, your final system will possess this skill and nothing else. It won't generalize to any other task.

To generalize, you must optimize for generality itself.
To be clear, optimizing for task-specific skill can be valuable. It gets you somewhere. But now, we're at a stage in the development of AI where generalization has become, inevitably, the bottleneck to skill acquisition.
Of course, if you can amass a sufficiently dense sampling of situations within a sufficiently narrow domain, you can always train a machine model -- but it will break down as soon as it encounters anything it has never seen before.

That's modern deep learning.
Read 5 tweets
Profile picture
François Chollet
@fchollet
Woke up to 3 new hate emails from PyTorch fanboys, a new record (was this coordinated?).

I got my very first hate message in 2017, from a PyTorch fan, and it has been a recurring thing since. The degree of toxicity of that community is insane.

It's exhausting.
If you are a Keras user, I beg you: never behave like these people. Don't be a jerk. Don't harass anyone. Not that I've had reports of that ever happening.

Just be cool. Build great things. Help others.
If you are a PyTorch fan: please leave me alone. My work is not related to you. You don't need to feel threatened by my work. I'm just doing the best I can. I've been at it for 5 years, it has been a lot of work.

So please leave me alone, and leave the Keras & TF team alone.
Read 5 tweets

Embed code for your website

×
Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!